Skip to content

an open-source alternative to the dots, bots and muses of the world, run an office of claude code/codex like agents on your laptop, sandboxes or anywhere, uses your existing subscriptions

blog: rewrite the top 10 posts for 0.5.2 and the current product

chaitanyagiri02b05f75e426
diff --git a/blog/src/posts/best-claude-code-multi-agent-tools.md b/blog/src/posts/best-claude-code-multi-agent-tools.md
index 0e167892..ed484c0d 100644
--- a/blog/src/posts/best-claude-code-multi-agent-tools.md
+++ b/blog/src/posts/best-claude-code-multi-agent-tools.md
@@ -1,170 +1,185 @@
---
title: "The Best Tools to Run Multiple Claude Code Agents (2026)"
-description: "An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared."
+description: "An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one."
date: 2026-06-04
+updated: 2026-09-10
category: comparisons
categoryLabel: Comparisons
type: Non-technical
primaryKeyword: "best claude code multi-agent tools"
-secondaryKeywords: ["claude code multi-agent tool", "agentic coding tools", "best tools to run multiple claude code agents"]
+secondaryKeywords: ["claude code multi-agent tool", "best tools to run multiple claude code agents", "claude code agent teams", "crystal nimbalyst", "emdash coding agents", "conductor claude code"]
tags: ["Comparisons", "Multi-Agent", "Claude Code", "Tools"]
author:
name: Chaitanya Giri
initials: CG
faq:
- q: "What is the best tool to run multiple Claude Code agents?"
- a: "There's no single best — it depends on what you want. Claude Squad is the leanest terminal option; Conductor is the most polished on macOS; Crystal is a good open-source desktop GUI; vibe-kanban gives you a task board; and Munder Difflin adds shared memory, inter-agent messaging, and a GOD orchestrator so the agents act as one coordinated team."
- - q: "Are these Claude Code multi-agent tools free?"
- a: "Most are free and several are open source (Claude Squad, Crystal, vibe-kanban, and Munder Difflin are open source; Munder Difflin is MIT-licensed). Conductor is a free, native macOS app. Always check each project's current license and pricing before you commit."
- - q: "Do I need a multi-agent tool to use Claude Code?"
- a: "No. One Claude Code session handles most tasks. You start wanting a multi-agent tool once you're running three or more sessions at once and the coordination overhead — who's doing what, who knows what, who edits which file — starts costing you time."
+ a: "It depends on your bottleneck. Claude Code agent teams are the built in option for a single session. Claude Squad is the leanest terminal manager. Conductor is the polished Mac app, with cloud workspaces on paid plans. Nimbalyst and Emdash are open source desktop apps. Munder Difflin adds long term memory, messaging and an orchestrator across twelve CLIs."
+ - q: "Does Claude Code have a built in way to run multiple agents?"
+ a: "Yes, two. Subagents handle short helper jobs inside one session. Agent teams, which are experimental and stay off until you set CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, let a lead session spawn teammates that share a task list and message each other."
+ - q: "What happened to Crystal?"
+ a: "Its maker deprecated Crystal in February 2026 and replaced it with Nimbalyst, an MIT licensed visual workspace for Claude Code and Codex."
+ - q: "Is Vibe Kanban still maintained?"
+ a: "Its website says Vibe Kanban is sunsetting and will continue as an open source, community maintained project."
+ - q: "Are these tools free?"
+ a: "Most have a free option. Claude Squad (AGPL 3.0), Nimbalyst (MIT), Emdash (Apache 2.0), Vibe Kanban (Apache 2.0) and the Munder Difflin app (MIT) are open source. Conductor has a free plan for local workspaces and paid plans for cloud and team features. Check pricing before you commit, because this space moves fast."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p>The tools for running
-<strong>multiple Claude Code agents</strong> fall into three camps: <strong>terminal session
-managers</strong> (Claude Squad), <strong>parallel-worktree desktop apps</strong> (Conductor,
-Crystal), <strong>task boards</strong> (vibe-kanban), and <strong>coordinated hives</strong> (Munder
-Difflin). The first three help you run agents <em>side by side</em>; a hive adds shared memory,
-messaging, and an orchestrator so they run <em>as a team</em>. Pick by how much coordination you
-actually need.</p></div>
-
-If you've outgrown a single Claude Code window, the next question is *which tool* helps you run
-several at once without the chaos. The landscape moved fast in 2025–2026, so here's an honest,
-plain-English roundup of the main options — what each does well, where it stops, and who it's for.
-
-> **A note on fairness:** this space changes quickly. Features, platforms, and licenses below are
-> accurate to the best of our knowledge at the time of writing — always check each project's repo or
-> site for current details. Munder Difflin is our own tool; we've tried to describe the others on
-> their own terms, not as straw men.
-
-## The four shapes of "multi-agent Claude Code"
-
-Before the tools, the categories — because they answer different questions:
-
-- **Terminal session managers** run each agent in its own shell and let you switch between them from
- one keyboard-driven interface. Minimal, fast, SSH-friendly.
-- **Parallel-worktree desktop apps** give each agent an isolated [git
- worktree](/blog/how-to-run-multiple-claude-code-agents/) and a GUI to launch tasks and review
- diffs. Great for "try five approaches in parallel."
-- **Task boards** model the work as cards you assign to agents and move through columns — a
- project-management view over agentic coding.
-- **Coordinated hives** add the layer the others leave to you: shared long-term memory, direct
- agent-to-agent messaging, and an orchestrator that routes work. That's the [multi-agent
- harness](/#what) idea.
-
-Most tools live cleanly in one camp. Knowing which camp you need is most of the decision.
-
-## The tools, at a glance
-
-| Tool | What it is | How it parallelizes | Shared memory | Orchestrator | License |
-|---|---|---|---|---|---|
-| Claude Squad | Terminal (TUI) session manager | tmux + git worktrees | No | No (you assign) | Open source |
-| Conductor | Native macOS desktop app | Parallel git worktrees | No | No (you assign) | Free, macOS-only |
-| Crystal | Open-source desktop app | Parallel sessions + worktrees | No | No (you assign) | Open source |
-| vibe-kanban | Kanban board for agents | Task cards across agents | No | Board, not auto-routing | Open source |
-| Munder Difflin | Local coordinated hive | Roles + mailboxes + orchestrator | Yes (MemPalace) | Yes (GOD agent) | Open source (MIT) |
-
-The single row that most separates the field is **shared memory + orchestrator**: that's the jump
-from "running agents in parallel" to "running a team."
+<div class="callout tldr"><span class="ic">TL;DR</span><p>Tools for running <strong>multiple Claude Code
+agents</strong> come in four shapes: <strong>built in</strong> (Claude Code agent teams),
+<strong>session managers</strong> (Claude Squad), <strong>parallel workspace apps</strong> (Conductor,
+Nimbalyst, Emdash, and Vibe Kanban as a board), and <strong>coordinated offices</strong> (Munder Difflin).
+The first three help you run agents side by side. An office adds shared memory, messaging and an
+orchestrator so they run as a team.</p></div>
+
+A lot changed since we first wrote this in June. Claude Code grew its own agent teams, Crystal became
+Nimbalyst, Conductor moved into the cloud, Vibe Kanban announced it is winding down, and Emdash arrived.
+Here is the current picture, checked on 10 September 2026.
+
+<div class="callout note"><span class="ic">Fair warning</span><p>Munder Difflin is our own tool. Everything
+about the others comes from their own sites, READMEs and docs as of 10 September 2026, and every tool is
+linked so you can check it yourself.</p></div>
+
+## What kinds of tools run multiple Claude Code agents?
+
+Four kinds, and knowing which one you need is most of the decision.
+
+- **Built in.** Features inside Claude Code itself. Nothing new to install, and the team lives in one session.
+- **Session managers.** Many agents in terminal sessions, driven from one keyboard friendly interface.
+- **Parallel workspace apps.** A desktop app that gives each agent an isolated workspace and a place to
+ review what it changed.
+- **Coordinated offices.** Agents with roles, mailboxes, shared long term memory and an orchestrator that
+ routes the work.
+
+## The tools at a glance
+
+| Tool | Shape | Agents it runs | Platforms | License |
+|---|---|---|---|---|
+| Claude Code agent teams | Built in, experimental | Claude Code | Wherever Claude Code runs | Part of Claude Code |
+| Claude Squad | Session manager | Claude Code, Codex, Gemini, Aider and more | Terminal, needs tmux | AGPL 3.0 |
+| Conductor | Parallel workspaces, cloud on paid plans | Claude Code, Codex, Cursor, OpenCode | macOS, iOS listed as coming | Free and paid plans |
+| Nimbalyst | Visual workspace, parallel sessions | Claude Code, Codex | macOS, Windows, Linux | MIT |
+| Emdash | Parallel workspaces | Any provider, including Claude Code, Codex, Amp, Antigravity | macOS, Windows, Linux | Apache 2.0 |
+| Vibe Kanban | Task board, sunsetting | Claude Code, Codex and others | Runs with npx | Apache 2.0 |
+| Munder Difflin | Coordinated office | Twelve CLIs, including Claude Code, Codex, Gemini CLI, Copilot, Cursor | macOS, Windows, Linux | MIT (free app) |
{% img "note-1" %}
-## Claude Squad — the lean terminal option
+## Claude Code agent teams: the built in option
+
+[Agent teams](https://code.claude.com/docs/en/agent-teams) are Claude Code's own way to run a team, and
+they are still experimental. Set `CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1` and a lead session can spawn
+teammates. Each teammate is a separate Claude Code instance with its own context window. They share a task
+list, claim work, and message each other directly. You can keep them all in one terminal, or give each one
+a split pane with tmux or iTerm2.
-[Claude Squad](https://github.com/smtg-ai/claude-squad) is a terminal UI that manages multiple AI
-coding agents — Claude Code among them — each in its own tmux session, with git worktrees keeping
-their changes isolated. You launch, switch, background, and review agents without leaving the
-terminal.
+- **Strengths:** nothing to install, first party, and teammates genuinely talk to each other.
+- **Where it stops:** Anthropic's docs list the limits. A session has exactly one team, in process
+ teammates are not restored when you resume a session, and token use grows with every teammate. Every
+ teammate is Claude.
-- **Strengths:** featherweight, keyboard-driven, works great over SSH, no GUI overhead. If you live
- in the terminal and want parallel agents *now*, it's hard to beat for speed of setup.
-- **Where it stops:** agents don't share memory or message each other, and there's no orchestrator —
- you decide who does what. It's a superb *session manager*, not a coordination layer.
+Best for: research, review and debugging jobs you finish in one sitting.
-Best for: terminal purists running a handful of independent tasks. If you want a deeper look at why
-isolation alone isn't coordination, see [Claude Squad alternative](/blog/claude-squad-alternative/).
+## Claude Squad: the lean terminal manager
-## Conductor — the polished macOS experience
+[Claude Squad](https://github.com/smtg-ai/claude-squad) is a terminal app that manages several agents,
+including Claude Code, Codex, Gemini and Aider, each in its own workspace. It needs tmux and installs as `cs`.
-[Conductor](https://conductor.build) is a native macOS app for running Claude Code agents in
-parallel, each in its own isolated git worktree, with a clean Mac-native interface for kicking off
-tasks and reviewing their diffs.
+- **Strengths:** light, fast, keyboard driven, and comfortable over SSH.
+- **Where it stops:** its README describes managing sessions. It does not describe shared memory or agents
+ messaging each other, so deciding who does what stays with you.
-- **Strengths:** genuinely nice UX, fast diff review, and a low-friction way to run several
- worktree-isolated agents at once on a Mac.
-- **Where it stops:** it's macOS-only, and it's built around *parallel isolated workspaces* rather
- than a coordinated team — no shared cross-agent memory, no agent-to-agent messaging, no
- plain-language orchestrator. Closed-source, so you can't read or extend the internals.
+Best for: terminal people running a handful of independent tasks. More in
+[Claude Squad vs Munder Difflin](/blog/claude-squad-vs-munder-difflin/).
-Best for: Mac users who want a refined parallel-worktree workflow. If you're cross-platform or want a
-coordination layer, see [a Conductor alternative](/blog/conductor-claude-code-alternative/).
+## Conductor: the polished Mac app, now with a cloud
-## Crystal — the open-source desktop GUI
+[Conductor](https://conductor.build) started as a Mac app for running coding agents in parallel, each in its
+own workspace. It now runs Claude Code, Codex, Cursor and OpenCode, and pitches itself as a way to run a team
+of coding agents in the cloud. The free plan covers running agents in parallel in local workspaces on your
+Mac. Cloud workspaces come with paid plans, and its Teams plan adds live collaboration on shared workspaces.
-[Crystal](https://github.com/stravu/crystal) is an open-source desktop app that runs multiple Claude
-Code sessions in parallel, each in its own git worktree. You spin up sessions from prompts, run them
-concurrently, and review diffs and test results across worktrees in one window.
+- **Strengths:** a refined interface, quick review, and cloud workspaces that can run for hours.
+- **Where it stops:** the desktop app is macOS only today, with iOS listed as coming soon, and the cloud and
+ multiplayer parts are paid.
-- **Strengths:** open source, a real GUI, and a tidy model for running several experiments in
- parallel and comparing the results.
-- **Where it stops:** like the others in this camp, it's parallel isolated sessions — agents don't
- pool memory or talk to each other, and there's no orchestrator routing work between them.
+Best for: Mac users who want polish and are happy to pay for cloud runs. See
+[a Conductor alternative](/blog/conductor-claude-code-alternative/).
-Best for: developers who want an open, visual, worktree-based parallel workflow without the terminal.
-More on the trade-offs in [a Crystal alternative](/blog/crystal-claude-code-alternative/).
+## Nimbalyst: what Crystal became
+
+[Crystal](https://github.com/stravu/crystal) was deprecated in February 2026 and replaced by
+[Nimbalyst](https://nimbalyst.com), which describes itself as an open source visual workspace for building
+with Codex, Claude Code and more. It runs parallel sessions with optional git worktree isolation, adds visual
+editors for markdown, mockups, diagrams and data models, and organises sessions on a kanban. The desktop apps
+for macOS, Windows and Linux are MIT licensed.
+
+- **Strengths:** open source, cross platform, and good at reviewing what agents change, including non code files.
+- **Worth checking:** its pricing page, for the team and collaboration features.
+
+Best for: developers who want an open, visual, cross platform workspace. Our older
+[Crystal comparison](/blog/crystal-claude-code-alternative/) still explains the worktree model it grew from.
{% img "note-2" %}
-## vibe-kanban — the task board for agents
+## Emdash: the open source newcomer
+
+[Emdash](https://emdash.sh) is an open source agentic development environment backed by Y Combinator. It runs
+coding agents in parallel with any provider, schedules repeat work, previews apps in a built in browser, pulls
+issues from Linear, Jira and GitHub, and manages prompts, skills and MCP tools. It runs on macOS, Windows and
+Linux under Apache 2.0.
+
+- **Strengths:** wide agent support, issue tracker integrations, and every desktop platform.
+- **Worth checking:** remote development is its newest area, so test it on your own setup first.
+
+Best for: teams that live in an issue tracker and want agents fed straight from it.
+
+## Vibe Kanban: the board, winding down
-[vibe-kanban](https://github.com/BloopAI/vibe-kanban) takes a different angle: a kanban board where
-you create tasks as cards, assign them to coding agents (it's agent-agnostic — Claude Code, Gemini,
-Codex, and more), and move work through columns from to-do to review.
+[Vibe Kanban](https://github.com/BloopAI/vibe-kanban) turns agent work into cards you plan, prompt and review.
+It runs with `npx vibe-kanban` and works with Claude Code, Codex and other agents. Its site now says the project
+is sunsetting and will continue as open source, maintained by the community.
-- **Strengths:** the board is a familiar, legible mental model; it's agent-agnostic; and it's great
- for managing a *queue* of work and reviewing each agent's output before it lands.
-- **Where it stops:** a board is something *you* drive. Cards don't carry shared long-term memory
- between agents, agents don't message each other, and nothing auto-routes work — the orchestration
- is you, moving cards.
+- **Strengths:** a familiar board for planning work and reviewing agent output.
+- **Where it stops:** with its company stepping back, future fixes depend on the community.
-Best for: people who think in tasks and want a board over their agents. When you'd rather the routing
-happen for you, see [a vibe-kanban alternative](/blog/vibe-kanban-alternative/).
+Best for: people already on it who are comfortable with a community maintained tool. See
+[a Vibe Kanban alternative](/blog/vibe-kanban-alternative/).
-## Munder Difflin — the coordinated hive
+## Munder Difflin: the coordinated office
-[Munder Difflin](/#what) is our own take, and it's deliberately in a different camp. It turns the
-Claude Code terminals you already run into a *self-coordinating hive*: each agent gets a role and a
-mailbox, agents message each other directly, they share long-term **semantic memory** (MemPalace),
-and a [GOD orchestrator](/#how) you talk to in plain language decomposes your intent and routes work
-across the team. The whole floor is visualized as avatars at their desks, so you can actually watch
-it run.
+[Munder Difflin](https://munderdiffl.in/) is ours, and it lives in the fourth camp. It wraps twelve terminal
+coding CLIs (Claude Code, Codex, Gemini CLI, Antigravity, Grok, Kimi Code, Qwen, OpenCode, Crush, Pi, Copilot
+and Cursor) and runs them as real processes on your machine. Each agent gets a desk on an office floor, a
+mailbox and long term memory. Michael, your clone, turns what you ask for into tasks, hires workers and
+routes messages between them. Questions that need you land on an ASK ME board.
-- **Strengths:** it adds the coordination the other tools leave to you — [shared
- memory](/blog/give-claude-code-long-term-memory/), inter-agent messaging, an orchestrator, and
- visibility. Local-first and MIT-licensed, on macOS, Windows, and Linux.
-- **Where it stops (honestly):** it's a younger project, and the office-floor visualization is
- heavier than a TUI. For one or two quick parallel tasks, a lean session manager is less to think
- about — a hive earns its keep once coordination is the real cost.
+- **Strengths:** memory and messaging that outlast a session, one orchestrator for the whole floor, engines
+ from different vendors on the same team, and it runs on the subscriptions you already pay for. Free and MIT
+ licensed on macOS, Windows and Linux.
+- **Where it stops, honestly:** an office floor is more app than a terminal manager, and your machine has to
+ stay on for the agents to keep working. Hosted sandboxes are not available yet. For one or two quick
+ parallel tasks, a lighter tool is less to think about.
-Best for: people running enough agents that *coordination* — not just parallelism — is the problem.
+Best for: anyone running enough agents that coordination, not parallelism, is the real cost. The free app does
+all of the above. Optional paid plans exist too: Pro puts the office in one window and adds Stapler, and Teams
+lets your clone work with your teammates' clones. Details are on the [pricing page](https://munderdiffl.in/#pricing).
-## How to choose
+## How do you choose?
Match the tool to the bottleneck:
-- **"I just want parallel agents, fast."** → Claude Squad (terminal) or Crystal (GUI).
-- **"I'm on a Mac and want it to feel native."** → Conductor.
-- **"I want to manage work as tasks."** → vibe-kanban.
-- **"My agents need to share what they learn and stop colliding."** → Munder Difflin.
+- **"I only use Claude Code and need a team for one job."** Try agent teams first.
+- **"I want parallel agents in the terminal, now."** Claude Squad.
+- **"I am on a Mac, I want polish, and I will pay for cloud runs."** Conductor.
+- **"I want an open, visual, cross platform workspace."** Nimbalyst or Emdash.
+- **"My agents need to remember, talk to each other and stop colliding, across more than one vendor."**
+ Munder Difflin.
-If you want a structured way to weigh these, we wrote a [buyer's
-checklist](/blog/how-to-choose-a-multi-agent-tool/) and a criteria-based [orchestration tools
-comparison](/blog/claude-code-orchestration-tools-compared/). And if you've decided coordination is
-your real problem, the fastest way to feel the difference is to
-[download Munder Difflin](/#install) — it's free and open source.
+For a more structured pass, there is a [buyer's checklist](/blog/how-to-choose-a-multi-agent-tool/) and a
+criteria based [orchestration tools comparison](/blog/claude-code-orchestration-tools-compared/).
---
-The honest summary: there's no universal "best." The best tool is the one that solves *your*
-bottleneck — and the bottleneck shifts from parallelism to coordination the moment you're running a
-real team of agents.
+There is no universal best. The right tool is the one that removes your bottleneck, and that bottleneck moves
+from parallelism to coordination the day you start running a real team of agents.
diff --git a/blog/src/posts/command-center-guide.md b/blog/src/posts/command-center-guide.md
index 9ec6d129..45862262 100644
--- a/blog/src/posts/command-center-guide.md
+++ b/blog/src/posts/command-center-guide.md
@@ -1,7 +1,8 @@
---
-title: "The Command Center: Kanban, Fleet, and Budgets in One Place"
-description: "A guide to Munder Difflin's Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor."
+title: "The Command Center: Kanban, Fleet and Budgets in One Place"
+description: "A guide to Munder Difflin's Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent's budget, and when to watch the board instead of the floor."
date: 2026-07-03
+updated: 2026-09-10
category: guides
categoryLabel: Guides
type: Non-technical
@@ -13,83 +14,108 @@ author:
initials: CG
faq:
- q: "What is the Command Center in Munder Difflin?"
- a: "It's Michael's control surface — the management view of the whole hive. It has tabs for the orchestrator Terminal, the Floor (roster, dispatch, per-agent model selector, live fleet monitoring), Memory (semantic search plus a memory graph), Activity (event log, board, real token telemetry, observability, and a CI watcher), Tasks (a dependency-aware kanban board), and Triggers (everything that starts work without you — schedules with last/next-fired times, context, webhooks, and organization)."
+ a: "It is the management view of your office, next to the floor. Its tabs are terminal, monitor, tasks, ask me, triggers, history, memory, graph, activity, skills and workers. The floor shows you who is busy. The Command Center shows what is queued, what is blocked, what it costs and what runs next."
- q: "How does the task kanban work?"
- a: "The Tasks tab is a dependency-aware kanban board. You assign tasks to agents and track them across todo, doing, blocked, and done. Because tasks can declare dependencies, downstream work waits until its prerequisites finish, so a chain of related tasks starts in the right order without you babysitting the handoffs."
- - q: "How do per-agent budgets and cost tracking work?"
- a: "You set a token budget per agent, and live fleet monitoring tracks consumption across the roster. Cost numbers are real, not estimates from vibes: the Activity tab reads the JSONL transcripts Claude Code writes and surfaces actual token counts plus estimated USD cost per agent per session, backed by a durable cost ledger that survives restarts. The budget pairs with a circuit breaker that steers, then constrains, then stops an agent that loops or blows its cap."
- - q: "When should I watch the office floor versus the Command Center?"
- a: "The floor is for ambient awareness and single-agent moments — seeing who's working, watching envelopes fly, dropping into one terminal. The Command Center is for management questions: what's queued and blocked, what each agent has spent, whether CI is green, and what fires next on the schedule. Roughly: floor for one agent right now, board for the whole fleet over time."
- - q: "Does the Command Center include observability?"
- a: "Yes. The Activity tab includes a live OpenTelemetry collector with per-model cost attribution, a fleet grid, and a per-agent tool-span waterfall, so you can see exactly what every agent is doing and what it costs in real time. Each agent card also carries a context-window gauge showing how much of the model's context that agent has consumed."
- - q: "Can the Command Center pull in work from GitHub?"
- a: "Yes. You can ingest open issues from any registered repo via the gh CLI and assign them to agents with one click, and a CI status watcher shows live pass/fail/in-progress state for GitHub Actions runs in the Activity tab. Work flows in from GitHub, gets tracked on the kanban, and its CI result lands back in the same view."
+ a: "Tasks move across todo, doing, blocked and done, and each one is assigned to an agent. A task can depend on other tasks, so downstream work waits for its prerequisites and you stop being the one who sequences handoffs."
+ - q: "How do budgets and cost tracking work?"
+ a: "Each agent can have a token budget, and a circuit breaker steers, then constrains, then stops an agent that loops or runs past its cap. The activity tab shows live token use from each agent's telemetry against its own limit, or against the floor budget when an agent has none."
+ - q: "Where do questions for me show up?"
+ a: "On the ask me tab. When the team blocks a task on your input, whether a question to answer or a to do only you can do, it lands there and on the ASK ME board on the floor. Answer it and the work carries on, or dismiss it and the history is kept."
+ - q: "When should I watch the floor instead of the Command Center?"
+ a: "Watch the floor for one agent right now: you just sent work, you want to type into a session, or you want ambient awareness. Watch the Command Center for the whole fleet over time: what is queued or blocked, who is near budget, and what runs tonight."
+ - q: "Can the Command Center take work from GitHub?"
+ a: "Yes. Open issues from a registered GitHub repo can come onto the board as tasks and be assigned to agents."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p>The office floor is fun to watch, but you don't manage a team by staring at desks. Munder Difflin's <strong>Command Center</strong> is the management view: a <strong>dependency-aware task kanban</strong>, <strong>live fleet monitoring with per-agent token budgets</strong>, <strong>real cost telemetry</strong> backed by a durable ledger, <strong>OpenTelemetry observability</strong> with a per-agent tool-span waterfall, plus GitHub issue ingestion, a CI watcher, and a Triggers tab. Rule of thumb: <strong>watch the floor for one agent right now, watch the board for the whole fleet over time</strong>.</p></div>
+<div class="callout tldr"><span class="ic">TL;DR</span><p>The floor is fun to watch, but you do not manage a team
+by staring at desks. The <strong>Command Center</strong> is the management view: a <strong>kanban with task
+dependencies</strong>, an <strong>ask me</strong> tab for decisions only you can make, <strong>triggers</strong> for work
+that starts without you, <strong>memory</strong> and its <strong>graph</strong>, <strong>activity</strong> with live token use
+against each agent's budget, and a <strong>skills</strong> catalog. Rule of thumb: <strong>the floor for one agent right now,
+the Command Center for the whole fleet over time</strong>.</p></div>
<video controls preload="none" playsinline poster="/media/demo/features-poster.jpg" style="width:100%; border-radius:12px; margin:12px 0 24px;">
<source src="/media/demo/features.mp4" type="video/mp4" />
</video>
-Munder Difflin has two ways of looking at the same hive. The **office floor** is the ambient view: avatars at desks, envelopes flying between them, speech bubbles when a tool fires. The **Command Center** is the management view — Michael's control surface, the place you go when the question stops being "what is Dwight doing?" and becomes "what is *everyone* doing, what has it cost, and what's next?"
+Munder Difflin gives you two views of the same office. The **floor** is the ambient one: characters at desks and envelopes
+flying between them. The **Command Center** is the management one, where you go when the question changes from "what is Dwight
+doing?" to "what is everyone doing, what has it cost, and what happens next?" This guide walks through it as of 0.5.2.
-This guide walks through what's in it and how the pieces fit together.
+## What is in the Command Center?
-## What's in the Command Center
+Eleven tabs on one control surface: terminal, monitor, tasks, ask me, triggers, history, memory, graph, activity, skills and
+workers. The ones you will use most:
-Six tabs, one control surface:
+- **terminal:** Michael's own terminal, where you talk to your clone.
+- **monitor:** the roster at a glance, with a context gauge on every agent.
+- **tasks:** the kanban board.
+- **ask me:** everything that is waiting on your answer.
+- **triggers:** everything that starts work without you.
+- **memory** and **graph:** search the office's shared memory, and see how it connects.
+- **activity:** live token use and each agent's tool calls.
+- **skills:** a catalog of skills to install for your agents.
-- **Terminal** — Michael's own terminal, the [GOD orchestrator](/blog/how-the-god-orchestrator-works/) you talk to directly.
-- **Floor** — the roster and dispatch controls: hire, dispatch work, pick a model per agent, and watch live fleet monitoring across everyone at once.
-- **Memory** — semantic search over the shared memory palace, plain text search, and a memory graph.
-- **Activity** — the append-only event log, the blackboard, real token telemetry, the observability view, and a CI watcher.
-- **Tasks** — the kanban board.
-- **Triggers** — everything that starts work without you, grouped by kind: schedules (recurring missions and the adaptive heartbeat, with last/next-fired times), context, webhooks, and organization.
+You could run a small office from the floor alone. Past two or three agents, the Command Center is where the real decisions
+happen.
-You could run the hive from the floor alone. But once you have more than two or three agents, the Command Center is where the actual decisions happen.
+## How does the task kanban work?
-## The task kanban: work as a board, not a chat log
+It turns work into a board instead of a chat log. Tasks move across four columns, todo, doing, blocked and done, and every task
+is assigned to an agent.
-The Tasks tab is a **dependency-aware kanban board**. Tasks move across four columns — todo, doing, blocked, done — and each task is assigned to a specific agent.
+The part that matters is dependencies. A task can depend on other tasks, and it waits until they are done. The refactor lands
+before the tests are updated. The migration runs before the endpoint. You stop being the sequencer, because the board does it.
-The load-bearing word is *dependency-aware*. Real work has ordering: the refactor has to land before the tests get updated; the schema migration comes before the endpoint. On the board you wire those dependencies explicitly, and downstream tasks wait until their prerequisites finish. You're not the sequencer anymore — the board is.
-
-Work gets onto the board a few ways: you create tasks yourself, Michael creates and assigns them as he routes requests, and you can **pull open GitHub issues from any registered repo** (via the `gh` CLI) and assign them to agents with one click. Issue in, task tracked, agent dispatched — and when the work ships, the **CI status watcher** in the Activity tab shows live pass/fail/in-progress for the repo's GitHub Actions runs. The loop closes in the same window it opened in.
+Work reaches the board three ways: you add it, Michael creates and assigns it while he routes your requests, or it comes in from
+an open issue on a registered GitHub repo.
{% img "note-1" %}
-## Fleet status: the roster at a glance
-
-The Floor tab is where the fleet stops being a set of individual terminals and becomes a roster. Each agent card shows what that agent is up to, and the card's progress bar doubles as a **context-window gauge** — a glanceable read on how much of the model's context each agent has burned. An agent near the top of its gauge is an agent about to compact or slow down; you can see it coming instead of discovering it.
-
-Dispatch also lives here: pick an agent, pick a model (the **per-agent model selector** means your cheap-tier worker and your frontier-tier reviewer are one dropdown apart), and send work. Routine tasks go to the cheap tier; the frontier tier is reserved for the reasoning that needs it.
+## What does the ask me tab do?
-## Budgets and cost: real numbers, not vibes
+It collects every decision that needs you. When the team blocks a task on your input, whether that is a question to answer or a
+to do only you can do, it shows up here and on the ASK ME board on the floor. Questions render as formatted text, so a list of
+options reads like a list. Answer it and the work carries on. Dismiss it and the history is kept.
-This is the part that makes a 24/7 fleet safe to leave running.
+This is what keeps a busy office from hiding its questions in five different scrollbacks.
-**Per-agent token budgets.** You set a budget per agent, and live fleet monitoring tracks consumption across the whole roster against it. Budgets pair with the **circuit breaker**, which handles the runaway case on a ladder: *steer* (nudge the agent), then *constrain*, then *stop*. An agent that loops, storms errors, or blows through its cap gets caught by machinery, not by you noticing the bill.
+## How do budgets and cost tracking work?
-**Real telemetry.** The Activity tab doesn't estimate from message counts — it reads the JSONL transcripts Claude Code writes to `~/.claude/projects/` and surfaces **actual token counts and estimated USD cost per agent, per session**, backed by a **durable cost ledger** that survives app restarts. Yesterday's spend is still there this morning, attributed to the agent that spent it.
+This is the part that makes a floor safe to leave running.
-**Observability.** Also in Activity: a live **OpenTelemetry collector** with per-model cost attribution, a fleet grid, and a **per-agent tool-span waterfall** — the "what exactly is this agent doing, step by step, and what does each step cost" view. It's the same discipline we covered in [observability for agent fleets](/blog/observability-for-agent-fleets/), built into the harness instead of bolted on.
+- **Per agent token budgets.** Set a budget for each agent in Settings, under Autonomy & Budgets. The activity tab shows live
+ token use from each agent's telemetry as bars against its limit, or against the floor budget when an agent has no limit of its own.
+- **A circuit breaker.** An agent that loops or blows through its cap is steered, then constrained, then stopped, by the app
+ rather than by you noticing a bill.
+- **A tool waterfall.** Activity also lays out each agent's tool calls, so you can see exactly what it did, step by step.
+- **A context gauge.** On the monitor, every agent shows how much of its context window it has used, so you see a slowdown coming
+ before it arrives.
{% img "note-2" %}
-## Triggers: the board's fourth dimension
+## What starts work without you?
-The kanban answers "what's in flight." The **Triggers tab** answers "what happens next without me" — every way the hive wakes up when you aren't typing. **Schedules** are the first and oldest of them: recurring missions carry a label, an interval, a target agent, and a body, and the tab shows last-fired and next-fired times, so drift is visible. A scheduler **heartbeat** re-engages the floor when it goes quiet. Together with the board, it turns the hive from something you drive into something you supervise — the full pattern is in [scheduling autonomous agent missions](/blog/scheduling-autonomous-agent-missions/).
+The triggers tab. **Schedules** run a prompt on an interval or on chosen weekdays at a set time, and the prompt is sent word for
+word on every run. **Context** rules decide what happens as an agent's memory fills up. **Webhooks** let outside systems post work
+in, and **organisation** lets a teammate's office send work to yours. Together with the board, triggers turn the office from
+something you drive into something you supervise. The full pattern is in
+[scheduling autonomous agent missions](/blog/scheduling-autonomous-agent-missions/).
-## Floor or board? A simple rule
+## Should you watch the floor or the Command Center?
-Both views are the same hive, so this is about attention, not data:
+Both show the same office, so this is about attention, not data:
-- **Watch the floor** when you care about *one agent right now* — you just dispatched something, you want to type into a session, or you want ambient awareness while doing other work. (Also: it's charming.)
-- **Watch the board** when you care about *the fleet over time* — what's queued, what's blocked on what, who's near budget, whether CI is green, what fires tonight.
+- **Watch the floor** for one agent right now: you just sent work, you want to type into a session, or you want ambient
+ awareness while you do something else. It is also, admittedly, charming.
+- **Watch the Command Center** for the whole fleet over time: what is queued, what is blocked on what, who is close to budget, and
+ what runs tonight.
-In practice: floor open while you work, Command Center when you check in. Morning check-in is the board — done column, spend per agent, blocked tasks, next scheduled missions. That's a two-minute read that would take twenty minutes of terminal-scrolling without it. And when something needs a human decision, escalations land in the [approvals queue](/blog/human-in-the-loop-approving-ai-agents/) rather than hiding in a scrollback buffer.
+In practice, keep the floor open while you work and check the Command Center when you check in. A morning check in takes two
+minutes: the done column, the ask me tab, spend per agent, and what is scheduled next. When something needs a human decision, it is
+already waiting for you on [the ask me tab](/blog/human-in-the-loop-approving-ai-agents/) rather than buried in a scrollback.
## Try it
-The Command Center ships in the current release, free and open source. [Download Munder Difflin](https://github.com/chaitanyagiri/munder-difflin/releases/latest) — and if the board earns its place in your morning routine, a [star on GitHub](https://github.com/chaitanyagiri/munder-difflin) helps more people find it.
+The Command Center is part of the free classic office. [Download Munder Difflin](https://munderdiffl.in/), and if the board earns a
+place in your morning routine, [a star on GitHub](https://github.com/chaitanyagiri/munder-difflin) helps other people find it.
diff --git a/blog/src/posts/deploy-a-blog-writer-agent.md b/blog/src/posts/deploy-a-blog-writer-agent.md
index c48326fd..a9fbb043 100644
--- a/blog/src/posts/deploy-a-blog-writer-agent.md
+++ b/blog/src/posts/deploy-a-blog-writer-agent.md
@@ -1,160 +1,138 @@
---
-title: "Deploy a Blog-Writer Agent: The One That Wrote This Post"
-description: "How Munder Difflin's blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own."
+title: "Deploy a Blog Writer Agent: The One That Wrote This Post"
+description: "How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own."
date: 2026-06-10
-updated: 2026-08-20
+updated: 2026-09-10
category: use-cases
categoryLabel: Use Cases
type: Non-technical
primaryKeyword: "automated blog writer agent"
-secondaryKeywords: ["ai blog automation", "content agent", "multi-agent blogging", "automated content pipeline"]
+secondaryKeywords: ["ai blog automation", "content agent", "multi-agent blogging", "automated content pipeline", "ai blog writing workflow"]
tags: ["Use Cases", "Automation", "Content", "Multi-Agent", "Open Source"]
author:
name: Chaitanya Giri
initials: CG
faq:
- q: "Did an AI actually write the Munder Difflin blog?"
- a: "Yes — most of it. A writer agent in the hive drafts each post from a topic brief and a house-style reference, Michael integrates it as the single committer, Eleventy builds the markdown into the static site, and a human approves the final deploy to munderdiffl.in. This very post is an example of that pipeline running — and as of August 2026, the same office also draws every post's illustrations as code."
- - q: "Is the blog-writer agent fully autonomous?"
- a: "Almost — it's deliberately human-gated at one point: publish. The agent drafts and self-reviews against a style reference in an isolated worktree; the orchestrator integrates and builds; a person reviews the diff and approves the deploy. That keeps the volume high and hands-off without putting an unreviewed post on the live domain."
- - q: "How do I build my own blog-writer agent?"
- a: "Give one agent in the hive three things: a topic or brief, a house-style reference (a few of your best existing posts), and write access to an isolated worktree. Let it draft, have a reviewer agent or the orchestrator check it, then human-approve the build and deploy. Munder Difflin gives you the worktree isolation, single-committer git, skills, and approval queue out of the box."
- - q: "Why use a multi-agent hive instead of one prompt to write blog posts?"
- a: "One prompt writes one post. A hive runs a content function: a writer drafts, a reviewer checks, the orchestrator integrates and builds, and a scheduled mission can fire the loop on a cadence — so you get a steady, compounding stream of on-topic posts rather than a single output you have to re-prompt for each time."
+ a: "Mostly, yes. Agents in our own Munder Difflin office draft posts from a brief and a house style, and the illustrations are drawn in code. The change goes up as a pull request, a person reads it and merges it, and the merge publishes it. This post was rewritten that way in September 2026."
+ - q: "Is the blog writer agent fully autonomous?"
+ a: "Everything before publishing is. Research, drafting, checks and the pull request run without a person. Publishing waits for a human merge on purpose, so nothing unreviewed reaches the live site."
+ - q: "How do I build my own blog writer agent?"
+ a: "Give one agent a brief, a house style reference and its own workspace. Have a second agent or a checklist review the draft, then send it wherever a human approves changes, like a pull request. Put the loop on a schedule once a few posts come out right."
+ - q: "What makes an AI written post worth reading?"
+ a: "Something only you can say: a command you ran, a screenshot of the real screen, a result you measured and dated, a mistake you made. Without that, a draft is a summary of whatever already ranks, and readers can tell."
+ - q: "Why use a team of agents instead of one prompt?"
+ a: "One prompt writes one post. A team runs a content function: one agent researches, one drafts, one reviews, and a schedule starts the loop again. You stop prompting for posts and start reviewing them."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p>The Munder Difflin blog is written by an
-<strong>automated writer agent</strong> living in the hive. The loop: a writer agent drafts a post in an
-<strong>isolated worktree</strong> from a topic brief + a house-style reference; <strong>Michael</strong>
-integrates it as the single committer; <strong>Eleventy</strong> builds
-<code>blog/src/posts</code> → <code>docs/blog</code>; and a human approves the deploy to
-<strong>munderdiffl.in</strong>. The outcome is a steady, on-topic stream — <strong>well over a hundred
-posts and counting</strong>, every one now illustrated by code the same office wrote.
-<em>This post was made that way.</em></p></div>
+<div class="callout tldr"><span class="ic">TL;DR</span><p>The blog you are reading is <strong>mostly
+written by agents</strong> in our own office. An agent takes a <strong>brief</strong> and a <strong>house
+style</strong>, drafts in its <strong>own worktree</strong>, the draft gets checked, the pictures are
+<strong>drawn in code</strong>, and the change goes up as a <strong>pull request</strong>. A person merges
+it, and the merge publishes it. <em>This post was rewritten exactly that way.</em></p></div>
-Here's a fact that's either a confession or a flex, depending on how you read it: the blog you're reading
-is mostly written by one of our own agents. Not "AI-assisted." Not "drafted then heavily rewritten." A
-**writer agent** in the Munder Difflin hive takes a brief, drafts a full post, and hands it down a pipeline
-that ends — after one human nod — on the live site.
+Here is a fact that is either a confession or a flex: most of this blog is written by agents. Not "AI
+assisted" in the vague marketing sense. Agents in a Munder Difflin office take a brief, write the post and
+open the pull request. A human does one thing: reads it and merges it.
-This very post is an instance of that system working. So let me do the most on-brand thing possible and use
-it as the worked example. Here's how the blog-writer agent is automated, why the outcome compounds, and how
-to stand up your own.
+This post describes that pipeline, and in September 2026 the pipeline rewrote this post, so it doubles as
+the worked example. Here is how it runs and how to build your own.
-## The outcome first: a library, not a launch post
+## What does a blog writer agent actually do?
-Before the *how*, the *why it matters*. This blog now holds **well over a hundred published posts** —
-the topic chips at the top of [the index](/blog/) show the live counts. They're not filler — they
-cluster into a real content strategy: guides, internals deep-dives, concept explainers, comparisons,
-orchestration patterns, and a pointed story set, including [why we built
-Munder Difflin](/blog/why-we-built-munder-difflin/) and the
-[launch-week retros](/blog/what-reddit-told-us-about-munder-difflin/). The thought-leadership thread
-does the heavy SEO lifting — the [multi-agent cost playbook](/blog/the-multi-agent-cost-playbook/),
-[compressing agent memory](/blog/compressing-agent-memory/),
-[context engineering for AI agents](/blog/context-engineering-for-ai-agents/), and more.
+It turns a brief into a finished change that is ready for review. It reads the brief and the house style,
+researches the topic, drafts the post with its frontmatter and the questions a reader would type, checks its
+own work, and hands the result on. It does not publish.
-That's the whole point of automating the writer: **volume that stays on-topic compounds**. A hundred-plus
-internally-linked, keyword-targeted posts is a discoverability moat you cannot hand-write at a startup's
-spare-time pace. When someone — or an AI answer engine — searches "single-committer git multi-agent" or
-"compressing agent memory," there's a post for that, and it links to five neighbors. The blog-writer agent
-is how a one-person project publishes like a content team.
+## How does the pipeline work, end to end?
-And crucially: it's *largely hands-off*. The expensive part of blogging isn't typing — it's the discipline
-to keep shipping. An agent has no problem with discipline.
+Five stages, and only the last one needs a person.
-{% img "note-1", "The content function: brief in, draft in a worktree, one committer, one build, one human gate — on a timer." %}
+### 1. Brief
-## The pipeline, end to end
+Every post starts from a brief: the topic, the question a reader is trying to answer, and the angle. Briefs
+come from a backlog, so the agent is always writing something somebody is looking for, not whatever it
+happens to feel like.
-The writer agent isn't a magic monolith. It's one role in a hive, and the post moves through the same
-machinery any work does. Four stages:
+### 2. Draft, in its own worktree
-### 1. Draft — in an isolated worktree
+The writer works in its **own git worktree**, a separate checkout, so a half written draft never collides
+with anyone else's files. ([Why that isolation matters](/blog/claude-code-git-worktrees-vs-hive/).) It reads a
+handful of existing posts for structure, and it can carry the house style as an installed **skill**, a
+checklist it rereads on every draft instead of a prompt we hope it remembers.
-The writer agent gets a **brief** (a topic + intent, usually straight from our SEO backlog) and a
-**house-style reference**: a handful of existing posts to mirror for voice, front-matter, and structure. It
-works in its **own git worktree** — a separate working directory so its in-progress draft never collides
-with anyone else's files. (We wrote about why that isolation matters in
-[git worktrees vs a hive](/blog/claude-code-git-worktrees-vs-hive/).) Since v0.4.4 the writer can also
-carry an installed **skill** — a house-style checklist it re-reads on every draft, instead of a prompt
-we hope it remembers.
+Our house style is short and blunt: no dashes, get to the point, one light joke at most, and nothing we cannot
+back up. The output is a single markdown file, and its filename becomes the URL.
-The agent reads the reference posts, picks internal links to neighbors, writes the front-matter (title,
-description, category, keywords, FAQ schema), and drafts ~1,000–1,400 words of body. It self-checks
-against the style reference before handing off. The output is a single `.md` file in
-`blog/src/posts/` — filename becomes the URL slug.
+### 3. Check
-### 2. Integrate — the orchestrator is the single committer
+A second pass holds the draft against the brief and the style. Are the facts sourced and dated? Does every
+heading answer a question a real person asks? Do the links work? A draft that fails goes back with the reason.
-The writer **never commits**. It writes a plain file; **Michael** owns every commit. This is
-the [single-committer pattern](/blog/single-committer-git-pattern/) — agents write files, one process
-serializes all the git, so parallel agents never race on `.git/index.lock` and the repo stays a clean audit
-log. The orchestrator picks up the finished draft, reviews routing, and commits it into the real tree.
+### 4. Pictures, drawn in code
-### 3. Build — Eleventy turns markdown into a site
+Every hero image and inline sketch on this blog is [drawn as code by an agent](/blog/an-agent-redesigned-this-blog/):
+a library of SVG parts, a set of scene layouts and one spec file, rendered in a headless browser. No image API
+bills. One manifest lists every image, and a post shows a designed placeholder until its drawings land, so a page
+never renders broken.
-A static [Eleventy](https://www.11ty.dev/) build compiles `blog/src/posts` → `docs/blog`. Dropping one
-markdown file is the entire authoring action: the build auto-adds the post to the index, its topic page,
-each tag page, the sitemap, and the RSS feed — plus the SEO that's already wired (canonical URLs, OpenGraph,
-`BlogPosting` + `FAQPage` JSON-LD). No other file gets touched. That's deliberate: the agent's job is *write
-one file correctly*, and the build does the rest.
+### 5. Pull request, then one human merge
-**And since August 2026, the pictures are part of the pipeline too.** Every post's hero and inline
-sketches are [drawn as code by an agent](/blog/an-agent-redesigned-this-blog/) — an SVG parts library,
-composition archetypes, and one spec file, rendered through a headless browser for exactly $0 in image
-APIs. One `media.json` manifest drives every image on the blog; a post ships with designed placeholders
-until its drawings land, so nothing is ever broken.
+The agent opens a pull request with the new post and the rebuilt pages. A person reads it and merges it. Merging
+to main rebuilds the site with [Eleventy](https://www.11ty.dev/), and GitHub Pages serves it at munderdiffl.in/blog.
+The build adds the post to the index, its topic and tag pages, the sitemap and the RSS feed, with the structured
+data already wired in.
-### 4. Deploy — human-gated, on purpose
+That split is the whole lesson: **writing is automated, publishing is reviewed.** You get an agent's stamina and
+a human's final read. Nobody wants a confidently wrong claim on their front page.
-This is the one stage that is **not** automated, by design. The build output under `docs/blog` is served by
-GitHub Pages at **munderdiffl.in/blog**. Before that goes live, a person reviews the diff and approves the
-deploy. The orchestrator [escalates exactly this kind of "publish to the world"
-action](/blog/how-the-god-orchestrator-works/) to the human-approval queue rather than shipping it itself.
+{% img "note-1", "Brief in, draft in its own worktree, one human gate at the very end." %}
-The split is the lesson: **drafting and integration are autonomous; publishing is human-gated.** You get the
-throughput of an agent and the safety of a final human read. Nobody wants a hallucinated claim on their
-front page — so that one gate stays manual while everything upstream runs hands-off.
+## How do you build your own blog writer agent?
-{% img "note-2", "One gate stays human on purpose: the deploy. Everything upstream of it runs itself." %}
+Five steps, and none of them needs our exact stack.
-## Build your own blog-writer agent
+**1. Keep a backlog of briefs.** One topic, one reader question and one angle per brief.
-You don't need our exact stack. The pattern transfers to any static site or CMS. Here's the recipe.
+**2. Write the house style down.** Point the agent at three to five of your best posts and list the rules you
+actually care about. Install it as a skill so every draft rereads it. A writer with a sharp reference produces
+something publishable. A writer without one produces beige.
-**1. Give it a brief.** One topic, the search intent, and the angle. Pull it from a keyword backlog so the
-agent is always writing something discoverable, not random.
+**3. Give the draft its own workspace.** A worktree, a branch or a scratch folder. In Munder Difflin this is the
+git isolation toggle on the agent.
-**2. Give it a house-style reference.** This is the highest-leverage input. Point the agent at three to five
-of your *best* existing posts and tell it to mirror their front-matter, voice, length, and link density —
-or install that guidance as a **skill**, so every draft re-reads it. A writer with a strong reference
-produces something publishable; a writer without one produces generic AI slop.
+**4. Check, then gate publishing.** A reviewer agent or a checklist first, then a pull request or an approval that
+only a human can complete.
-**3. Isolate the draft.** Let the agent write in its own worktree (or branch, or scratch directory) so an
-in-flight draft can't clobber live files. In Munder Difflin this is a per-agent Git isolation toggle.
+**5. Put it on a schedule.** Once a few posts come out right, open the Triggers tab and create a schedule: a label,
+who it goes to, and the prompt that starts the loop. It is the same move that stood up
+[an hourly PR reviewer](/blog/one-prompt-automated-pr-review/) for us.
-**4. Review before publish.** Have a reviewer agent or your orchestrator check the draft against the brief,
-then put the **deploy** behind a human approval. Draft and integrate automatically; publish on a click.
+{% img "note-2", "One gate stays human on purpose: publishing. Everything before it runs on its own." %}
-**5. Make it recurring.** The real unlock is a scheduled mission: fire the writer on a cadence, feeding it
-the next backlog item each time. One prompt to the orchestrator stood up [an hourly PR
-reviewer](/blog/one-prompt-automated-pr-review/) for us the same way — automation that just keeps running.
-Point that same scheduling at content and the blog writes itself on a timer.
+## What makes an AI written post worth reading?
-## The meta-point
+Something only you can say. A draft built only from what already ranks is a summary of other people's posts, and
+readers smell it quickly. Give the writer real material: a command you ran, a screenshot of the actual screen, a
+number you measured and dated, a mistake you made and fixed. The agent brings structure and stamina. You bring the
+proof.
-A multi-agent hive isn't only for code. Once you have a writer that drafts, an orchestrator that integrates
-and commits, a build that publishes, and one human gate, you have a **content function** — not a one-off
-prompt. The difference shows up as a library instead of a launch post.
+Two more rules we follow:
-So consider this post Exhibit A. It was briefed, drafted in a worktree against a style reference, integrated
-single-committer, Eleventy-built, illustrated by code, and human-approved to the domain — the exact loop it
-describes. The system is, quite literally, writing about itself.
+- **Write for a job someone is doing.** "How do I run pi on a local model?" beats "Some thoughts on local AI".
+- **Refresh instead of duplicating.** When the product changes, update the post that already exists rather than
+ writing a second one that competes with it. This rewrite is that rule in action.
+
+## The meta point
+
+A multi agent office is not only for code. Once you have a writer that drafts, a check that catches mistakes, a build
+that publishes and one human gate, you have a content function instead of a one off prompt.
+
+So this post is exhibit A: briefed, rewritten in a worktree against a house style, checked, and handed to a human as
+a pull request.
---
-Munder Difflin runs a [hive of agents on ten engines](https://munderdiffl.in/#how) on
-your own machine — with isolated worktrees, single-committer git, skills, and a human-approval queue built
-in, so an agent can draft and integrate while you keep the one gate that matters.
-[Download Munder Difflin](https://munderdiffl.in/#install) to put a blog-writer (or any worker) on your
-floor; it's free and open source.
+Munder Difflin runs an office of agents on twelve terminal CLIs on your own machine, with worktrees, skills, schedules
+and an ASK ME board for the decisions that need you. [Download it free](https://munderdiffl.in/) and put a writer, or
+any other worker, on your floor.
diff --git a/blog/src/posts/how-to-hire-from-the-agent-gallery.md b/blog/src/posts/how-to-hire-from-the-agent-gallery.md
index 8089b9e1..e8fe6910 100644
--- a/blog/src/posts/how-to-hire-from-the-agent-gallery.md
+++ b/blog/src/posts/how-to-hire-from-the-agent-gallery.md
@@ -1,84 +1,111 @@
---
title: "How to Hire From the Agent Gallery"
-description: "A practical guide to Munder Difflin's Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself."
+description: "A practical guide to Munder Difflin's Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself."
date: 2026-07-03
+updated: 2026-09-10
category: guides
categoryLabel: Guides
type: Non-technical
primaryKeyword: "agent gallery"
-secondaryKeywords: ["hire an ai agent", "off-the-shelf agent roles", "import agent manifest", "munder difflin hires", "ready-made ai coworkers"]
+secondaryKeywords: ["hire an ai agent", "ready made ai agent roles", "import agent manifest", "munder difflin hires", "ai coworker templates"]
tags: ["Guides", "Multi-Agent", "Agent Design", "Getting Started", "Security"]
author:
name: Chaitanya Giri
initials: CG
faq:
- q: "What is the Agent Gallery?"
- a: "It's a free gallery of ready-made agent roles for Munder Difflin at munderdiffl.in/hires — no login, no trackers. Each role, called a hire, is a small JSON manifest that packages a configured agent: name, avatar, provider, model, flags, a goal, capability tags, and a token budget. It ships with six off-the-shelf hires you can import in one click and adapt to your project."
- - q: "Does importing a hire from the gallery start an agent automatically?"
- a: "No. Import only pre-fills the Add-Agent modal behind an explicit imported banner. You review every field, change anything you like, and the agent spawns only when you click spawn. There is no code path that runs an agent on import — the human is always the spawn gate."
- - q: "Is it safe to import a hire manifest someone sent me?"
- a: "The import pipeline is built on the assumption that it isn't — every manifest is validated as untrusted input. There is no executable field, so a manifest can never name a program to run; command flags pass a default-deny allowlist; and deep-link fetches are https-only and bounded. Even after all that, you still review the pre-filled form before anything spawns. Treat the goal text with the same skepticism you'd give any file from the internet."
- - q: "Can I change the engine or model a gallery hire uses?"
- a: "Yes. The manifest is a starting point, not a lock. In the Add-Agent modal you can switch the hire onto any engine you have — Claude Code, Antigravity, Codex, OpenCode, Crush, pi.dev, or GitHub Copilot CLI — pick a different model, and on local-capable engines use the OSS-model quick-picks to run it on open models via Ollama or a BYOK provider."
- - q: "What should I customize before spawning a hire?"
- a: "Four things. Identity: the name and avatar you'll recognize on the floor. Workspace: the working directory it operates in, ideally with the git-isolation toggle on so it gets its own worktree. Engine: the CLI and model that match the job's difficulty and your budget. Briefing: rewrite the goal so it names your repo, your conventions, and what done means — a generic goal produces generic work."
- - q: "How much does hiring from the gallery cost?"
- a: "The gallery and the app are free and open source (MIT-licensed code). You pay only for the model usage of whatever CLI or key the agent runs on. Every hire carries a token budget field, and the harness pairs per-agent budgets with live fleet monitoring and a circuit breaker, so an imported role can't quietly outspend the ceiling you gave it."
+ a: "A free gallery of ready made agent roles for Munder Difflin at munderdiffl.in/hires. Each role, called a hire, is one JSON file describing a configured agent: its name and avatar, engine and model, flags, goal, skills and token budget. As of September 2026 it lists 80 roles."
+ - q: "How do I import a hire?"
+ a: "Download the role's .json from its card, open Munder Difflin, click Add agent, then import hire, and pick the file. The form fills in with every field from the manifest."
+ - q: "Does importing a hire start an agent?"
+ a: "No. Importing only fills in the Add agent form for you to review. Nothing spawns until you click spawn yourself."
+ - q: "Is it safe to import a manifest someone sent me?"
+ a: "The import treats every manifest as untrusted. A manifest cannot name a program to run or carry shell syntax, and you review the final command before anything starts. Still read the goal text with the care you would give any file from the internet, because it is someone else's prompt."
+ - q: "Can I change the engine or model a hire uses?"
+ a: "Yes. The manifest is a starting point. In Add agent you can switch the role onto any of the twelve supported CLIs you have installed, pick a different model, and on OpenCode, Crush or Pi choose one of the open model quick picks."
+ - q: "What does hiring from the gallery cost?"
+ a: "The gallery and the classic app are free. You pay only for the model usage of the CLI or key the agent runs on, and every agent's token budget is enforced by the app with a circuit breaker behind it."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p>The <strong>Agent Gallery</strong> (<a href="https://munderdiffl.in/hires/">munderdiffl.in/hires</a>) is where you hire ready-made AI coworkers for Munder Difflin — it ships with <strong>six off-the-shelf hires</strong>. A hire is a small JSON manifest: provider, model, flags, goal, capabilities, token budget. Importing one — by <code>munderdifflin://hire</code> link or a local file — <strong>never runs anything</strong>: it only pre-fills the Add-Agent modal behind an "imported" banner, and the manifest is <strong>validated as untrusted input</strong> along the way. You review, customize the <strong>identity, workspace, engine, and briefing</strong>, and <em>you</em> click spawn. Five minutes from browsing to a working agent on your floor.</p></div>
+<div class="callout tldr"><span class="ic">TL;DR</span><p>The <strong>Agent Gallery</strong> at
+<a href="https://munderdiffl.in/hires/">munderdiffl.in/hires</a> holds <strong>80 ready made roles</strong> for Munder Difflin.
+Download a role's <code>.json</code>, click <strong>Add agent</strong>, then <strong>import hire</strong>, and the form fills in
+for you. <strong>Nothing runs on import.</strong> You review every field, customise the <strong>identity, workspace, engine and
+briefing</strong>, and you click spawn. About five minutes from browsing to a working agent on your floor.</p></div>
-The slowest part of running a multi-agent office isn't running the agents. It's the blank Add-Agent form: which engine, which model, which flags, and — the one that actually stalls people — what is this thing's *job*? The Agent Gallery exists so you never start from that blank box. You start from a role somebody already got right, and you edit.
+The slowest part of running an office of agents is not the agents. It is the blank Add agent form: which engine, which model, which
+flags, and the question that really stalls people, what is this agent's job? The gallery exists so you never start from a blank
+form. You start from a role somebody already got right, and you edit it.
-This is the practical walkthrough: browse, import, review, customize, spawn. (If you haven't installed the app yet, do [that first](/blog/how-to-install-and-use-munder-difflin/) — the gallery assumes you have a floor to hire onto.)
+This is the walkthrough as of Munder Difflin 0.5.2: browse, import, review, customise, spawn. Not installed yet?
+[Do that first](/blog/how-to-install-and-use-munder-difflin/), because the gallery assumes you have a floor to hire onto.
-## Step 1: Browse the gallery
+## Where is the Agent Gallery, and what is in it?
-The gallery lives at [munderdiffl.in/hires](https://munderdiffl.in/hires/) — a static page, no login, no trackers. It's stocked with **six off-the-shelf hires**, each a complete role: a PR reviewer, a docs writer, a QA enforcer, a security auditor, a token-spend auditor, and a migrations grinder, each fronted by a pixel avatar from the app's affectionate Office parody cast.
+At [munderdiffl.in/hires](https://munderdiffl.in/hires/), a static page with no login. It lists 80 roles, and they are not all
+engineers. Next to a PR reviewer, a QA enforcer and a security auditor you will find roles for research, data analysis, docs, design,
+customer support, sales outreach and marketing. Each card shows what the role is for, and you can search the page.
-Every card tells you what you're getting before you click anything: the role's goal, its capability tags, and which provider it's written for. Don't overthink the choice — you're picking a starting point, not signing a contract. Everything on the card is editable after import.
+Do not overthink the choice. You are picking a starting point, not signing a contract, and every field is editable after import.
{% img "note-1" %}
-## Step 2: Import the hire
+## How do you import a hire?
-Two ways in, both landing in the same place:
+Three steps:
-- **Deep link.** Click the hire button on a card and the `munderdifflin://hire` link opens the app directly.
-- **Local file.** Save the manifest JSON (or receive one from a teammate) and import it from disk.
+1. On the role's card, click **download** to save its `.json`.
+2. In Munder Difflin, click **Add agent**, then **import hire**.
+3. Pick the file you just downloaded.
-Either way, one thing is worth being precise about: **import never spawns anything.** What opens is the app's ordinary Add-Agent modal, pre-filled with the manifest's values and marked with an explicit **"imported" banner** so you know these fields came from outside. The agent starts only when you click spawn. A hire you got from the internet is inert data until a human acts on it.
+The Add agent form opens with every field filled in from the manifest. Because a hire is just a file, a teammate can also send you one
+directly, and you import it the same way.
-That's not an accident of UI — it's a security posture. A manifest arrives by clicked link or downloaded file, which makes it attacker-controlled bytes, and the import pipeline treats it that way: there's **no executable field** (the binary always comes from your own locally configured provider presets), command flags pass a **default-deny allowlist**, and the deep-link fetch is **https-only and bounded**. The full threat model is its own post: [treating a hire manifest as untrusted input](/blog/hire-manifest-untrusted-input/).
+The one thing to be precise about: **importing never spawns anything.** It only fills in the form. The agent starts when you click spawn,
+and not a moment before.
-## Step 3: Actually read the manifest
+## Why read the manifest before you spawn?
-The imported banner is your cue to slow down for thirty seconds. A hire manifest encodes the whole role — name, avatar, provider, model, command flags, goal, capability tags, and a token budget — and every one of those is now sitting in front of you as a form field. Read them like you'd read a job description before forwarding it to HR:
+Because the goal text is somebody else's prompt. The import treats every manifest as untrusted input: a manifest cannot name a program to
+run or carry shell syntax, so the program always comes from your own installed engine. What validation cannot vouch for is the goal,
+which is free text. So take thirty seconds:
-- **Goal text** is the prompt this agent will work from. It's free text from someone else, which makes it the one field validation can't vouch for. If anything in it looks off, rewrite it.
-- **Model and flags** tell you what it costs to run and how it behaves. The allowlist keeps flags safe; it doesn't make them right for you.
-- **Token budget** is the spend ceiling the author suggested. Set it to what *you* are comfortable with — the harness enforces per-agent budgets with live fleet monitoring and a circuit breaker behind it.
+- **Goal.** The prompt this agent will work from. If anything in it looks off, rewrite it.
+- **Model and flags.** What the agent costs to run and how it behaves.
+- **Token budget.** The ceiling the author suggested. Set the one you are comfortable with, because the app enforces it.
+- **Skills.** The skills this hire activates. Keep only what the job needs.
+
+The full threat model is its own post: [treating a hire manifest as untrusted input](/blog/hire-manifest-untrusted-input/).
{% img "note-2" %}
-## Step 4: Customize the four things that matter
+## What should you customise before spawning?
+
+Four edits turn a generic role into your hire.
-A gallery hire is deliberately generic. Four edits turn it into *your* hire:
+**Identity.** Rename it and pick the avatar you want to see walking the floor. When six agents are working at once, you triage by face and
+name, so this matters more than it sounds.
-**Identity.** Rename it and pick the avatar you want to see walking the floor. This sounds cosmetic; it isn't. When six agents are working at once, you triage by face and name.
+**Workspace.** Point it at the project it should work in and turn on git isolation, so it gets its own worktree. Agents that share one
+checkout collide on branches, and agents with worktrees do not. ([Why worktrees](/blog/claude-code-git-worktrees-vs-hive/).)
-**Workspace.** Point the hire at the repo it should work in, and flip on the **git isolation** toggle so it auto-provisions its own worktree on spawn — agents that share a checkout collide on branches; agents with worktrees don't ([why worktrees, in depth](/blog/claude-code-git-worktrees-vs-hive/)).
+**Engine.** The manifest suggests an engine, but the picker is yours. Run the role on any of the twelve supported CLIs you have installed.
+If the job is routine, OpenCode, Crush and Pi offer open model quick picks, so it can run on a local or cheap model.
-**Engine.** The manifest names a provider, but the modal's picker is yours: run the role on Claude Code, Antigravity, Codex, OpenCode, Crush, pi.dev, or GitHub Copilot CLI. Match the engine to the job — and if the role is routine enough, the local-capable engines surface **OSS-model quick-picks** so it can run on open models for pennies.
+**Briefing.** Rewrite the goal until it names your project, your conventions and what done looks like. "Review pull requests" becomes "review
+PRs on this repo, enforce the lint config, and never approve without a green typecheck". This is the most valuable minute of the whole flow.
-**Briefing.** Rewrite the goal until it names your project, your conventions, and what "done" looks like. This is the highest-leverage sixty seconds of the whole flow: a hire that arrives with "review pull requests" leaves with "review PRs on repo X, enforce the lint config, never approve without a green typecheck."
+## What happens after you spawn?
-One more gate you'll meet after spawning: hires carry a manifest of allowed **skills and MCP servers**, default-deny over a shared catalog. Anything the hire wants to use surfaces in a consent UI for you to grant — imported input is reviewed, never auto-granted.
+The agent takes a desk on the floor, gets its own mailbox and long term memory, and is ready for work straight away. Assign it tasks on the
+kanban, or tell Michael what you need and let him [route the work](/blog/how-the-god-orchestrator-works/).
-## Step 5: Spawn, and put it to work
+## Can you add your own hire?
-Click spawn. The agent appears at a desk, joins the hive with its own mailbox and long-term memory, and is immediately routable: assign it kanban tasks, or just tell the GOD orchestrator what you need and let [Michael route the work](/blog/how-the-god-orchestrator-works/).
+Yes. A hire is one JSON file that follows the hire manifest spec. Start from any card's JSON, check your version with the validator on the
+gallery page, then import it with Add agent and import hire. Manifests are plain files, so you can share them anywhere.
-From browsing a card to a working coworker is genuinely a five-minute path — and every minute of it kept you in the loop: you read the manifest, you set the budget, you clicked spawn.
+---
-Grab the latest build from the [releases page](https://github.com/chaitanyagiri/munder-difflin/releases/latest), and if the gallery saves you a blank-form afternoon, a [GitHub star](https://github.com/chaitanyagiri/munder-difflin) is appreciated.
+Grab the latest build from [munderdiffl.in](https://munderdiffl.in/), and if the gallery saves you an afternoon of blank forms,
+[a GitHub star](https://github.com/chaitanyagiri/munder-difflin) is appreciated.
diff --git a/blog/src/posts/how-to-install-and-use-munder-difflin.md b/blog/src/posts/how-to-install-and-use-munder-difflin.md
index 511e69ee..fbc6d617 100644
--- a/blog/src/posts/how-to-install-and-use-munder-difflin.md
+++ b/blog/src/posts/how-to-install-and-use-munder-difflin.md
@@ -1,8 +1,8 @@
---
title: "How to Install and Use Munder Difflin: A Beginner's Guide"
-description: "A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you'll ever run, and how to install and set it up on macOS, Windows or Linux."
+description: "A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you'll ever run, and how to install and set it up on macOS, Windows or Linux."
date: 2026-06-05
-updated: 2026-08-27
+updated: 2026-09-10
category: guides
categoryLabel: Guides
type: Non-technical
@@ -16,11 +16,11 @@ faq:
- q: "Do I need to know how to code to use Munder Difflin?"
a: "No. The onboarding asks whether you are technical or not on its very first screen, and the non-technical path replaces every piece of jargon with plain language. You will open a terminal exactly once, to install and log into your AI engine, and you can copy and paste those commands."
- q: "Does Munder Difflin cost anything?"
- a: "The app is free and open source. What you may pay for is the AI engine behind it. Antigravity is available at no charge and OpenCode has a free path, so you can run the whole thing without paying anything. If you already pay for ChatGPT or Claude, you can use those subscriptions instead."
+ a: "The classic app is free and open source, with unlimited local agents. Pro and Teams are optional paid plans with a 14 day trial. What you may pay for is the AI engine behind it. Antigravity is available at no charge and OpenCode has a free path, so you can run the whole thing without paying anything. If you already pay for ChatGPT or Claude, you can use those subscriptions instead."
- q: "Can these agents really change files on my computer?"
a: "Yes, and that is the point, so it is worth understanding before you start. An agent can read files, write files and run commands in the folders you give it. You choose how much freedom it has during setup, and you can set it to ask permission before every change."
- q: "Which operating systems does Munder Difflin run on?"
- a: "macOS, Windows and Linux. macOS ships as a .dmg, Windows as an installer and a portable build, and Linux as an AppImage and a .deb."
+ a: "macOS, Windows and Linux. macOS ships as one universal .dmg for Apple Silicon and Intel, Windows 10 and 11 as a setup installer or a portable build, and Linux as an AppImage."
- q: "Is my code or data sent anywhere?"
a: "The app runs on your machine and stores its files there. Your prompts and the files an agent reads do go to whichever AI engine you picked, the same as if you used that tool directly. Anonymous usage telemetry is opt-out and every event is listed publicly in TELEMETRY.md."
---
@@ -44,7 +44,8 @@ working on something you asked for. One of them is your clone, the boss of the f
you ask for, breaks it into jobs, and hands those jobs to the others. You talk to your clone, and
your clone manages everyone else.
-The app is free and open source. It runs on your machine, and it keeps its files on your machine.
+The app is free and open source. It runs on your machine, and it keeps its files on your machine. There are
+optional paid plans, Pro and Teams, but nothing in this guide needs them.
The thing it is not: a chatbot. You are not sitting there typing and waiting for replies. You give
your clone a job, close the laptop, and come back to work that was done while you were gone.
@@ -279,16 +280,19 @@ straight from the
{% img "shot-download", "The download page. Pick the file that matches your computer." %}
-**macOS.** Download the `.dmg`. Open it, drag Munder Difflin into Applications. The first time you
-open it, macOS may say it cannot verify the developer. Right-click the app icon, choose Open, then
-click Open in the dialog. You only do this once.
+**macOS.** Download the universal `.dmg`, which runs on Apple Silicon and Intel. Open it and drag
+Munder Difflin into Applications. The app is signed and notarized by Apple, so the first launch only
+shows the usual question about opening an app downloaded from the internet. Click Open.
-**Windows.** Download the `.exe` installer and run it. Windows SmartScreen will likely warn you,
-because the installer is not yet EV code signed. Click More info, then Run anyway. If the installer
-fails to start at all, download the portable build instead, which needs no installation.
+**Windows.** Download the setup `.exe` and run it. If Windows SmartScreen shows a warning, click
+More info, then Run anyway. If the installer will not start at all, download the portable `.exe`
+from the release instead, which needs no installation.
-**Linux.** Download the `AppImage`, make it executable (`chmod +x` on the file) and run it. There is
-a `.deb` if you prefer to install it properly.
+**Linux.** Download the `.AppImage`, make it executable (`chmod +x` on the file) and run it.
+
+**Want to be sure the file is genuine?** Every release carries a `SHA256SUMS.txt`. Run
+`shasum -a 256` on the file you downloaded (or `Get-FileHash` in Windows PowerShell) and check that
+the value matches that file's line.
## Step 4: Onboarding, six screens
@@ -434,10 +438,11 @@ it alone. The default is sensible and most people never touch it.
Seven sections. These are the ones that matter early:
- **General.** Change your home folder, switch between technical and plain language, pick your
- language. As of v0.4.6 the app runs in English, Chinese and Arabic.
+ language. The app runs in English, Simplified Chinese and Arabic.
- **Prerequisites.** Checks whether your engines are actually installed and logged in. **This is the
first place to look when something is not working.**
-- **Agents & Models.** Which engines are available and which models each uses.
+- **Agents & Models.** Which engines are available and which models each uses. New models show up
+ without an app update, because the model pickers read the latest list when the app launches.
- **Autonomy & Budgets.** The permission choice from onboarding, plus spending limits per agent.
Worth setting a cap before leaving anything running overnight.
- **Connections.** Slack, webhooks, MCP and the REST API.
@@ -464,10 +469,11 @@ sign-in.
**"Engine not installed" during onboarding, but you installed it.** The app looks for the command on
your system path. Close the app completely and reopen it, since it reads your environment at launch.
-**Windows blocks the installer.** SmartScreen flags it because the installer is not EV code signed.
-Choose More info, then Run anyway, or use the portable build.
+**Windows blocks the installer.** If SmartScreen flags it, choose More info, then Run anyway, or use
+the portable build.
-**macOS says the developer cannot be verified.** Right-click the app, choose Open, then Open again.
+**macOS asks whether to open an app downloaded from the internet.** That is normal for anything you
+download. Click Open.
**Nothing is happening.** Check whether message delivery is paused in the Command Center. Paused
means queued work is being held for every agent, and nothing is lost when you switch it back on.
diff --git a/blog/src/posts/qm-vs-munder-difflin.md b/blog/src/posts/qm-vs-munder-difflin.md
index e5fcfea2..d919ca8d 100644
--- a/blog/src/posts/qm-vs-munder-difflin.md
+++ b/blog/src/posts/qm-vs-munder-difflin.md
@@ -1,116 +1,106 @@
---
-title: "qm vs Munder Difflin: YC's Multiplayer Harness or a Local-First Agent Office?"
-description: "An honest comparison of YC's qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE."
+title: "qm vs Munder Difflin: YC's Multiplayer Harness or an Office of Your Clones?"
+description: "An honest September 2026 comparison of YC's qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them."
date: 2026-08-06
+updated: 2026-09-10
category: comparisons
categoryLabel: Comparisons
type: Non-technical
primaryKeyword: "qm vs munder difflin"
-secondaryKeywords: ["yc qm alternative", "qm claude code", "qm agent harness", "multiplayer agent harness", "local first agent orchestration", "qm y combinator agents"]
+secondaryKeywords: ["yc qm alternative", "qm agent harness", "multiplayer agent harness", "local first agent orchestration", "qm y combinator agents", "self hosted agent harness"]
tags: ["Comparisons", "Multi-Agent", "Tools", "Claude Code", "Open Source"]
author:
name: Chaitanya Giri
initials: CG
faq:
- q: "What is qm?"
- a: "qm (github.com/yc-software/qm) is Y Combinator's open-source 'multiplayer agent harness for work' — a headless Node.js core with a web UI and a Slack plugin. Team members get agent 'employees' with isolated workspaces, shared skills with scope-based permissions, background crons and watches, and three security postures (Strict, Auto, Dangerous). It's MIT-licensed and installs via npm."
- - q: "What's the main difference between qm and Munder Difflin?"
- a: "Where the harness lives and what you can see. qm is multiplayer-first: a shared server your whole team reaches through Slack and the browser. Munder Difflin is local-first: a desktop app on your machine where every agent is a real CLI process in a real terminal you can watch, with a visual office floor, voice orchestration, and a built-in IDE. They cover the same core jobs — postures, scheduled work, Slack, skills — from opposite ends."
- - q: "Does Munder Difflin do what qm's security postures do?"
- a: "Yes, the same job with different controls: a floor-wide auto-mode toggle (the equivalent of switching between approval-gated and autonomous), per-agent pause/halt, voice-level confirm words for destructive actions, tool gating, a circuit breaker for runaway agents, and spend/scope escalation to a human. qm's Strict/Auto/Dangerous is org-level configuration; Munder Difflin's controls are per-floor and per-agent."
- - q: "Does qm have voice control, an IDE, or git visualization?"
- a: "Not as of this writing. qm's interfaces are Slack and a web UI. Munder Difflin ships realtime voice orchestration (a talking Michael with live floor context), a built-in Monaco IDE with commit history, branch compare and side-by-side diffs, markdown previews, real watchable terminals, and auto-update."
- - q: "Which one should I use?"
- a: "Use qm if your whole team wants one shared harness in Slack with org-level admin control. Use Munder Difflin if you want your agent office on your own machine — watchable, local-first, with voice and an IDE — riding the CLI subscriptions you already pay for. Both are MIT-licensed; some teams will genuinely want both: qm as the shared org layer, Munder Difflin as the personal floor."
- - q: "Is Munder Difflin older than qm?"
- a: "Munder Difflin has been shipping public releases since v0.2.0 in mid-2026 and is on v0.3.5 with a stable auto-updating release train, while qm is a newer project that grew popular quickly. Popularity and maturity are different axes — evaluate both against your workflow."
+ a: "qm (github.com/yc-software/qm) is Y Combinator's open source multiplayer agent harness for work, used in Slack and on the web. A company deploys it from its own deployment repository. Each person and each room gets scoped memory, files, permissions, crons and a durable sandbox, and Pi, OpenCode, Codex and Claude Code can all drive the same core. It is MIT licensed."
+ - q: "What is the main difference between qm and Munder Difflin?"
+ a: "Where the agents run and who they belong to. qm is a shared service a company deploys, and everyone reaches it through Slack and a browser. Munder Difflin is a desktop app where your agents run as real terminal processes on your own machine, coordinated by your clone, Michael."
+ - q: "Can Munder Difflin be multiplayer like qm?"
+ a: "Yes, in a different shape. The Munder Difflin Teams plan lets clones message each other, and every message is sealed for the device that opens it. Each person's clone still runs on their own machine under their own keys, and the relay carries messages it cannot read."
+ - q: "How do their safety controls compare?"
+ a: "qm has one security posture for the whole org: Strict pauses harness tool calls for human approval, Auto screens external data with a classifier, and Dangerous does neither. Munder Difflin sets autonomy on your floor: an ask first mode where agents pause for tool approval, per agent token budgets, a circuit breaker that steers, constrains and then stops a runaway agent, and an ASK ME board for decisions that need you."
+ - q: "Which should I use?"
+ a: "qm if your company wants one shared agent service in Slack, run by an admin on your own infrastructure. Munder Difflin if you want agents you can watch in real terminals on your own machine, across twelve CLIs, on the subscriptions you already pay for. Both are MIT licensed, so trying both costs you an afternoon."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p><strong>qm</strong> is YC's
-multiplayer agent harness — headless, org-shaped, living in <strong>Slack and a web
-UI</strong>, great when a whole team shares one fleet. <strong>Munder Difflin</strong> is
-the same idea grown from the other end: a <strong>local-first desktop office</strong> where
-every agent is a real CLI process in a terminal you can watch, with <strong>voice
-orchestration, a built-in IDE with git history, and auto-update</strong> — things a
-headless harness doesn't have. Same jobs, opposite ends of the wire. Both MIT-licensed.</p></div>
+<div class="callout tldr"><span class="ic">TL;DR</span><p><strong>qm</strong> is Y Combinator's
+multiplayer agent harness. A company deploys it, and everyone works with agents through <strong>Slack and a
+web app</strong>, each person and room with its own scoped memory, crons and sandbox. <strong>Munder
+Difflin</strong> runs an <strong>office of your clones on your own machine</strong>: real terminal CLIs you
+can watch, coordinated by Michael, with <strong>Teams</strong> so your clone can work with your teammates'
+clones. Same destination, opposite starting points. Both MIT licensed.</p></div>
-[qm](https://github.com/yc-software/qm) earned its stars fast — "a multiplayer agent
-harness for work, in Slack and on the web" is a great pitch, and Y Combinator shipping
-open-source agent tooling is good for everyone. We get asked about it a lot, so here's the
-honest comparison.
+[qm](https://github.com/yc-software/qm) keeps picking up stars, and people keep asking how it compares. We
+wrote the first version of this in August. Both projects have shipped a lot since, so this is the updated
+comparison, checked against qm's README and our own 0.5.2 release on 10 September 2026.
-## The same problem, from opposite ends
+## What problem do both tools solve?
-Both tools exist because one CLI agent in one terminal doesn't scale: you want several
-agents working in parallel, safely, with scheduled background work and a way to reach them
-from where you already are.
+One agent in one terminal does not scale to real work. Both tools let several agents work in parallel, safely,
+on schedules, and reachable from where you already are. They just start from opposite ends.
-**qm answers it org-first.** A headless Node core (installed via `npm exec qm init`) that
-your team reaches through Slack and a web UI. Each "employee" gets an isolated workspace;
-skills are shared with scope-based permissions and can be promoted org-wide; admins set the
-security posture — **Strict** (approval-gated), **Auto** (default screening), or
-**Dangerous** — for everyone at once. Background work runs as **crons and watches**.
+**qm starts from the company.** You create a deployment repository that depends on `@yc-software/qm`, and the
+qm CLI validates and deploys it. A headless core backed by Postgres runs the sessions, and people reach it
+through Slack and a web app. Each person and each room gets its own scoped memory, files, keychain view,
+permissions, crons, web apps and durable sandbox. Admins choose which harnesses and models are available, and
+Pi, OpenCode, Codex and Claude Code can all drive the core.
-**Munder Difflin answers it desk-first.** A desktop app on your machine where every agent
-is a **real CLI process in a real terminal** — Claude Code, Codex, Antigravity, Copilot,
-Grok, Kimi, and more, riding the subscriptions you already pay for. The floor is visual
-(yes, it looks like The Office), a GOD orchestrator routes work, agents share long-term
-memory, and background work runs as **scheduled missions** with Slack and webhook triggers.
+**Munder Difflin starts from your desk.** It is a desktop app for macOS, Windows and Linux. Every agent is a
+real CLI process in a real terminal on your machine, running on the subscriptions you already pay for, and
+twelve CLIs are supported. The office floor shows who is doing what. Michael, your clone, turns requests into
+tasks and routes the work. Agents share long term memory and message each other through mailboxes.
-If you map the concepts across, most of qm's vocabulary has a Munder Difflin counterpart:
+## How do the concepts map across?
| Job to be done | qm | Munder Difflin |
| --- | --- | --- |
-| Safety levels | Strict / Auto / Dangerous postures | Auto-mode toggle, per-agent pause/halt, tool gating, circuit breaker, confirm-word voice safety |
-| Background work | Crons and watches | Scheduled missions + Slack/webhook triggers |
-| Reusable behaviors | Shared skills with scoped permissions | Agent Gallery roles + shareable agent recipes |
-| Reach it from anywhere | Slack + web UI | Slack + Remote Control from your phone |
-| Isolation | Isolated workspaces per employee | Isolated git worktrees per agent |
+| Safety | One org posture: Strict, Auto or Dangerous | Ask first or autonomous agents, per agent token budgets, a circuit breaker, an ASK ME board |
+| Background work | Crons, watches, inbound webhooks | Scheduled triggers on an interval or on chosen weekdays, webhooks, Slack |
+| Reusable behaviour | Skills owned by a scope, shared by grant, skill packs from git | A skills catalog, plus ready made hires from the Agent Gallery |
+| Isolation | A durable sandbox per person and room | A git worktree per agent when isolation is on |
+| Where you use it | Slack and a web app | A desktop app, plus Slack |
+| Multiplayer | Built in: one deployment for the company | Teams plan: clones message each other, sealed per device |
+| Where it runs | Your company's infrastructure | Your own machine |
| License | MIT | MIT |
{% img "note-1" %}
-## What each has that the other doesn't
-
-**qm's edge is multiplayer.** One shared harness, unified identity, org-level admin,
-skills promotion across scopes. If your team wants a *single* fleet everyone shares from
-Slack, that's qm's home turf, and Munder Difflin doesn't try to be that.
-
-**Munder Difflin's edge is everything you can see and touch:**
-
-- **Real terminals.** Every agent is a watchable CLI process — when something goes
- sideways, you read the actual session, not a log abstraction.
-- **Voice orchestration.** Flip on talk mode and Michael greets you already knowing every
- agent's status, steers work, changes settings from an allowlist, and waits for a spoken
- confirm word before anything destructive.
-- **A built-in IDE.** Monaco (the VS Code engine) one click over the floor: side-by-side
- diffs of agent changes, clickable commit history, branch compare, guarded checkout, live
- markdown previews.
-- **Local-first by construction.** No shared server, no code upload — agents, state, and
- memory live on your machine.
-- **Auto-update.** The app ships its own updates from GitHub releases; you just click
- "restart to update."
-- **A longer release train.** Public releases since v0.2.0, now at v0.3.5, with the
- reliability work battle-tested by a growing contributor community.
+## What does qm do that Munder Difflin does not?
+
+Company wide sharing, run by an admin. That is qm's home turf. One deployment serves everyone, the admin
+controls which harnesses and models people can use, the whole company inherits one security posture, and
+people can build internal web apps and publish them to each other. A sandbox per scope also means the tools
+someone installs stay installed. If your company wants a single agent service living in Slack, qm was built
+for exactly that.
+
+## What does Munder Difflin do differently?
+
+It puts everything on your own machine, where you can watch and touch it:
+
+- **Real terminals.** Every agent is a CLI process you can open and type into. When something goes sideways,
+ you read the actual session.
+- **Twelve CLIs on your own subscriptions.** Claude Code, Codex, Gemini CLI, Copilot, Cursor and seven more,
+ mixed on one floor, within the usage limits you already pay for.
+- **Your clone in charge.** Michael takes your request, hires workers and routes messages, and you can talk
+ to him by voice.
+- **An IDE over the floor.** Browse and edit an agent's workspace and see its uncommitted changes as a diff.
+- **Nothing to deploy.** Download it, answer a short setup, give Michael a job. No server, no database.
+- **Local first.** Code, keys and personal context stay on your machine. With Teams, the only thing that
+ travels is clone to clone messages, sealed so the relay in the middle cannot read them.
{% img "note-2" %}
-## Which should you use?
-
-- **Your whole team wants one shared fleet with org-level control → qm.** That's what
- multiplayer-first buys you.
-- **You want your own agent office, on your own machine, with everything visible → Munder
- Difflin.** Simpler to adopt (download, open, hire), nothing leaves your computer, and
- the interfaces — floor, voice, IDE — are things a headless harness structurally can't
- offer.
-- **Honestly? Some teams will run both.** qm as the shared org layer, Munder Difflin as
- the personal floor where you watch and steer your own agents. They don't compete for
- the same seat.
-
-Facts about qm above come from its public README as of August 2026 — check
-[the repo](https://github.com/yc-software/qm) for the current state, because both projects
-move fast.
-
-**[Try Munder Difflin free](https://munderdiffl.in/)** — macOS, Windows, Linux, MIT-licensed.
-And if this comparison helped, a [GitHub star](https://github.com/chaitanyagiri/munder-difflin)
-helps more people find it.
+## Which one should you use?
+
+- **Your company wants one shared agent service in Slack, with admin control:** qm.
+- **You want your own office of agents on your machine, visible and vendor neutral:** Munder Difflin. Add
+ Teams when you want your clone working with your teammates' clones.
+- **Some companies will run both.** qm as the company layer in Slack, and Munder Difflin as the personal floor
+ where each person watches and steers their own agents. They do not compete for the same seat.
+
+Facts about qm above come from its public README as of 10 September 2026. Check
+[the repo](https://github.com/yc-software/qm) for the current state, because both projects ship often.
+
+**[Download Munder Difflin free](https://munderdiffl.in/)** for macOS, Windows or Linux. MIT licensed.
diff --git a/blog/src/posts/run-munder-difflin-on-a-mac-mini.md b/blog/src/posts/run-munder-difflin-on-a-mac-mini.md
index 22ac2364..96861b25 100644
--- a/blog/src/posts/run-munder-difflin-on-a-mac-mini.md
+++ b/blog/src/posts/run-munder-difflin-on-a-mac-mini.md
@@ -1,232 +1,173 @@
---
title: "Run Munder Difflin Locally on a Mac Mini"
-description: "A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint."
+description: "Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini."
date: 2026-06-22
+updated: 2026-09-10
category: guides
categoryLabel: Guides
type: Technical
draft: false
primaryKeyword: "run munder difflin on a mac mini"
-secondaryKeywords: ["local llm mac mini", "ollama mac mini", "apple silicon unified memory llm", "offline ai agents mac", "lm studio mac mini agents"]
+secondaryKeywords: ["local llm mac mini", "ollama mac mini", "apple silicon unified memory llm", "offline ai agents mac", "lm studio mac mini agents", "m6 mac mini local llm"]
tags: ["Guides", "Local-First", "Mac Mini", "Ollama", "LM Studio", "BYOK"]
author:
name: Chaitanya Giri
initials: CG
faq:
- - q: "Can a Mac Mini run a Munder Difflin hive offline?"
- a: "Yes. An Apple Silicon Mac Mini runs the model locally through Ollama or LM Studio, and Munder Difflin points OpenCode and Crush at that local OpenAI-compatible endpoint (in v0.3.1 pi reaches models through a third-party provider key rather than a local server). The hive's control plane (router, scheduler, mailboxes, audit log) is already local-first, so once the model is local there is no cloud dependency at all."
- - q: "How much RAM does the Mac Mini need for local AI agents?"
- a: "It depends on the model size, not the agent count — every worker shares one local model server. As a rule of thumb on Apple Silicon's unified memory: 16GB comfortably runs 7–8B models, 24GB runs up to ~14B, 32GB up to ~32B (tight), and 48–64GB (M4 Pro) runs a 70B-class model quantized to 4-bit. See the RAM-tier table below."
- - q: "Ollama or LM Studio — which should I use on a Mac Mini?"
- a: "Either works; both expose an OpenAI-compatible server that Munder Difflin's engines can target. Ollama is a lightweight CLI/daemon (great for headless, always-on hives); LM Studio is a GUI with a model browser and a one-click local server. You can even run both and point different engines at different ports."
- - q: "Which open-source models run best on each engine?"
- a: "All three engines — OpenCode, Crush, and pi — select models in `provider/model` form. For the fully-local path, OpenCode and Crush point at a local OpenAI-compatible endpoint, so the same locally-served model works across both; in v0.3.1 pi runs through a third-party provider key instead. For the exact, version-verified model slugs and per-RAM-tier picks, see the open-source model catalog linked below — that's the single source of truth we keep current."
+ - q: "Can a Mac mini run a Munder Difflin office offline?"
+ a: "Yes. The Mac mini serves an open model through Ollama or LM Studio, and OpenCode, Crush, Qwen and Pi agents point at that local server. Munder Difflin's routing, mailboxes, memory and schedules already run on your machine, so once the model is local, the agents' work loop does not need the internet."
+ - q: "How much memory does a Mac mini need for local AI agents?"
+ a: "It depends on the size of the model, not the number of agents, because every agent shares one model server. As a rule of thumb for 4 bit models: 16 GB runs about 8B comfortably, 24 GB up to about 14B, 32 GB up to about 32B, and 64 GB runs a 70B class model."
+ - q: "Which Mac mini should I buy for local models?"
+ a: "Buy for the biggest model you want to run, because the memory cannot be upgraded later. As of September 2026 the M6 Mac mini goes up to 32 GB and the M5 Pro model goes up to 64 GB. For 30B class models, 32 GB is the sweet spot. For 70B class models, get the M5 Pro with 64 GB."
+ - q: "Ollama or LM Studio on a Mac mini?"
+ a: "Either works. Both serve an OpenAI compatible endpoint the engines can use. Ollama is a light background service that suits an always on box, and LM Studio is an app with a model browser and a one click local server. You can run both on different ports."
+ - q: "How does Pi use a local model?"
+ a: "Through Pi's own config. Add the local server to ~/.pi/agent/models.json, and Munder Difflin copies that file into every Pi agent it starts."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p>A <strong>Mac Mini</strong> with Apple
-Silicon can run an entire <strong>Munder Difflin</strong> hive <em>offline</em>. Serve an open-source
-model locally with <strong>Ollama</strong> or <strong>LM Studio</strong>, size the model to your
-<strong>unified memory</strong> (16GB → ~7–8B, 24GB → ~14B, 32GB → ~32B, 48–64GB → 70B-class at 4-bit),
-then point <a href="/blog/why-cli-agents-are-powerful/">OpenCode and Crush</a> at the local endpoint via
-<strong>Settings → AI Engines</strong> (pi uses a third-party provider key in v0.3.1). The hive's control plane is
-<a href="/blog/local-first-ai-agent-orchestration/">already local-first</a>, so nothing leaves the box.</p></div>
-
-It's one thing to run a single local model in a chat window. It's another to put a whole *team* of
-agents to work on your own hardware, with no API bill and no data leaving the room. That's the promise
-of running Munder Difflin on a Mac Mini: the [orchestration is already
-local-first](/blog/local-first-ai-agent-orchestration/) — the message router, scheduler, mailboxes, and
-git audit log are all processes on your machine — so the only remaining cloud dependency is the model
-call itself. Move that local, and the hive is fully offline.
-
-This guide walks the whole setup: picking a Mac Mini tier, understanding how Apple Silicon's unified
-memory bounds what you can run, installing a local model server, and wiring each CLI engine to it.
-
-## Why the Mac Mini is a good hive box
-
-The Mac Mini is the cheapest way into Apple Silicon's **unified memory** architecture, and unified
-memory is exactly what local LLMs want. On a Mac, the CPU, GPU, and Neural Engine all share one pool of
-high-bandwidth RAM — so the model's weights live in the same memory the GPU computes against, with no
-copy across a PCIe bus. A 32GB Mac Mini can hold a model that would need a 32GB *discrete* GPU on a PC,
-at a fraction of the price and power draw. And because the Mini is small, quiet, and sips power, it's a
-natural always-on box for a hive you leave running for hours.
-
-One catch worth knowing up front: **Apple Silicon RAM is soldered to the chip and cannot be upgraded
-after purchase.** Buy the memory you'll want for the largest model you intend to run — it's the one
-spec you can't change later.
-
-## Step 1 — Pick a Mac Mini tier for the model you want to run
-
-Model count doesn't drive your RAM — *model size* does. Every worker in the hive shares one local model
-server, so a ten-agent hive and a two-agent hive on the same model need roughly the same memory. What
-you're really choosing is **how large a model the box can hold.**
-
-A 4-bit-quantized model needs roughly **0.6 GB of memory per billion parameters** for its weights, plus
-headroom for the KV cache (grows with context length), the model server, and macOS itself. macOS also
-caps how much unified memory the GPU may pin — by default around 65–75% — though you can raise that
-ceiling on a dedicated box with `sudo sysctl iogpu.wired_limit_mb=<MB>` when you need to fit a large
-model. A practical rule: **plan for the model to use about two-thirds of your unified memory** and leave
-the rest for context and the OS.
-
-Current Mac Mini lineup and what each tier comfortably runs (4-bit quant):
-
-| Mac Mini | Unified memory | Memory bandwidth | Comfortable model size (Q4) |
-|---|---|---|---|
-| M4 (base) | 16GB | 120 GB/s | ~7–8B (e.g. an 8B-class coder) |
-| M4 | 24GB | 120 GB/s | up to ~14B |
-| M4 | 32GB | 120 GB/s | up to ~32B (tight; short context) |
-| M4 Pro | 48GB | 273 GB/s | ~32B with room, or a 70B at heavy quant |
-| M4 Pro | 64GB | 273 GB/s | 70B-class at 4-bit |
-
-*(Newer Mac Mini chips raise bandwidth and the memory ceiling, but the same "model ≈ two-thirds of
-unified memory" sizing rule carries over — pick the tier by the model size you want.)*
-
-Here's a concrete pick for each tier. Pull these with Ollama (LM Studio carries the same models as
-GGUF/MLX): `ollama pull <tag>`, then reference the model as `local/<tag>` in OpenCode or `ollama/<tag>` in
-Crush (more on that in Step 3).
-
-| Mac Mini RAM | Recommended pick (Ollama tag) | Weights (Q4) | Good for |
-|---|---|---|---|
-| 16GB | `gpt-oss:20b` (tight) — or lighter `qwen3:8b` / `deepseek-r1:8b` | ~14GB / ~5GB | Smallest genuinely capable default; the 8B picks leave more context headroom |
-| 24GB | `qwen3:14b` · `mistral-small:24b` | 9–14GB | Roomy generalist with context to spare |
-| 32GB | `qwen3:30b-a3b` (MoE) · `qwen3-coder:30b` · `deepseek-r1:32b` | ~19–20GB | The sweet spot — fast MoE generalist, coding, or reasoning |
-| 48GB (M4 Pro) | `glm-4.7-flash` (q8) · `mixtral:8x7b` | 26–32GB | Bigger context, plus the only Mac-viable GLM |
-| 64GB (M4 Pro) | `llama3.3:70b` · `deepseek-r1:70b` | ~43GB | A 70B-class generalist or reasoner at 4-bit |
-| 96GB+ (Studio-class) | `gpt-oss:120b` · `llama4:scout` | 65–67GB | Top local models; need ~80GB resident |
-
-These picks are drawn from the model families our open-source model catalog tracks — the single source of
-truth we keep version-current.
-The companion guide, <a href="/blog/run-munder-difflin-on-open-models/">running Munder Difflin on open
-models</a>, pairs each tier with the same picks and adds the third-party-provider route.
-
-<div class="callout"><span class="ic">⚠️</span><p><strong>What a Mac Mini <em>can't</em> run
-locally.</strong> The frontier open-weight flagships — <strong>Kimi K2.x</strong> (~1&nbsp;trillion
-parameters) and <strong>GLM-5.2</strong> (744B) — are server-class: their 4-bit weights run to hundreds of
-gigabytes, far past any Mac Mini. To use those, skip the local server and point an engine at a third-party
-open-source-model provider (OpenRouter, Groq, DeepInfra, and friends) with an API key — covered in the
-<a href="/blog/run-munder-difflin-on-open-models/">open-models guide</a>. For a genuinely local GLM,
-<code>glm-4.7-flash</code> (~30B) is the Mac-friendly member of the family.</p></div>
+<div class="callout tldr"><span class="ic">TL;DR</span><p>An Apple silicon <strong>Mac mini</strong> can run
+a whole <strong>Munder Difflin</strong> office offline. Serve an open model with <strong>Ollama</strong> or
+<strong>LM Studio</strong>, size it to your <strong>unified memory</strong> (16 GB about 8B, 24 GB about 14B, 32 GB
+about 32B, 64 GB a 70B class model at 4 bit), then point <strong>OpenCode</strong>, <strong>Crush</strong> and
+<strong>Qwen</strong> at it in <strong>Settings → AI Engines</strong> and add it to <strong>Pi</strong>'s own models file.
+The office's routing, mailboxes and schedules are <a href="/blog/local-first-ai-agent-orchestration/">already local
+first</a>.</p></div>
+
+Running one local model in a chat window is a hobby. Putting a whole team of agents to work on your own hardware, with no
+API bill and nothing leaving the room, is a different project. Munder Difflin's plumbing already runs on your machine: message
+routing, mailboxes, memory, schedules and git history. The one piece that usually calls out to the internet is the model. Move
+that onto the Mac mini and the agents' whole loop stays in the room.
+
+This guide covers picking a Mac mini, sizing a model to its memory, installing a model server and wiring each engine, checked
+against Munder Difflin 0.5.2 and Apple's Mac mini lineup on 10 September 2026.
+
+## Why is a Mac mini a good box for local agents?
+
+Unified memory. On Apple silicon the CPU and GPU share one pool of fast memory, so the model's weights sit right where the GPU
+works on them, with no copying across a bus. A 32 GB Mac mini holds a model that would need a 32 GB graphics card on a PC, for
+a fraction of the power. It is also small and quiet, which makes it a natural always on box.
+
+One catch up front: **the memory is part of the chip and cannot be upgraded later.** Buy for the largest model you plan to run.
+
+## How much memory do you need?
+
+Enough for the model, not for the agents. Every agent shares one model server, so ten agents and two agents on the same model
+need about the same memory.
+
+A 4 bit model needs roughly **0.6 GB per billion parameters**, plus room for context, the model server and macOS. macOS also
+caps how much memory the GPU can hold by default, somewhere around two thirds to three quarters of it. On a dedicated box you can
+raise that ceiling with `sudo sysctl iogpu.wired_limit_mb=<MB>`. A practical rule: **plan for the model to use about two thirds of
+your memory.**
+
+Here is the Mac mini range as of September 2026, and what each memory size runs comfortably at 4 bit:
+
+| Mac mini | Memory | Comfortable model size |
+|---|---|---|
+| M6 | 16 GB | About 8B |
+| M6 | 24 GB | Up to about 14B |
+| M6 | 32 GB | Up to about 32B, with shorter context |
+| M5 Pro | 24 GB | Up to about 14B, generated faster |
+| M5 Pro | 48 GB | 32B with room, or 70B at heavy quantization |
+| M5 Pro | 64 GB | 70B class at 4 bit |
+
+The M5 Pro also has much faster memory, 307 GB/s against up to 170 GB/s on the M6, and that shows up as tokens per second.
+
+A concrete pick for each memory size, pulled with `ollama pull <tag>`:
+
+| Memory | Pick (Ollama tag) | Good for |
+|---|---|---|
+| 16 GB | `gpt-oss:20b` (tight), or `qwen3:8b` or `deepseek-r1:8b` for more context room | The smallest capable default |
+| 24 GB | `qwen3:14b` or `mistral-small:24b` | A roomy generalist |
+| 32 GB | `qwen3:30b-a3b`, `qwen3-coder:30b` or `deepseek-r1:32b` | The sweet spot: generalist, coding or reasoning |
+| 48 GB | `glm-4.7-flash` at 8 bit, or `mixtral:8x7b` | Bigger context |
+| 64 GB | `llama3.3:70b` or `deepseek-r1:70b` | A 70B class generalist or reasoner |
+
+<div class="callout note"><span class="ic">Heads up</span><p><strong>What a Mac mini cannot run.</strong> The frontier open
+flagships, like Kimi K2.6 and Qwen3 235B, are server class and far beyond 64 GB. Use a provider for those, which the
+<a href="/blog/run-munder-difflin-on-open-models/">open models guide</a> covers. And <code>gpt-oss:120b</code> needs about 96 GB,
+which means a Mac Studio rather than a mini.</p></div>
{% img "note-1" %}
-## Step 2 — Install a local model server
+## Step 1: Install a model server
-You have two solid options. Both expose an **OpenAI-compatible HTTP endpoint**, which is exactly what
-Munder Difflin's engines target.
+Ollama or LM Studio. Both serve an OpenAI compatible endpoint, which is what the engines talk to.
-### Option A — Ollama (lightweight, headless-friendly)
+### Ollama: light, good for an always on box
```bash
-# Install (Homebrew) and start the daemon
brew install ollama
-ollama serve # runs the API on http://localhost:11434
-
-# Pull a model (pick the tag for your RAM tier — see Step 1)
-ollama pull gpt-oss:20b # 16GB-friendly default; swap for your tier's tag
+ollama serve # serves the API on http://localhost:11434
+ollama pull gpt-oss:20b # swap in the tag for your memory size
```
-Ollama's OpenAI-compatible API is served at **`http://localhost:11434/v1`**. It runs as a background
-daemon, which makes it the natural choice for an always-on hive box.
+Its OpenAI compatible API lives at **`http://localhost:11434/v1`**. It runs as a background service, which suits a box you leave on.
+
+### LM Studio: an app with a model browser
-### Option B — LM Studio (GUI, model browser, one-click server)
+Download it from [lmstudio.ai](https://lmstudio.ai), find and download a model inside the app, then start its local server from the
+Developer section. It serves an OpenAI compatible API at **`http://localhost:1234/v1`**.
-Download LM Studio from [lmstudio.ai](https://lmstudio.ai), search and download a model in its UI, then
-start the **Local Server** (the "Developer" / server tab). LM Studio serves an OpenAI-compatible API at
-**`http://localhost:1234/v1`**.
+You can run both, Ollama on 11434 and LM Studio on 1234, and point different engines at different ports to compare models.
-<div class="callout"><span class="ic">💡</span><p>You can run <em>both</em> — Ollama on :11434 and LM
-Studio on :1234 — and point different engines at different ports if you want to compare models side by
-side across your workers.</p></div>
+## Step 2: Wire each engine to the local server
-## Step 3 — Wire each CLI engine to the local endpoint
+Four engines can use a local model on your Mac mini floor.
+
+### OpenCode
-Munder Difflin runs each agent on a pluggable [CLI engine](/blog/why-cli-agents-are-powerful/). All three engines select models in **`provider/model`** form, but for the fully-local path it's **two** —
-OpenCode and Crush — that wire a local OpenAI-compatible endpoint in v0.3.1, so the same locally-served
-model works across both. (pi reaches models through a third-party provider key in this release — see its
-note below.) The fastest path
-is **Settings → AI Engines**: set the per-engine base URL to your local server and pick the model; the
-harness writes the right per-agent config for you. Here's what that maps to under the hood for each
-engine.
+Open **Settings → AI Engines** and set OpenCode's local base URL to `http://localhost:11434/v1`. The app injects it as a local
+provider when the agent starts. Pick the model as `local/<tag>`, for example `local/gpt-oss:20b`.
### Crush
-Crush reads a JSON config. Define the local server as an **`openai-compat`** provider (this keeps it on
-the OpenAI wire, which is what the harness proxy expects — don't use Crush's native `ollama`/`lmstudio`
-discovery types for the hive path):
+Set Crush's local base URL in the same panel. Crush runs through a small local proxy inside the app, and your server becomes that
+proxy's upstream, so there is no Crush config to edit by hand. Pick the model as `ollama/<tag>`, for example `ollama/gpt-oss:20b`.
+
+### Qwen
+
+Set Qwen's local base URL and default model in the same panel. Qwen uses the same proxy approach as Crush.
+
+### Pi
+
+Pi reads local models from its own file, `~/.pi/agent/models.json`, and Munder Difflin copies that file into every Pi agent it
+starts. The app's base URL field for Pi stays reserved. Add Ollama like this, then pick `ollama/gpt-oss:20b`:
```json
{
"providers": {
"ollama": {
- "name": "Ollama",
- "type": "openai-compat",
- "base_url": "http://localhost:11434/v1",
- "api_key": "ollama",
- "models": [
- { "name": "gpt-oss 20B", "id": "gpt-oss:20b", "context_window": 131072, "default_max_tokens": 8192 }
- ]
+ "baseUrl": "http://localhost:11434/v1",
+ "api": "openai-completions",
+ "apiKey": "ollama",
+ "models": [{ "id": "gpt-oss:20b" }]
}
}
}
```
-Then select the model as `ollama/gpt-oss:20b` (swap in the tag for your RAM tier from Step 1). For LM
-Studio, use `base_url: "http://localhost:1234/v1"`, `api_key: "lm-studio"`, and a `lmstudio` provider key,
-then select `lmstudio/<model-id>`.
-
-### OpenCode
-
-OpenCode reaches a local model through a custom OpenAI-compatible provider (the `@ai-sdk/openai-compatible`
-adapter with `options.baseURL`). In Munder Difflin, set the OpenCode base URL in Settings → AI Engines to
-your local server and choose the model in `provider/model` form; the harness supplies the engine config.
-
-### pi
-
-A heads-up specific to **v0.3.1**: pi's local-endpoint field is **reserved.** This release wires a local
-base URL for OpenCode and Crush only — it does not yet write a local provider into pi's `models.json` — so
-the fully-offline path today runs on **OpenCode or Crush.** To put pi to work in this release, point it at
-a third-party open-source-model provider instead of a local server: set a provider key in **Settings → AI
-Engines** (e.g. an OpenRouter or Groq key) and pick an open-model slug such as
-`openrouter/openai/gpt-oss-120b` or `groq/llama-3.3-70b-versatile`. pi joins the local path once its
-local-endpoint field ships; until then, build your offline hive on OpenCode and Crush.
-
-<div class="callout"><span class="ic">📋</span><p><strong>The same local model, three prefixes.</strong>
-Once a model is served locally, each engine names it slightly differently: OpenCode as
-<code>local/&lt;tag&gt;</code>, Crush as <code>ollama/&lt;tag&gt;</code>. For example, gpt-oss 20B is
-<code>local/gpt-oss:20b</code> in OpenCode and <code>ollama/gpt-oss:20b</code> in Crush. (In v0.3.1, pi
-points at a third-party <code>provider/model</code> slug instead, per the note above.) The exact slugs for
-every tier live in the <a href="/blog/run-munder-difflin-on-open-models/">open-source model catalog</a>,
-kept in one place so they don't drift.</p></div>
+For LM Studio, use `http://localhost:1234/v1` and the model id LM Studio shows you.
{% img "note-2" %}
-## Step 4 — Run the hive offline
+## Step 3: Run the office
-With a model served locally and the engines pointed at it, the rest of Munder Difflin is already
-local-first. Spin up your workers from the [Agent Gallery or Add Agent](/blog/how-to-install-and-use-munder-difflin/),
-give the hive a goal, and let it run. The [message router](/blog/atomic-file-mailboxes-for-agents/) moves
-work between agents through file mailboxes, the scheduler fires tasks on local timers, and git records
-the audit trail — none of it touches a network. Pull the Ethernet cable and the hive keeps working.
+With a model served and the engines pointed at it, hire workers from the
+[Agent Gallery or Add agent](/blog/how-to-hire-from-the-agent-gallery/), give Michael a goal, and let it run. Messages move between
+agents through [file mailboxes](/blog/atomic-file-mailboxes-for-agents/), schedules fire on local timers, and git keeps the history.
-A few operational notes for an offline Mac Mini hive:
+A few notes for an always on Mac mini:
-- **One model, many workers.** Every agent shares the local server, so concurrency is bounded by the
- Mac Mini's compute, not by per-agent memory. Watch tokens/sec under load and keep contexts trim on the
- 16/24GB tiers.
-- **Keep it awake.** A headless always-on hive wants the Mini to not sleep — set Energy settings to
- prevent sleep (or `caffeinate` the session) so the router and scheduler keep ticking.
-- **Quantization is your lever.** If a model won't fit, drop to a heavier quant before dropping a tier —
- 4-bit is the usual sweet spot for coder models on these boxes.
-- **Fully offline means OpenCode or Crush.** In v0.3.1 those are the two engines that wire a local base
- URL, so a hive built on them runs with the cable pulled. pi currently reaches models through a
- third-party provider key (its local field is reserved), so pi workers need a network — mix engines
- accordingly if total isolation is the goal.
+- **One model, many workers.** Every agent shares the model server, so the limit is the chip's speed, not memory per agent. Watch
+ tokens per second under load, and keep contexts trim on 16 and 24 GB.
+- **Keep it awake.** Set the Mac mini not to sleep, or run `caffeinate`, so schedules keep firing. Munder Difflin also holds the
+ machine awake while at least one agent is running.
+- **Quantization is your lever.** If a model does not fit, try a smaller quantization before you try a smaller model.
+- **Mix engines freely.** OpenCode, Crush, Qwen and Pi can all share the same local server, so choose per agent.
## Where to go next
-- The companion guide, [run Munder Difflin on open models](/blog/run-munder-difflin-on-open-models/),
- covers running fully local *or* through a third-party open-source provider, and pairs each RAM tier
- with a concrete model pick.
-- For the *why* behind keeping it all on your machine, see [why local-first matters for AI
- agents](/blog/why-local-first-matters-for-ai-agents/).
-- New to the app? Start with [how to install and use Munder
- Difflin](/blog/how-to-install-and-use-munder-difflin/).
+- [Run Munder Difflin on open models](/blog/run-munder-difflin-on-open-models/) adds the provider route and the full list of quick picks.
+- [Why local first matters for AI agents](/blog/why-local-first-matters-for-ai-agents/), for the reasoning behind all of this.
+- New to the app? Start with [how to install and use Munder Difflin](/blog/how-to-install-and-use-munder-difflin/).
diff --git a/blog/src/posts/run-munder-difflin-on-open-models.md b/blog/src/posts/run-munder-difflin-on-open-models.md
index 8aa323ad..a7a251f9 100644
--- a/blog/src/posts/run-munder-difflin-on-open-models.md
+++ b/blog/src/posts/run-munder-difflin-on-open-models.md
@@ -1,152 +1,180 @@
---
-title: "Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider"
-description: "Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker's own CLI. Here's the wiring, current as of v0.4.4."
+title: "Run Munder Difflin on Open Source Models: Fully Local or Through a Provider"
+description: "Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2."
date: 2026-06-22
-updated: 2026-08-20
+updated: 2026-09-10
category: guides
categoryLabel: Guides
type: Technical
primaryKeyword: "run ai agents on open source models"
-secondaryKeywords: ["local llm coding agent", "ollama coding agent", "openrouter coding agent", "gpt-oss", "byok open models", "opencode crush pi", "qwen cli agent"]
+secondaryKeywords: ["local llm coding agent", "ollama coding agent", "openrouter coding agent", "gpt-oss", "byok open models", "opencode crush pi", "pi models.json ollama"]
tags: ["Guides", "Local-First", "Open Source", "CLI Agents", "Tutorial"]
author:
name: Chaitanya Giri
initials: CG
faq:
- - q: "Can Munder Difflin run entirely on open-source models?"
- a: "Yes. The OpenCode, Crush, and pi engines all support bring-your-own-key (BYOK) and local models, and the Qwen and Kimi CLIs are first-class engines whose flagship models are open-weight. You can run every agent — workers and Michael himself — on open models like gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, or Kimi: fully local on your own hardware, or through a third-party OSS provider with your own API key."
- - q: "What's the difference between running local and using a third-party provider?"
- a: "Local (Ollama, LM Studio, vLLM) runs the weights on your own machine — fully private, no per-token bill, but bounded by your RAM and GPU. A third-party OSS provider (OpenRouter, Groq, Together, Fireworks, DeepInfra) hosts the same open weights on their hardware and you pay per token with your own key — no local hardware limit, so you can reach the 100B–1T-parameter frontier models a laptop can't hold."
- - q: "Which open model should I pick for the orchestrator seat?"
- a: "Michael does the reasoning and long-context coordination, so give him a strong model: locally, gpt-oss-120b or Llama 3.3 70B on a 64–96 GB machine; via a provider, DeepSeek-V4-Flash, GLM-4.6, or Kimi-K2.6 on OpenRouter. Small local models (8B and under) are fine for routine workers but underpowered for orchestration."
- - q: "Do I need a different model id for each engine?"
- a: "No — the upstream model id is the same. All three BYOK engines use a provider/model slug; only the provider prefix and how the key or base-URL is wired differ. For local models, OpenCode uses local/<id> while Crush uses ollama/<id>."
+ - q: "Can Munder Difflin run entirely on open source models?"
+ a: "Yes. OpenCode, Crush, Qwen and Pi can all run open weight models, either on your own machine or through a provider with your own API key. You can put open models in every seat on the floor, including Michael's."
+ - q: "What is the difference between running local and using a provider?"
+ a: "Local means Ollama, LM Studio or vLLM runs the weights on your machine: private, no per token bill, and limited by your memory. A provider such as OpenRouter or Groq hosts the same open weights on its hardware and bills your own key per token, so you can reach models far too big for a laptop."
+ - q: "Which open model should I give the orchestrator?"
+ a: "A strong one, because Michael does the reasoning and the long context coordination. Locally that means gpt-oss 120B or Llama 3.3 70B on a machine with 64 to 96 GB of memory. Through a provider, DeepSeek V4 Flash or Kimi K2.6 on OpenRouter. Models of 8B and under make fine workers and thin orchestrators."
+ - q: "Does each engine need a different model name?"
+ a: "The model id stays the same and only the prefix changes. A local model is local/<tag> on OpenCode and ollama/<tag> on Crush and Pi, and a provider model carries the provider first, like openrouter/openai/gpt-oss-120b."
+ - q: "How do I run Pi on a local model?"
+ a: "Add your local server to Pi's own config file, ~/.pi/agent/models.json. Munder Difflin copies that file into every Pi agent it starts, so the models you define there are available to your Pi agents. The Pi base URL field in the app stays reserved."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p><strong>Munder Difflin runs entirely on open models.</strong> Three routes: <strong>fully local</strong> (Ollama / LM Studio / vLLM on your own machine — private, no per-token bill, bounded by RAM), a <strong>third-party OSS provider</strong> (OpenRouter, Groq, Together, Fireworks, DeepInfra — their hardware, your key, reaching frontier 100B–1T models a laptop can't hold), or the <strong>model-maker's own CLI</strong> — the Qwen and Kimi engines are first-class hires whose flagship models are open-weight. The BYOK engines (<strong>OpenCode</strong>, <strong>Crush</strong>, <strong>pi</strong>) all use a <code>provider/model</code> slug; keys and local base-URLs live in <strong>Settings → AI Engines</strong>, and <strong>Settings → Prerequisites</strong> tells you which engine binaries the app can actually see.</p></div>
+<div class="callout tldr"><span class="ic">TL;DR</span><p><strong>Munder Difflin can run on open models
+end to end.</strong> Two routes: <strong>fully local</strong> with Ollama, LM Studio or vLLM (private, no per
+token bill, limited by your memory), or a <strong>provider</strong> such as OpenRouter or Groq (their hardware,
+your key, much bigger models). Four engines do the wiring: <strong>OpenCode</strong>, <strong>Crush</strong> and
+<strong>Qwen</strong> take a local base URL in <strong>Settings → AI Engines</strong>, and <strong>Pi</strong> reads
+your own <code>~/.pi/agent/models.json</code>. Keys go in the same panel and are stored write only.</p></div>
-Munder Difflin started as a harness for the closed frontier CLIs — Claude Code, Codex, Antigravity. Useful, but it tied your agent floor to a handful of vendors and their pricing. That's long since broken open: of the **ten engines** the app now ships — Claude Code, Antigravity, Codex, Grok, Kimi, Qwen, OpenCode, Crush, pi, and GitHub Copilot CLI — three ([OpenCode](https://opencode.ai), [Crush](https://charm.land), [pi](https://pi.dev)) were built from the start to point at *any* model, and two more (Qwen, Kimi) are the model-makers' own CLIs for families whose weights are public.
+Munder Difflin started out wrapping the closed frontier CLIs. Today it supports twelve, and several of them will
+point at any model you like. That means a whole office of agents can run on models whose weights anyone can download.
-That means you can run an entire office of agents on models anyone can download. There are two honest ways to do it, and they trade off differently. This guide walks both, then shows the exact wiring for each engine — current as of **v0.4.4**. (For the why-bother, see [why local-first matters for AI agents](/blog/why-local-first-matters-for-ai-agents/) and [why CLI agents are so powerful](/blog/why-cli-agents-are-powerful/).)
+There are two honest ways to do it, and they trade off differently. This guide walks through both, then gives the exact
+wiring for each engine, checked against Munder Difflin 0.5.2 on 10 September 2026. (For the why, see
+[why local first matters for AI agents](/blog/why-local-first-matters-for-ai-agents/).)
-## Two routes: your hardware, or someone else's
+## Should you run open models locally or through a provider?
-"Open source models" is one phrase covering two very different setups. Pick by what you're optimizing for.
+Local if privacy and a fixed cost matter and your machine can hold the model. A provider if you want the biggest models
+or have no spare hardware.
-| | Fully local | Third-party OSS provider |
+| | Fully local | Through a provider |
|---|---|---|
-| **Runs on** | Your machine (Ollama, LM Studio, vLLM) | Their GPUs (OpenRouter, Groq, Together, Fireworks, DeepInfra, Novita) |
-| **Cost** | Electricity. No per-token bill. | Per-token, billed to your own key. |
-| **Privacy** | Total — code never leaves the box. | Prompts transit a third party. |
-| **Ceiling** | Bounded by RAM/VRAM (≈7B–70B realistic on a Mac). | The whole frontier — 235B, 480B, even 1T-parameter models. |
-| **Setup** | Pull a model + point the engine at `localhost`. | Paste one API key. |
-| **Best for** | Private work, 24/7 unattended, fixed cost. | Frontier quality, zero local hardware, bursty use. |
+| **Runs on** | Your machine (Ollama, LM Studio, vLLM) | Their GPUs (OpenRouter, Groq and others) |
+| **Cost** | Electricity. No per token bill. | Per token, billed to your own key. |
+| **Privacy** | Prompts never leave the machine. | Prompts go to the provider. |
+| **Ceiling** | Your memory: roughly 8B to 70B on a Mac, 120B on a very big one. | Open models with hundreds of billions of parameters. |
+| **Setup** | Pull a model and point the engine at localhost. | Paste one key. |
+| **Best for** | Private work, always on floors, fixed cost. | Top quality, bursty use, no local hardware. |
-You don't have to choose globally — Munder Difflin sets the engine and model *per agent*. A common pattern: a strong provider-hosted model in the orchestrator seat, and cheap local workers for the routine majority. That's exactly the [capability-routing](/blog/do-more-with-less-model-routing/) idea, now with open weights on both ends.
+You do not have to choose once for the whole floor. Engine and model are set per agent, so a strong provider model can sit
+in Michael's seat while cheap local workers handle the routine majority. That is
+[capability routing](/blog/do-more-with-less-model-routing/) with open weights at both ends.
-And route three, for the least wiring of all: **hire the model-maker's own CLI.** The Qwen and Kimi engines are first-class in the Add-Agent dialog — their vendor CLIs, their auth, no slug to compose — and Qwen3 and Kimi K2.6 are open-weight families. If all you want is "an open-model worker on my floor, now," that's one dropdown.
+## Which engines can run open models?
-## How the three BYOK engines name a model
+Four, and each one wires it a little differently.
-One thing to internalize before any wiring: **OpenCode, Crush, and pi all use the same `provider/model` slug form.** The *model* part is just the upstream's id (e.g. `openai/gpt-oss-120b`, `qwen3:30b-a3b`). The *provider* prefix resolves one of two ways:
+| Engine | Where the local server goes | What the app does with it | Local model slug |
+|---|---|---|---|
+| **OpenCode** | Base URL in Settings → AI Engines | Injects it as a local OpenAI compatible provider named `local` | `local/<tag>` |
+| **Crush** | Base URL in Settings → AI Engines | Uses it as the upstream of the local proxy Crush runs through | `ollama/<tag>` |
+| **Qwen** | Base URL and default model in Settings → AI Engines | The same proxy approach as Crush | the default model you set |
+| **Pi** | Your own `~/.pi/agent/models.json` | Copies that file into every Pi agent it starts | `ollama/<tag>` |
-- **A built-in provider** the engine already knows — supply the matching API-key env var and you're done: `openrouter`, `openai`, `anthropic`, `groq`, `deepseek`, `mistral`, and the local ones.
-- **A custom OpenAI-compatible provider** you define once (a `base_url` + key block) for any host that isn't built in — Together, Fireworks, DeepInfra, Novita, Z.ai, Moonshot. Then the slug is `<your-name>/<model-id>`.
+The usual endpoints: Ollama at `http://localhost:11434/v1`, LM Studio at `http://localhost:1234/v1`, and vLLM wherever you
+expose it, often `:8000/v1`.
-The *local* route differs slightly by engine — same id, different prefix and wiring:
+{% img "note-1", "Same model id everywhere. Only the prefix changes: local/ on OpenCode, ollama/ on Crush and Pi, and the provider name when you use a key." %}
-| Engine | How the app wires local | Local slug |
-|---|---|---|
-| **OpenCode** | Injects a custom provider named `local` (OpenAI-compatible) with your base-URL. | `local/<id>` |
-| **Crush** | Writes a provider block (`type: ollama / lmstudio / openai-compat`) into the agent's config. | `ollama/<id>` |
-| **pi** | Still **reserved** as of v0.4.4 — the harness doesn't yet write pi a `models.json`. Run open models on pi via a provider key (Path B). | — |
+## How do you run a model fully locally?
-Default endpoints are the usual ones: Ollama `http://localhost:11434/v1`, LM Studio `http://127.0.0.1:1234/v1`, vLLM whatever you exposed (often `:8000/v1`). You set these in the app — no shell exports required.
+Pull a model, tell the engine where it lives, and pick it for an agent.
-{% img "note-1", "Same model id everywhere — only the prefix changes: local/ on OpenCode, ollama/ on Crush, provider/ for BYOK." %}
+**1. Pull a model.** With [Ollama](https://ollama.com) installed, grab one sized to your memory:
-## Path A — fully local (Ollama / LM Studio / vLLM)
+```bash
+ollama pull gpt-oss:20b # about 14 GB, fits a 16 GB Mac
+ollama pull qwen3:30b-a3b # about 19 GB, a fast generalist for 32 GB
+ollama pull deepseek-r1:32b # about 20 GB, strong reasoning for 32 GB
+ollama serve # serves the OpenAI compatible API on port 11434
+```
-Three steps: pull a model, tell Munder Difflin where it lives, pick it for an agent.
+LM Studio works the same way: load a model in the app and start its local server on port 1234.
-**1. Pull a model.** With [Ollama](https://ollama.com) installed, grab one sized to your RAM:
+**2. Point the engine at it.** For OpenCode, Crush or Qwen, open **Settings → AI Engines** and set that engine's local base
+URL, for example `http://localhost:11434/v1`. Local servers need no key. For Pi, add the server to `~/.pi/agent/models.json`
+instead:
-```bash
-ollama pull gpt-oss:20b # 14 GB — runs on a 16 GB Mac, the safe default
-ollama pull qwen3:30b-a3b # 19 GB — fast MoE generalist, 32 GB
-ollama pull deepseek-r1:32b # 20 GB — strong reasoning, 32 GB
-ollama serve # exposes the OpenAI-compatible API on :11434
+```json
+{
+ "providers": {
+ "ollama": {
+ "baseUrl": "http://localhost:11434/v1",
+ "api": "openai-completions",
+ "apiKey": "ollama",
+ "models": [{ "id": "gpt-oss:20b" }]
+ }
+ }
+}
```
-(LM Studio works the same way — load the model in the app and it serves on `:1234`. vLLM and llama.cpp expose their own OpenAI-compatible endpoint.)
+Munder Difflin copies that file into each Pi agent when it starts, so start the Pi agent again after you edit it.
-**2. Point the engine at it.** Open **Settings → AI Engines**, find the engine you'll use (**OpenCode** or **Crush**), and set its **local base-URL** field to your endpoint — e.g. `http://localhost:11434/v1` for Ollama. The harness uses it to inject the right provider config when it spawns the agent. No API key needed for local.
+**3. Hire an agent on that model.** In **Add agent**, choose the engine and pick one of the open model quick picks, or type
+the slug yourself: `local/gpt-oss:20b` on OpenCode, `ollama/gpt-oss:20b` on Crush and Pi. Keep the colon in the tag. That
+agent now runs entirely on your hardware.
-**3. Hire an agent on that model.** In the **Add-Agent** modal, choose the engine, then pick the local model. The picker offers the open-model quick-picks; the slug it sends is `local/gpt-oss:20b` on OpenCode, or `ollama/gpt-oss:20b` on Crush (keep the colon in the tag). That agent now runs fully on your hardware.
+Which local model? These are the app's local quick picks, by memory:
-Which local model? Match it to your machine. These picks are from the project's open-model catalog, by RAM tier:
-
-| Model | Ollama tag | Min RAM | Good for |
+| Model | Ollama tag | Memory | Good for |
|---|---|---|---|
-| gpt-oss 20B | `gpt-oss:20b` | 16 GB | Smallest capable default |
-| Mistral Small 24B | `mistral-small:24b` | 16–32 GB | Lightweight generalist |
-| Qwen3 30B-A3B (MoE) | `qwen3:30b-a3b` | 32 GB | Fast MoE generalist |
-| Qwen3-Coder 30B | `qwen3-coder:30b` | 32 GB | Coding |
-| DeepSeek-R1 32B | `deepseek-r1:32b` | 32 GB | Reasoning |
-| GLM-4.7-Flash | `glm-4.7-flash` | 32 GB | The only Mac-viable GLM |
-| Llama 3.3 70B | `llama3.3:70b` | 64 GB | Bigger generalist |
-| gpt-oss 120B | `gpt-oss:120b` | 96 GB | Top local (Studio-class) |
+| gpt-oss 20B | `gpt-oss:20b` | 16 GB | The smallest capable default |
+| Mistral Small 24B | `mistral-small:24b` | 16 to 32 GB | A light generalist |
+| Qwen3 30B A3B | `qwen3:30b-a3b` | 32 GB | A fast generalist |
+| Qwen3 Coder 30B | `qwen3-coder:30b` | 32 GB | Coding |
+| DeepSeek R1 32B | `deepseek-r1:32b` | 32 GB | Reasoning |
+| GLM 4.7 Flash | `glm-4.7-flash` | 32 GB | The GLM that fits on a Mac |
+| Llama 3.3 70B | `llama3.3:70b` | 64 GB | A bigger generalist |
+| gpt-oss 120B | `gpt-oss:120b` | 96 GB | The top local pick |
-A note on what *won't* fit: the headline frontier open models — DeepSeek-V4, Kimi K2.6, GLM-5.2, Qwen3-235B — are server-class. The Ollama tags exist, but no consumer Mac holds them. For those, you want Path B. (Choosing a local model by RAM is the whole subject of the companion [Mac Mini setup guide](/blog/run-munder-difflin-on-a-mac-mini/).)
+The headline open flagships, like DeepSeek V4, Kimi K2.6 and Qwen3 235B, are server class. No consumer Mac holds them, so
+use a provider for those. Sizing a model to your memory is the whole subject of the
+[Mac mini guide](/blog/run-munder-difflin-on-a-mac-mini/).
-## Path B — a third-party OSS provider (BYOK)
+## How do you run open models through a provider?
-Same open weights, hosted on someone else's GPUs, billed to your own key. This is how you reach the big models, and it's a two-field setup.
+Same open weights, someone else's GPUs, your own key.
-**1. Get a key.** Sign up with a provider and copy an API key. [OpenRouter](https://openrouter.ai) is the easiest start — one key, the widest catalog. [Groq](https://groq.com) is the fastest for the models it carries. Together, Fireworks, and DeepInfra host the heavyweights.
+**1. Get a key.** [OpenRouter](https://openrouter.ai) is the easiest start, with one key for a wide catalog.
+[Groq](https://groq.com) is quick for the models it carries.
-**2. Paste it into the app.** In **Settings → AI Engines**, enter the key in the matching field. Keys are stored **write-only through the broker** — the renderer can set a key but never read it back; only the main process injects it as the right env var (`OPENROUTER_API_KEY`, `GROQ_API_KEY`, …) when an agent spawns, and only for the provider that agent actually uses. Nothing lands in plaintext config.
+**2. Paste it into Settings → AI Engines.** There are key fields for Anthropic, OpenAI, Google Gemini, OpenRouter and Groq.
+Keys are write only: the app stores a key but never shows it back, and it reaches an agent as the right environment variable
+only when that agent starts.
-**3. Pick a model.** In Add-Agent, select the engine and a provider-hosted model. The slug carries the provider prefix — e.g. `openrouter/deepseek/deepseek-v4-flash` or `groq/openai/gpt-oss-120b`. The recommended BYOK quick-picks:
+**3. Pick a model.** In Add agent, choose OpenCode, Crush or Pi and a provider hosted quick pick:
-| Model | Route | Slug | Key env |
+| Model | Route | Slug | Key |
|---|---|---|---|
-| gpt-oss 120B (fastest) | Groq | `groq/openai/gpt-oss-120b` | `GROQ_API_KEY` |
+| gpt-oss 120B | Groq | `groq/openai/gpt-oss-120b` | `GROQ_API_KEY` |
| Llama 3.3 70B | Groq | `groq/llama-3.3-70b-versatile` | `GROQ_API_KEY` |
-| DeepSeek-V4-Flash | OpenRouter | `openrouter/deepseek/deepseek-v4-flash` | `OPENROUTER_API_KEY` |
-| GLM-4.6 | OpenRouter | `openrouter/z-ai/glm-4.6` | `OPENROUTER_API_KEY` |
+| DeepSeek V4 Flash | OpenRouter | `openrouter/deepseek/deepseek-v4-flash` | `OPENROUTER_API_KEY` |
+| GLM 4.6 | OpenRouter | `openrouter/z-ai/glm-4.6` | `OPENROUTER_API_KEY` |
| Kimi K2.6 | OpenRouter | `openrouter/moonshotai/kimi-k2.6` | `OPENROUTER_API_KEY` |
-| Qwen3-Coder 480B | OpenRouter | `openrouter/qwen/qwen3-coder` | `OPENROUTER_API_KEY` |
+| Qwen3 Coder 480B | OpenRouter | `openrouter/qwen/qwen3-coder` | `OPENROUTER_API_KEY` |
+| Qwen3 235B | OpenRouter | `openrouter/qwen/qwen3-235b-a22b-2507` | `OPENROUTER_API_KEY` |
| gpt-oss 120B | OpenRouter | `openrouter/openai/gpt-oss-120b` | `OPENROUTER_API_KEY` |
-Prefer a model maker's own API? Those work too: DeepSeek (`deepseek/deepseek-v4-flash`, `DEEPSEEK_API_KEY`), Mistral (`mistral/...`, `MISTRAL_API_KEY`), Z.ai for GLM, Moonshot for Kimi. On Groq, stick to gpt-oss and `llama-3.3-70b-versatile`. The full slug-by-slug table, with citations, lives in the project's open-model catalog (the single source of truth this post and the Mac Mini guide both cite).
-
-One honest fix from v0.4.4 worth knowing: OpenCode used to preselect a BYOK slug **and silently fall back** to a different model when the key was absent — while every surface kept reporting the model it had asked for. That's gone. If a key is missing now, you find out; you never unknowingly run a model you didn't pick.
-
-{% img "note-2", "One key, the whole frontier: paste it once in Settings, and the write-only broker injects it per spawn — never into plaintext config." %}
-
-## Per-engine cheat-sheet
-
-You rarely touch these directly — the AI Engines panel writes them — but here's what each engine does under the hood, so the model field makes sense.
-
-**OpenCode** is OpenAI-SDK native and knows most providers out of the box. BYOK is just the env var; local is a custom provider named `local`. Slugs: `openrouter/openai/gpt-oss-120b`, `local/qwen3:30b-a3b`.
-
-**Crush** reads BYOK env vars for its built-in providers and uses a written config block for anything custom or local. For local Ollama it's literally:
-
-```json
-{ "providers": { "ollama": { "type": "ollama", "base_url": "http://localhost:11434/v1" } } }
-```
+{% img "note-2", "One key, the whole catalog: paste it once, and it reaches an agent only when that agent starts." %}
-then select `ollama/qwen3:30b-a3b`. For a host like Together, it's an `openai-compat` block with that provider's `base_url` and your key.
+## What should go in Michael's seat?
-**pi** ships 15+ built-in providers, so BYOK is just the provider key. Slugs look like `groq/llama-3.3-70b-versatile` or `openrouter/qwen/qwen3-coder`. Its local base-URL field remains **reserved** as of v0.4.4 — run open models on pi through a provider key rather than a local endpoint.
+A strong model. Michael does the reasoning, holds the long context and decides who does what. Locally, that is `gpt-oss:120b`
+or `llama3.3:70b` on 64 to 96 GB of memory. Through a provider, DeepSeek V4 Flash or Kimi K2.6. Models under 8B make good
+workers and thin orchestrators.
-**Qwen and Kimi** are the zero-wiring route: vendor CLIs as first-class engines. Sign in the way each CLI wants and hire away — no slugs, no base-URLs.
+## What if it does not work?
-All the BYOK engines are orchestrator-eligible, so you can put an open model in Michael's seat, not just the workers'. Give the seat a strong one — `gpt-oss:120b` or `llama3.3:70b` locally (64–96 GB), or a frontier provider model like `openrouter/deepseek/deepseek-v4-flash`. Sub-8B models are great workers but thin for orchestration.
+- **The model name is rejected.** Check the prefix: `local/` on OpenCode, `ollama/` on Crush and Pi. Keep the colon in the
+ Ollama tag.
+- **The agent cannot reach the server.** Make sure `ollama serve` or LM Studio's server is running, and that the base URL
+ ends in `/v1`.
+- **A Pi agent cannot see your model.** Check that `~/.pi/agent/models.json` is valid JSON, then start the Pi agent again so
+ the app copies the new file.
+- **An engine shows as missing.** Settings → Prerequisites lists which engine commands the app can actually find.
## The bottom line
-Open weights turn Munder Difflin from "a harness for three vendors' CLIs" into "a harness for the whole open ecosystem." Run it **fully local** when privacy and fixed cost matter and your RAM can hold the model; run it on a **third-party provider** when you want frontier quality or no local hardware at all; hire the **vendor CLI** when you want zero wiring — and mix all three across your floor, agent by agent. The setup is two fields in **Settings → AI Engines** and a pick in Add-Agent, with **Settings → Prerequisites** confirming every engine binary the app can see.
+Open weights turn Munder Difflin from a harness for a few vendors' CLIs into a harness for the whole open ecosystem. Go local
+when privacy and a fixed cost matter, use a provider when you want the biggest models, and mix both across your floor, agent
+by agent.
-That's the promise kept: a virtual office of CLI agents on your own computer, running on models whose weights anyone can read. [Download Munder Difflin](https://munderdiffl.in/#install) — free, open source, local-first — and point your favorite open model at it. (On a Mac and want the hardware-by-RAM walkthrough? Read the [Mac Mini setup guide](/blog/run-munder-difflin-on-a-mac-mini/).)
+[Download Munder Difflin](https://munderdiffl.in/), free and open source, and point your favourite open model at it. On a Mac
+and want the sizing walkthrough? Read the [Mac mini guide](/blog/run-munder-difflin-on-a-mac-mini/).
diff --git a/blog/src/posts/what-is-a-multi-agent-harness.md b/blog/src/posts/what-is-a-multi-agent-harness.md
index 66176ece..451f7656 100644
--- a/blog/src/posts/what-is-a-multi-agent-harness.md
+++ b/blog/src/posts/what-is-a-multi-agent-harness.md
@@ -1,90 +1,123 @@
---
-title: "What Is a Multi-Agent Harness? (Plain-English Guide)"
-description: "A multi-agent harness coordinates several AI coding agents into one team — here's what that means, and how it differs from a single agent or a framework."
+title: "What Is a Multi-Agent Harness? A Plain English Guide"
+description: "A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one."
date: 2026-05-22
+updated: 2026-09-10
category: concepts
categoryLabel: Concepts
type: Non-technical
primaryKeyword: "claude code multi-agent"
-secondaryKeywords: ["multi-agent harness", "multi-agent ai framework", "ai agent harness"]
+secondaryKeywords: ["multi-agent harness", "what is an agent harness", "multi-agent ai framework", "ai agent harness", "claude code agent teams vs harness"]
tags: ["Concepts", "Multi-Agent", "Claude Code"]
author:
name: Chaitanya Giri
initials: CG
faq:
- q: "What is a multi-agent harness?"
- a: "A multi-agent harness is software that runs several AI agents at once and coordinates them — giving each a role, letting them message each other, sharing memory, and routing work — so they act as one team instead of isolated sessions."
+ a: "Software that runs several AI agents at the same time and makes them work as one team. Each agent gets a role, agents pass messages to each other, they share long term memory, and a coordinator hands out the work."
- q: "How is a harness different from a framework like LangGraph or CrewAI?"
- a: "A framework is a library you write an agent app with. A harness wraps agents you already run (like Claude Code terminals) and adds coordination — messaging, memory, orchestration, and visibility — without you rebuilding your agent from scratch."
+ a: "A framework is a library you write a new agent application with. A harness wraps agents you already run, like Claude Code or Codex in a terminal, and adds the coordination around them. You do not rebuild anything."
+ - q: "Is Claude Code's agent teams feature a multi-agent harness?"
+ a: "It covers part of the idea inside Claude Code. Agent teams are experimental and off by default. A lead session spawns teammates that share a task list and message each other, and a team belongs to that one session. A harness keeps agents working across sessions, remembers between them, and can mix CLIs from different vendors."
- q: "Do I need a multi-agent harness to use Claude Code?"
- a: "No. One Claude Code session is plenty for many tasks. A harness helps once you're running several at once and the coordination overhead (who does what, who knows what) starts to cost you time."
+ a: "No. One session is plenty for most tasks. A harness starts paying for itself when you run three or more agents at once and keeping them coordinated costs you more time than the work does."
+ - q: "Is Munder Difflin a multi-agent harness?"
+ a: "Yes. It is a free, open source harness that runs twelve terminal coding CLIs, including Claude Code, Codex, Gemini CLI and Copilot, on your own machine, with your clone Michael coordinating them."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p><strong>A multi-agent harness</strong> is
-software that runs several AI agents at once and makes them act like a team: each gets a role, they
-message each other, share long-term memory, and a coordinator routes work between them. It's the
-layer that turns "five chat windows" into "one office."</p></div>
+<div class="callout tldr"><span class="ic">TL;DR</span><p>A <strong>multi-agent harness</strong> runs
+several AI coding agents at once and makes them behave like a team: each has a role, they message each
+other, they share memory, and a coordinator routes the work. It is the layer that turns five terminal
+windows into one office.</p></div>
-If you've run more than one AI coding agent at the same time, you've felt the problem: two agents
-edit the same file, neither remembers what the other did, and you become the human message bus
-alt-tabbing between windows. A **multi-agent harness** is the software that fixes that.
+If you have ever run two AI coding agents at the same time, you already know the problem. They edit the
+same file, neither knows what the other did, and you become the message bus, copying context from one
+window to the next. A **multi-agent harness** is the software that takes that job off your hands.
-## A multi-agent harness, in one sentence
+## What is a multi-agent harness?
-> A multi-agent harness runs several AI agents concurrently and coordinates them — roles, messaging,
-> shared memory, and work routing — so they behave as a single team.
+It is software that runs several AI agents at the same time and coordinates them. That coordination comes
+down to four things: roles, messaging, shared memory and routing. Everything else is detail about how a
+particular harness does those four things.
-That's the whole idea. Everything else is detail about *how* the coordination happens.
+The easiest picture is an office. The agents are the staff. The harness is the building: desks, mailboxes,
+a filing cabinet everyone can read, and a manager who decides who does what.
{% img "note-1" %}
-## What it adds on top of a single agent
+## What does a harness add on top of a single agent?
-A single agent is a loop: read context, take an action, repeat. A harness wraps **many** of those
-loops and adds the parts a lone agent doesn't have:
+Five things a lone agent does not have. A single coding agent is a loop: read the context, act, check the
+result, repeat. A harness wraps many of those loops and adds:
-- **Roles.** Each agent gets a job (researcher, builder, reviewer) instead of all of them doing
- everything.
-- **Messaging.** Agents pass information to each other directly, rather than through you.
-- **Shared memory.** Durable, cross-session knowledge so agent B can use what agent A learned.
-- **Orchestration.** A coordinator that decomposes your intent and assigns work — in Munder Difflin
- that's the [GOD orchestrator](https://munderdiffl.in/#how), an agent you talk to in plain language.
-- **Visibility.** A way to *see* what the team is doing, so it's not a black box.
+- **Roles.** A researcher, a builder and a reviewer, instead of five generalists stepping on each other.
+- **Messaging.** Agents hand findings to each other directly, not through you.
+- **Shared memory.** What one agent learned on Monday is still there for another agent on Thursday.
+- **Orchestration.** One coordinator turns your request into tasks and assigns them. In Munder Difflin that
+ coordinator is Michael, a clone of you that you talk to in plain language.
+- **Visibility.** A way to see what the whole team is doing without tailing five terminals.
-{% img "note-2" %}
-
-## Harness vs. framework vs. subagents
+## What is the difference between a harness, a framework, subagents and agent teams?
-These three get conflated. The difference is what you bring to the table:
+What you bring, and how long the team lasts. These four get mixed up constantly, so here they are side by side.
### Framework
-A library (LangGraph, CrewAI, AutoGen) you **build an agent application with**. You write the graph,
-the tools, the prompts. Powerful, but it's a from-scratch build.
+
+A library such as LangGraph, CrewAI or AutoGen that you **build an agent application with**. You write the
+graph, the tools and the prompts. Powerful, and a build from scratch.
### Subagents
-A single agent **spawning helpers** inside its own run. Useful for fan-out, but the helpers are
-short-lived and scoped to that one parent — no shared memory across your whole workflow.
+
+One agent **spawning short lived helpers** inside its own run. Great for fanning out a search or a review.
+The helpers report back to their parent and are gone when the job ends.
+
+### Agent teams inside Claude Code
+
+Claude Code's own [agent teams](https://code.claude.com/docs/en/agent-teams) sit in between, and they are
+still experimental and off by default. A lead session spawns teammates that share a task list and message
+each other directly. A team belongs to a single session, and every teammate is a Claude Code instance.
### Harness
-Software that **wraps agents you already run** and coordinates them. You don't rebuild your agent;
-you keep using Claude Code, and the harness adds messaging, memory, orchestration, and a view of the
-floor. That's the [multi-agent harness](https://munderdiffl.in/#what) approach.
-## When you actually need one
+Software that **wraps the agents you already run** and keeps them working together over time. You keep
+using Claude Code, Codex or Gemini CLI. The harness adds messaging, memory that survives a restart, an
+orchestrator, and a view of the floor. Because it sits outside any single CLI, it can put agents from
+different vendors on the same team.
+
+{% img "note-2" %}
+
+## When do you actually need one?
+
+When coordinating your agents costs more time than the agents save you. The usual signs:
+
+- you run **three or more** sessions and lose track of which one is doing what,
+- you keep **explaining the same context** because every session starts from zero,
+- two agents **collide** on the same files,
+- you want work to **keep moving** while you are in a meeting, with the machine left on.
+
+If none of that sounds familiar, one session is fine, and you can stop reading here with a clear conscience.
+
+## What does a harness look like in practice?
+
+Here is how [Munder Difflin](https://munderdiffl.in/) does it. Every agent is a real terminal CLI running
+on your machine, and twelve are supported: Claude Code, Codex, Gemini CLI, Antigravity, Grok, Kimi Code,
+Qwen, OpenCode, Crush, Pi, Copilot and Cursor. Each agent gets a desk on a 2D office floor, a mailbox and
+long term memory. Michael, your clone, takes your request, breaks it into tasks, hires workers and routes
+messages between them. When something needs a human decision, it lands on the ASK ME board instead of
+vanishing into a scrollback.
-You don't need a harness to use Claude Code. You start wanting one when:
+The floor is a simulation and uses no tokens. The real work happens in the terminals underneath, on the
+subscriptions you already pay for and within their normal usage limits.
-- you're running **three or more** sessions and losing track of which is doing what,
-- you keep **re-explaining context** because each session forgets, or
-- two agents **collide** on the same files.
+## Where to go next
-If that's you, the next step is a practical one:
-[how to run multiple Claude Code agents](/blog/how-to-run-multiple-claude-code-agents/) without
-losing track, and [how to give Claude Code long-term memory](/blog/give-claude-code-long-term-memory/)
-so the team stops forgetting.
+- [How to manage multiple Claude Code sessions](/blog/manage-multiple-claude-code-sessions/) without losing
+ track of them.
+- [How to give Claude Code long term memory](/blog/give-claude-code-long-term-memory/) so the team stops
+ forgetting.
+- [The best tools to run multiple Claude Code agents](/blog/best-claude-code-multi-agent-tools/), compared.
---
-Munder Difflin is exactly this: a local, open-source multi-agent harness for Claude Code. If you
-want to watch a coordinated team of agents work an office floor,
-[download Munder Difflin](https://munderdiffl.in/#install) — it's free and MIT-licensed.
+Munder Difflin is free and open source under the MIT license. [Download it](https://munderdiffl.in/) for
+macOS, Windows or Linux and watch a coordinated team work an office floor.
diff --git a/blog/src/posts/your-first-hour-with-munder-difflin.md b/blog/src/posts/your-first-hour-with-munder-difflin.md
index 6c2f5376..c402c6da 100644
--- a/blog/src/posts/your-first-hour-with-munder-difflin.md
+++ b/blog/src/posts/your-first-hour-with-munder-difflin.md
@@ -1,97 +1,145 @@
---
title: "Your First Hour With Munder Difflin"
-description: "A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running."
+description: "A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running."
date: 2026-07-03
-updated: 2026-08-20
+updated: 2026-09-10
category: guides
categoryLabel: Guides
type: Non-technical
primaryKeyword: "munder difflin onboarding"
-secondaryKeywords: ["getting started with munder difflin", "multi-agent harness tutorial", "first hour with a multi-agent harness", "ai agent approval queue", "scheduled agent missions", "munder difflin skills"]
+secondaryKeywords: ["getting started with munder difflin", "munder difflin tutorial", "first hour with a multi-agent harness", "munder difflin setup", "scheduled agent missions", "munder difflin skills"]
tags: ["Guides", "Onboarding", "Getting Started", "Multi-Agent", "Local-First"]
author:
name: Chaitanya Giri
initials: CG
faq:
- q: "How long does it take to get Munder Difflin running?"
- a: "About five minutes. Grab the signed build for macOS, Windows, or Linux from the releases page, or run from source with Node.js 18+. Onboarding validates your setup at the very first step — as of v0.4.4 it checks your home folder immediately instead of failing four steps later, and the Prerequisites page in Settings shows live status for every tool an engine needs, with a button that asks Michael to install what's missing."
+ a: "About ten minutes if one coding CLI, like Claude Code, Codex or Antigravity, is already installed and signed in. Download the build for macOS, Windows or Linux, answer a short setup, and you are on the floor. Settings, then Prerequisites, shows anything that is still missing."
- q: "What do I need before installing Munder Difflin?"
- a: "At least one supported agent CLI — Claude Code, Antigravity, Codex, Grok, Kimi, Qwen, OpenCode, Crush, pi, or GitHub Copilot CLI. If you're missing one (or git, Node, uv, MemPalace), Settings → Prerequisites shows exactly what's absent with the platform-correct install command, and can hand the whole job to Michael."
+ a: "One supported terminal coding CLI, installed and signed in. Twelve are supported: Claude Code, Codex, Gemini CLI, Antigravity, Grok, Kimi Code, Qwen, OpenCode, Crush, Pi, Copilot and Cursor. If you do not pay for any AI subscription, Antigravity is free."
- q: "Who is Michael and how do I give him work?"
- a: "Michael is a clone of you — the boss of the agents, while you stay the boss of him. You type a brief into his terminal (or talk to him by voice) and he adjudicates it: creating tasks, assigning them to workers, routing messages between inboxes, and escalating only critical items back to you. He seats himself in his office automatically on first launch."
+ a: "Michael is your clone, the boss of the floor, and you stay the boss of him. You type what you want into his terminal or talk to him by voice. He breaks it into tasks, hires workers, routes messages between them, and brings you only the decisions that need you."
- q: "Do I have to approve everything the agents do?"
- a: "No. Michael resolves routine requests himself so the system stays autonomous. Only critical items — spend, destructive operations, scope changes — land in the human-in-the-loop approvals queue for you to act on. A circuit breaker and per-agent token budgets guard the rest."
- - q: "How do I review what an agent actually changed?"
- a: "Open the built-in Monaco IDE from the title-bar IDE button. Its git CHANGES rail lists every modified file, and clicking one opens a read-only side-by-side diff against HEAD. There's a per-agent git tab with status, log, and commit graph — and since v0.4.4 the IDE previews images too, so a design change is as reviewable as a code change."
- - q: "Can Munder Difflin keep working after I walk away?"
- a: "Yes — that's the point. The Triggers tab in the Command Center holds recurring missions with a label, interval, target agent, and body, plus a heartbeat that re-engages the floor when it goes quiet. It's local-first, so the office runs on hardware you already own — and the app keeps itself current, telling you exactly what's in each update before you restart into it."
+ a: "Only if you want to. In ask first mode agents pause for tool approval, which is a good way to start. With auto mode on they carry on alone, and anything that genuinely needs you lands on the ASK ME board. Per agent token budgets and a circuit breaker catch runaway agents either way."
+ - q: "How do I review what an agent changed?"
+ a: "Open the IDE on the agent you are looking at. You can browse and edit files in its workspace and see its uncommitted changes as a diff before you accept anything."
+ - q: "Does Munder Difflin keep working after I walk away?"
+ a: "Yes, as long as the machine stays on. The Triggers tab holds schedules that start work on an interval or on chosen weekdays, and agents keep working through their queue while you are away. Hosted sandboxes that keep going with the lid closed are not available yet."
+ - q: "Is Munder Difflin free?"
+ a: "Yes. The classic office is free and MIT licensed, with unlimited local agents. Pro and Teams are optional paid plans, each with a 14 day trial."
---
-<div class="callout tldr"><span class="ic">TL;DR</span><p><strong>You can go from zero to a running AI office in one hour.</strong> Minute 0: install. Minute 5: onboarding — it opens on the honest pitch, <strong>a clone of you, working 24/7</strong>, and validates your setup at step one. Minute 10: your <strong>first brief to Michael</strong>. Minute 20: watch avatars work the floor and open a real desk terminal. Minute 40: your <strong>first approval</strong> and a side-by-side diff in the built-in <strong>Monaco IDE</strong>. Minute 60: set a <strong>schedule</strong> and walk away — the office keeps working.</p></div>
+<div class="callout tldr"><span class="ic">TL;DR</span><p><strong>Zero to a working AI office in an
+hour.</strong> Minute 0: download. Minute 5: a short setup, including your clone's engine. Minute 15: your
+<strong>first job for Michael</strong>. Minute 25: watch the floor and open a real terminal. Minute 40:
+answer an <strong>ASK ME</strong> card and read a <strong>diff</strong>. Minute 50: give an agent a
+<strong>skill</strong>. Minute 60: leave a <strong>schedule</strong> running and walk away.</p></div>
<video controls preload="none" playsinline poster="/media/demo/intro-poster.jpg" style="width:100%; border-radius:12px; margin:12px 0 24px;">
<source src="/media/demo/intro.mp4" type="video/mp4" />
</video>
-Most tools ask you to learn them before they do anything. Munder Difflin is the other kind: within an hour you've briefed an orchestrator, watched agents work, approved a real change, and left the office running without you. Here's that hour, minute by minute — current as of v0.4.4.
+Most tools want you to study them before they do anything useful. This one should be doing real work inside an
+hour: you brief an orchestrator, watch agents work, check a real change, and leave the office running without you.
+Here is that hour, minute by minute, current as of Munder Difflin 0.5.2.
-{% youtube "", "Your first hour with Munder Difflin — full walkthrough" %}
+## Minute 0: Download and install
-## Minute 0 — Install
+Grab the build for your system from [munderdiffl.in](https://munderdiffl.in/) or the
+[latest release](https://github.com/chaitanyagiri/munder-difflin/releases/latest):
-Two paths: grab a signed build (macOS, Windows, Linux) from the [releases page](https://github.com/chaitanyagiri/munder-difflin/releases/latest), or clone and run from source with `npm install && npm run dev`. You'll want at least one supported agent CLI on your `PATH` — the engine card now honestly names all ten: Claude Code, Antigravity, Codex, Grok, Kimi, Qwen, OpenCode, Crush, pi, and GitHub Copilot CLI.
+- **macOS:** one universal `.dmg` for Apple Silicon and Intel, signed and notarized by Apple.
+- **Windows 10 and 11:** a setup installer, or a portable `.exe` if you would rather not install anything.
+- **Linux:** an `.AppImage`.
-Missing something? You no longer have to find out the hard way. **Settings → Prerequisites** shows live status for git, Node, uv, MemPalace and every engine — real paths, platform-correct install commands, and a button that simply asks Michael to fill the gaps for you. The [install and usage guide](/blog/how-to-install-and-use-munder-difflin/) covers the deeper details. Budget five minutes.
+Each release also carries a `SHA256SUMS.txt`, so you can check the file you downloaded is the real one.
-{% img "note-1", "Prerequisites, before they become surprises: live status for every tool, and a button that hands the gaps to Michael." %}
+You need one terminal coding CLI installed and signed in. Twelve are supported: Claude Code, Codex, Gemini CLI,
+Antigravity, Grok, Kimi Code, Qwen, OpenCode, Crush, Pi, Copilot and Cursor. If you do not pay for any AI
+subscription, Antigravity is free. The [install guide](/blog/how-to-install-and-use-munder-difflin/) has the exact
+commands for every one of them.
-## Minute 5 — Onboarding: meet your clone
+## Minute 5: Setup
-First launch opens on the pitch the whole product keeps: **a clone of you, working 24/7.** The wizard's few questions are the actual shape of the product:
+First launch walks you through a short setup. It starts by asking whether you are technical, which only changes how
+much jargon the app shows you. Then four steps:
-- **Your clone's engine** — Michael runs on a pluggable engine; pick from the ten, change it later.
-- **Harness home** — the folder where the hive lives: per-agent memory, mailboxes, the blackboard, the event log. Plain files in a local git repo. The wizard validates this at step one — an empty or impossible folder fails immediately, not after step four.
-- **Registered repos** — the codebases your agents will work on.
-- **Auto-mode** — whether agents proceed without per-tool prompts (the approval gate still catches critical items).
+- **A home folder** for the app's own files: settings, agent memory and mailboxes. Use a new, empty folder.
+- **Your clone.** Name it and pick the engine that powers it. Give it the most capable model you have, because it
+ does the thinking and the delegating.
+- **Your projects.** The folders agents may work in. A project is just a folder.
+- **Permissions.** Whether agents act on their own or ask first. Start with ask first. You can change it any time.
-Finish the wizard and you land on the floor: a pixel-art office, empty except for Michael, who seats himself in his office automatically — with a **BOSS** tag on his card, because he's the boss of the agents while you stay the boss of him. Everything the hive needs starts with him: as of v0.4.4 a brand-new install boots its message router, hook server and mission scheduler on the very first run.
+Every install starts in the classic office, the free version, with the floor you are about to meet.
-## Minute 10 — Your first brief to Michael
+{% img "note-1", "Prerequisites in Settings shows the live status of every tool an engine needs, before it becomes a surprise." %}
-You don't manage the workers. You talk to your clone, and he runs the floor. Click into Michael's terminal and type a brief the way you'd brief a colleague: what you want, which repo, what "done" looks like. Prefer talking? His voice mode opens with a live snapshot of the floor and can run nearly the whole app.
+## Minute 15: Your first job for Michael
-Michael adjudicates. He creates tasks on the kanban, assigns them, and routes messages between agent inboxes. Workers you hire via **Add agent** each get a real CLI process in its own pseudo-terminal and, with the git-isolation toggle, their own worktree — so nobody collides on branches. How he decides what to route, resolve, or escalate is its own post: [how the orchestrator works](/blog/how-the-god-orchestrator-works/).
+You do not manage the workers. You talk to your clone, and he runs the floor. Click into Michael's terminal and
+describe the job the way you would brief a capable new hire: the outcome, the folder, and what done looks like.
+Prefer talking? He has a voice mode.
-A good first brief is small and self-contained: "read this repo and write a REPORT.md summarizing the architecture" beats "refactor everything" for hour one.
+A good first job is small and self contained. "Read this repo and write REPORT.md explaining how it fits together"
+beats "refactor everything" for hour one.
-## Minute 20 — Watch the floor, then open a desk
+Michael turns the job into tasks on the kanban, hires a worker if he needs one, and routes messages between their
+mailboxes. Each worker is a real CLI process in its own terminal and, with git isolation on, its own worktree. How he
+decides what to handle and what to send your way is its own post: [how Michael routes work](/blog/how-the-god-orchestrator-works/).
-Now the part that makes the product legible: the floor is not a decoration, it's the state of the system. Avatars walk to stations as they work. When the hive routes a message, an envelope flies from sender to recipient; escalations fly to the door. The cast is an affectionate parody of The Office, and every movement maps to a real event.
+## Minute 25: Watch the floor, then open a desk
-{% img "floor-view", "The floor mid-task: every movement maps to a real event." %}
+The floor is not decoration. It is the state of the system. Characters walk to their desks as they work, and envelopes
+fly between them when they message each other. The cast is an affectionate parody of The Office, and the whole floor is
+a simulation that uses no tokens.
-Click any agent and you get their desk: the live terminal (you can type back into it), a sandboxed file browser, and a git tab with status, log, and commit graph. Go fullscreen and the roster cards show each agent's model, project, and a live context gauge — the fuel dial for every worker at a glance. This is the moment the abstraction clicks: that avatar is a real CLI process, and you're reading its actual stdout.
+{% img "floor-view", "The floor mid task: every movement maps to a real event." %}
-{% img "note-2", "Every desk is real: a live terminal, a file browser, a git tab — and a context gauge telling you how much runway the agent has left." %}
+Click any agent to open its terminal. You can read the live output and type straight back into it. Want to work with
+one agent without the office around it? Focus mode gives you the terminal full width, and Esc brings the floor back.
+This is the moment it clicks: that character is a real CLI process, and you are reading its actual output.
-## Minute 40 — Your first approval, and the diff
+{% img "note-2", "Every desk is real: a live terminal, plus a context gauge showing how much runway the agent has left." %}
-Sometime in the first hour, something lands in the approvals queue — a spend threshold, a destructive operation, a scope change. This is by design: Michael resolves routine requests himself and escalates only critical items, so the queue stays short and every item in it deserves your attention. (The philosophy behind that gate: [approving AI agents without babysitting them](/blog/human-in-the-loop-approving-ai-agents/).)
+## Minute 40: Answer an ASK ME card, then read the diff
-Before you click approve, look at the work. Hit the title-bar **IDE** button and the built-in Monaco editor — the VS Code editor engine, fully self-hosted — opens over the floor. The git CHANGES rail lists what changed; click a file for a read-only side-by-side diff against HEAD. Images preview right in the IDE now, so a screenshot or SVG an agent produced is one click to inspect. Read the diff, approve the item, watch the envelope fly.
+At some point in the first hour an agent will need you. It might be a question only you can answer, or a step only you
+can take. It shows up on the **ASK ME** board instead of hiding in a scrollback, formatted so you can read it at a
+glance. Michael settles routine questions himself, so whatever reaches you deserves your attention.
-## Minute 50 — Give an agent a skill
+Before you say yes to a change, look at it. Open the **IDE** on the agent's workspace to browse its files and see its
+uncommitted changes as a diff. Read it, answer the card, and watch the work carry on. The thinking behind that gate is in
+[approving AI agents without babysitting them](/blog/human-in-the-loop-approving-ai-agents/).
-New in v0.4.4 and worth five minutes of your first hour: **Skills.** A browsable catalog of hundreds of skills — search, category and publisher filters — installable across Claude Code, OpenCode and Codex with scope precedence handled for you. Give your reviewer a review checklist, your writer a style guide, your researcher a source-citing discipline. It's the difference between hiring generalists and hiring people who've read the manual. (Background: [MCP and skills in a hive](/blog/mcp-and-skills-in-a-hive/).)
+## Minute 50: Give an agent a skill
-## Minute 60 — Leave it running
+The **skills** tab is a catalog of skills you can install for your agents. Give your reviewer a review checklist, your
+writer a style guide, and your researcher the habit of citing sources. It is the difference between hiring generalists
+and hiring people who actually read the manual. More in [MCP and skills in a hive](/blog/mcp-and-skills-in-a-hive/).
-The last move of the hour is the one that changes your relationship with the tool: schedule something. The Command Center's **Triggers** tab takes a label, an interval, a target agent, and a mission body — "every morning, triage new GitHub issues," say — and a heartbeat re-engages the floor if it goes quiet. Last-fired and next-fired times are right there in the tab.
+## Minute 60: Leave it running
-Close the laptop. It's local-first, so tomorrow the office is exactly where you left it — probably with mail in your queue. And when the next release ships, the app tells you itself: the update toast says what's actually inside before you restart into it. For patterns on what's worth automating, see [scheduling autonomous agent missions](/blog/scheduling-autonomous-agent-missions/).
+The last move of the hour is the one that changes how you use the tool: schedule something. Open the **Triggers** tab and
+create a schedule with a label, who it goes to, and the prompt, which is sent word for word on every run. It can repeat on
+an interval or on chosen weekdays at a set time. "Every weekday at 9, triage new GitHub issues and summarise them for me" is
+a good first one.
+
+Leave the machine on and walk away. The agents keep working, and when a new version ships the app updates itself. Hosted
+sandboxes that keep working with the lid closed are not available yet, so for now the office runs wherever you run it. For
+ideas on what is worth automating, read [scheduling autonomous agent missions](/blog/scheduling-autonomous-agent-missions/).
+
+## Want the office in one window?
+
+Everything above is the free classic office. If you would rather work in a single window, **Pro** puts the orchestrator,
+agents, tasks, inbox, automations and memory in one workspace and adds **Stapler**, a small floating window that sends a
+screenshot, a recorded message or a meeting transcript to your orchestrator with one line from you. **Teams** lets your
+clone work with your teammates' clones. Both come with a 14 day trial, and the [pricing page](https://munderdiffl.in/#pricing)
+has the details.
## The hour, in one line
-Install, meet your clone, brief him once, watch the floor, read one diff, install one skill, set one schedule — and you've gone from "a CLI in a terminal" to "an office that works while you don't."
+Install, set up your clone, give him one job, watch the floor, answer one card, read one diff, install one skill, set one
+schedule. That is the whole distance from "a CLI in a terminal" to "an office that works while you do not".
-[Download Munder Difflin](https://github.com/chaitanyagiri/munder-difflin/releases/latest) — free, MIT-licensed, local-first — and if the first hour earns it, [a GitHub star](https://github.com/chaitanyagiri/munder-difflin) helps more people find it.
+[Download Munder Difflin](https://munderdiffl.in/), free and MIT licensed. If the first hour earns it,
+[a GitHub star](https://github.com/chaitanyagiri/munder-difflin) helps other people find it.
diff --git a/docs/blog/agent-security-and-sandboxing/index.html b/docs/blog/agent-security-and-sandboxing/index.html
index 07a0e4f2..843dfcc5 100644
--- a/docs/blog/agent-security-and-sandboxing/index.html
+++ b/docs/blog/agent-security-and-sandboxing/index.html
@@ -313,13 +313,13 @@ git, lifecycle hooks, and an audit trail — so safe-by-default is the path of l
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -345,10 +345,10 @@ git, lifecycle hooks, and an audit trail — so safe-by-default is the path of l
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/best-ai-coding-agents/index.html b/docs/blog/best-ai-coding-agents/index.html
index a3fa46d6..3b0fd2c6 100644
--- a/docs/blog/best-ai-coding-agents/index.html
+++ b/docs/blog/best-ai-coding-agents/index.html
@@ -363,13 +363,13 @@ source.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/best-claude-code-multi-agent-tools/index.html b/docs/blog/best-claude-code-multi-agent-tools/index.html
index 2ba1ec31..dabc8a87 100644
--- a/docs/blog/best-claude-code-multi-agent-tools/index.html
+++ b/docs/blog/best-claude-code-multi-agent-tools/index.html
@@ -4,7 +4,7 @@
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>The Best Tools to Run Multiple Claude Code Agents (2026) — Munder Difflin Blog</title>
-<meta name="description" content="An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared." />
+<meta name="description" content="An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one." />
<link rel="canonical" href="https://munderdiffl.in/blog/best-claude-code-multi-agent-tools/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -15,7 +15,7 @@
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
<meta property="og:title" content="The Best Tools to Run Multiple Claude Code Agents (2026) — Munder Difflin Blog" />
-<meta property="og:description" content="An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared." />
+<meta property="og:description" content="An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one." />
<meta property="og:url" content="https://munderdiffl.in/blog/best-claude-code-multi-agent-tools/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -30,7 +30,7 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="The Best Tools to Run Multiple Claude Code Agents (2026) — Munder Difflin Blog" />
-<meta name="twitter:description" content="An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared." />
+<meta name="twitter:description" content="An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -44,10 +44,10 @@
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "The Best Tools to Run Multiple Claude Code Agents (2026)",
- "description": "An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared.",
+ "description": "An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-06-04T00:00:00.000Z",
- "dateModified": "2026-06-04T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -55,7 +55,7 @@
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/best-claude-code-multi-agent-tools/" },
- "keywords": "best claude code multi-agent tools, claude code multi-agent tool, agentic coding tools, best tools to run multiple claude code agents",
+ "keywords": "best claude code multi-agent tools, claude code multi-agent tool, best tools to run multiple claude code agents, claude code agent teams, crystal nimbalyst, emdash coding agents, conductor claude code",
"articleSection": "Comparisons",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/best-claude-code-multi-agent-tools/"
@@ -78,9 +78,11 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "What is the best tool to run multiple Claude Code agents?", "acceptedAnswer": { "@type": "Answer", "text": "There's no single best — it depends on what you want. Claude Squad is the leanest terminal option; Conductor is the most polished on macOS; Crystal is a good open-source desktop GUI; vibe-kanban gives you a task board; and Munder Difflin adds shared memory, inter-agent messaging, and a GOD orchestrator so the agents act as one coordinated team." } },
- { "@type": "Question", "name": "Are these Claude Code multi-agent tools free?", "acceptedAnswer": { "@type": "Answer", "text": "Most are free and several are open source (Claude Squad, Crystal, vibe-kanban, and Munder Difflin are open source; Munder Difflin is MIT-licensed). Conductor is a free, native macOS app. Always check each project's current license and pricing before you commit." } },
- { "@type": "Question", "name": "Do I need a multi-agent tool to use Claude Code?", "acceptedAnswer": { "@type": "Answer", "text": "No. One Claude Code session handles most tasks. You start wanting a multi-agent tool once you're running three or more sessions at once and the coordination overhead — who's doing what, who knows what, who edits which file — starts costing you time." } }
+ { "@type": "Question", "name": "What is the best tool to run multiple Claude Code agents?", "acceptedAnswer": { "@type": "Answer", "text": "It depends on your bottleneck. Claude Code agent teams are the built in option for a single session. Claude Squad is the leanest terminal manager. Conductor is the polished Mac app, with cloud workspaces on paid plans. Nimbalyst and Emdash are open source desktop apps. Munder Difflin adds long term memory, messaging and an orchestrator across twelve CLIs." } },
+ { "@type": "Question", "name": "Does Claude Code have a built in way to run multiple agents?", "acceptedAnswer": { "@type": "Answer", "text": "Yes, two. Subagents handle short helper jobs inside one session. Agent teams, which are experimental and stay off until you set CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, let a lead session spawn teammates that share a task list and message each other." } },
+ { "@type": "Question", "name": "What happened to Crystal?", "acceptedAnswer": { "@type": "Answer", "text": "Its maker deprecated Crystal in February 2026 and replaced it with Nimbalyst, an MIT licensed visual workspace for Claude Code and Codex." } },
+ { "@type": "Question", "name": "Is Vibe Kanban still maintained?", "acceptedAnswer": { "@type": "Answer", "text": "Its website says Vibe Kanban is sunsetting and will continue as an open source, community maintained project." } },
+ { "@type": "Question", "name": "Are these tools free?", "acceptedAnswer": { "@type": "Answer", "text": "Most have a free option. Claude Squad (AGPL 3.0), Nimbalyst (MIT), Emdash (Apache 2.0), Vibe Kanban (Apache 2.0) and the Munder Difflin app (MIT) are open source. Conductor has a free plan for local workspaces and paid plans for cloud and team features. Check pricing before you commit, because this space moves fast." } }
]
}
@@ -150,7 +152,7 @@
<a href="/blog/topics/comparisons/">Comparisons</a>
</nav>
<h1>The Best Tools to Run Multiple Claude Code Agents (2026)</h1>
- <p class="dek">An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared.</p>
+ <p class="dek">An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
@@ -166,182 +168,194 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p>The tools for running
-<strong>multiple Claude Code agents</strong> fall into three camps: <strong>terminal session
-managers</strong> (Claude Squad), <strong>parallel-worktree desktop apps</strong> (Conductor,
-Crystal), <strong>task boards</strong> (vibe-kanban), and <strong>coordinated hives</strong> (Munder
-Difflin). The first three help you run agents <em>side by side</em>; a hive adds shared memory,
-messaging, and an orchestrator so they run <em>as a team</em>. Pick by how much coordination you
-actually need.</p></div>
-<p>If you’ve outgrown a single Claude Code window, the next question is <em>which tool</em> helps you run
-several at once without the chaos. The landscape moved fast in 2025–2026, so here’s an honest,
-plain-English roundup of the main options — what each does well, where it stops, and who it’s for.</p>
-<blockquote>
-<p><strong>A note on fairness:</strong> this space changes quickly. Features, platforms, and licenses below are
-accurate to the best of our knowledge at the time of writing — always check each project’s repo or
-site for current details. Munder Difflin is our own tool; we’ve tried to describe the others on
-their own terms, not as straw men.</p>
-</blockquote>
-<h2 id="the-four-shapes-of-multi-agent-claude-code" tabindex="-1">The four shapes of “multi-agent Claude Code” <a class="anchor" href="#the-four-shapes-of-multi-agent-claude-code" aria-hidden="true">#</a></h2>
-<p>Before the tools, the categories — because they answer different questions:</p>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p>Tools for running <strong>multiple Claude Code
+agents</strong> come in four shapes: <strong>built in</strong> (Claude Code agent teams),
+<strong>session managers</strong> (Claude Squad), <strong>parallel workspace apps</strong> (Conductor,
+Nimbalyst, Emdash, and Vibe Kanban as a board), and <strong>coordinated offices</strong> (Munder Difflin).
+The first three help you run agents side by side. An office adds shared memory, messaging and an
+orchestrator so they run as a team.</p></div>
+<p>A lot changed since we first wrote this in June. Claude Code grew its own agent teams, Crystal became
+Nimbalyst, Conductor moved into the cloud, Vibe Kanban announced it is winding down, and Emdash arrived.
+Here is the current picture, checked on 10 September 2026.</p>
+<div class="callout note"><span class="ic">Fair warning</span><p>Munder Difflin is our own tool. Everything
+about the others comes from their own sites, READMEs and docs as of 10 September 2026, and every tool is
+linked so you can check it yourself.</p></div>
+<h2 id="what-kinds-of-tools-run-multiple-claude-code-agents" tabindex="-1">What kinds of tools run multiple Claude Code agents? <a class="anchor" href="#what-kinds-of-tools-run-multiple-claude-code-agents" aria-hidden="true">#</a></h2>
+<p>Four kinds, and knowing which one you need is most of the decision.</p>
<ul>
-<li><strong>Terminal session managers</strong> run each agent in its own shell and let you switch between them from
-one keyboard-driven interface. Minimal, fast, SSH-friendly.</li>
-<li><strong>Parallel-worktree desktop apps</strong> give each agent an isolated <a href="/blog/how-to-run-multiple-claude-code-agents/">git
-worktree</a> and a GUI to launch tasks and review
-diffs. Great for “try five approaches in parallel.”</li>
-<li><strong>Task boards</strong> model the work as cards you assign to agents and move through columns — a
-project-management view over agentic coding.</li>
-<li><strong>Coordinated hives</strong> add the layer the others leave to you: shared long-term memory, direct
-agent-to-agent messaging, and an orchestrator that routes work. That’s the <a href="/#what">multi-agent
-harness</a> idea.</li>
+<li><strong>Built in.</strong> Features inside Claude Code itself. Nothing new to install, and the team lives in one session.</li>
+<li><strong>Session managers.</strong> Many agents in terminal sessions, driven from one keyboard friendly interface.</li>
+<li><strong>Parallel workspace apps.</strong> A desktop app that gives each agent an isolated workspace and a place to
+review what it changed.</li>
+<li><strong>Coordinated offices.</strong> Agents with roles, mailboxes, shared long term memory and an orchestrator that
+routes the work.</li>
</ul>
-<p>Most tools live cleanly in one camp. Knowing which camp you need is most of the decision.</p>
-<h2 id="the-tools-at-a-glance" tabindex="-1">The tools, at a glance <a class="anchor" href="#the-tools-at-a-glance" aria-hidden="true">#</a></h2>
+<h2 id="the-tools-at-a-glance" tabindex="-1">The tools at a glance <a class="anchor" href="#the-tools-at-a-glance" aria-hidden="true">#</a></h2>
<div class="table-scroll">
<table>
<thead>
<tr>
<th>Tool</th>
-<th>What it is</th>
-<th>How it parallelizes</th>
-<th>Shared memory</th>
-<th>Orchestrator</th>
+<th>Shape</th>
+<th>Agents it runs</th>
+<th>Platforms</th>
<th>License</th>
</tr>
</thead>
<tbody>
<tr>
+<td>Claude Code agent teams</td>
+<td>Built in, experimental</td>
+<td>Claude Code</td>
+<td>Wherever Claude Code runs</td>
+<td>Part of Claude Code</td>
+</tr>
+<tr>
<td>Claude Squad</td>
-<td>Terminal (TUI) session manager</td>
-<td>tmux + git worktrees</td>
-<td>No</td>
-<td>No (you assign)</td>
-<td>Open source</td>
+<td>Session manager</td>
+<td>Claude Code, Codex, Gemini, Aider and more</td>
+<td>Terminal, needs tmux</td>
+<td>AGPL 3.0</td>
</tr>
<tr>
<td>Conductor</td>
-<td>Native macOS desktop app</td>
-<td>Parallel git worktrees</td>
-<td>No</td>
-<td>No (you assign)</td>
-<td>Free, macOS-only</td>
+<td>Parallel workspaces, cloud on paid plans</td>
+<td>Claude Code, Codex, Cursor, OpenCode</td>
+<td>macOS, iOS listed as coming</td>
+<td>Free and paid plans</td>
+</tr>
+<tr>
+<td>Nimbalyst</td>
+<td>Visual workspace, parallel sessions</td>
+<td>Claude Code, Codex</td>
+<td>macOS, Windows, Linux</td>
+<td>MIT</td>
</tr>
<tr>
-<td>Crystal</td>
-<td>Open-source desktop app</td>
-<td>Parallel sessions + worktrees</td>
-<td>No</td>
-<td>No (you assign)</td>
-<td>Open source</td>
+<td>Emdash</td>
+<td>Parallel workspaces</td>
+<td>Any provider, including Claude Code, Codex, Amp, Antigravity</td>
+<td>macOS, Windows, Linux</td>
+<td>Apache 2.0</td>
</tr>
<tr>
-<td>vibe-kanban</td>
-<td>Kanban board for agents</td>
-<td>Task cards across agents</td>
-<td>No</td>
-<td>Board, not auto-routing</td>
-<td>Open source</td>
+<td>Vibe Kanban</td>
+<td>Task board, sunsetting</td>
+<td>Claude Code, Codex and others</td>
+<td>Runs with npx</td>
+<td>Apache 2.0</td>
</tr>
<tr>
<td>Munder Difflin</td>
-<td>Local coordinated hive</td>
-<td>Roles + mailboxes + orchestrator</td>
-<td>Yes (MemPalace)</td>
-<td>Yes (GOD agent)</td>
-<td>Open source (MIT)</td>
+<td>Coordinated office</td>
+<td>Twelve CLIs, including Claude Code, Codex, Gemini CLI, Copilot, Cursor</td>
+<td>macOS, Windows, Linux</td>
+<td>MIT (free app)</td>
</tr>
</tbody>
</table>
</div>
-<p>The single row that most separates the field is <strong>shared memory + orchestrator</strong>: that’s the jump
-from “running agents in parallel” to “running a team.”</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/best-claude-code-multi-agent-tools/note-1.png" alt="Hand-drawn sketch from “The Best Tools to Run Multiple Claude Code Agents (2026)”" loading="lazy" decoding="async" /></figure>
-<h2 id="claude-squad-the-lean-terminal-option" tabindex="-1">Claude Squad — the lean terminal option <a class="anchor" href="#claude-squad-the-lean-terminal-option" aria-hidden="true">#</a></h2>
-<p><a href="https://github.com/smtg-ai/claude-squad">Claude Squad</a> is a terminal UI that manages multiple AI
-coding agents — Claude Code among them — each in its own tmux session, with git worktrees keeping
-their changes isolated. You launch, switch, background, and review agents without leaving the
-terminal.</p>
+<h2 id="claude-code-agent-teams-the-built-in-option" tabindex="-1">Claude Code agent teams: the built in option <a class="anchor" href="#claude-code-agent-teams-the-built-in-option" aria-hidden="true">#</a></h2>
+<p><a href="https://code.claude.com/docs/en/agent-teams">Agent teams</a> are Claude Code’s own way to run a team, and
+they are still experimental. Set <code>CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1</code> and a lead session can spawn
+teammates. Each teammate is a separate Claude Code instance with its own context window. They share a task
+list, claim work, and message each other directly. You can keep them all in one terminal, or give each one
+a split pane with tmux or iTerm2.</p>
<ul>
-<li><strong>Strengths:</strong> featherweight, keyboard-driven, works great over SSH, no GUI overhead. If you live
-in the terminal and want parallel agents <em>now</em>, it’s hard to beat for speed of setup.</li>
-<li><strong>Where it stops:</strong> agents don’t share memory or message each other, and there’s no orchestrator —
-you decide who does what. It’s a superb <em>session manager</em>, not a coordination layer.</li>
+<li><strong>Strengths:</strong> nothing to install, first party, and teammates genuinely talk to each other.</li>
+<li><strong>Where it stops:</strong> Anthropic’s docs list the limits. A session has exactly one team, in process
+teammates are not restored when you resume a session, and token use grows with every teammate. Every
+teammate is Claude.</li>
</ul>
-<p>Best for: terminal purists running a handful of independent tasks. If you want a deeper look at why
-isolation alone isn’t coordination, see <a href="/blog/claude-squad-alternative/">Claude Squad alternative</a>.</p>
-<h2 id="conductor-the-polished-macos-experience" tabindex="-1">Conductor — the polished macOS experience <a class="anchor" href="#conductor-the-polished-macos-experience" aria-hidden="true">#</a></h2>
-<p><a href="https://conductor.build">Conductor</a> is a native macOS app for running Claude Code agents in
-parallel, each in its own isolated git worktree, with a clean Mac-native interface for kicking off
-tasks and reviewing their diffs.</p>
+<p>Best for: research, review and debugging jobs you finish in one sitting.</p>
+<h2 id="claude-squad-the-lean-terminal-manager" tabindex="-1">Claude Squad: the lean terminal manager <a class="anchor" href="#claude-squad-the-lean-terminal-manager" aria-hidden="true">#</a></h2>
+<p><a href="https://github.com/smtg-ai/claude-squad">Claude Squad</a> is a terminal app that manages several agents,
+including Claude Code, Codex, Gemini and Aider, each in its own workspace. It needs tmux and installs as <code>cs</code>.</p>
<ul>
-<li><strong>Strengths:</strong> genuinely nice UX, fast diff review, and a low-friction way to run several
-worktree-isolated agents at once on a Mac.</li>
-<li><strong>Where it stops:</strong> it’s macOS-only, and it’s built around <em>parallel isolated workspaces</em> rather
-than a coordinated team — no shared cross-agent memory, no agent-to-agent messaging, no
-plain-language orchestrator. Closed-source, so you can’t read or extend the internals.</li>
+<li><strong>Strengths:</strong> light, fast, keyboard driven, and comfortable over SSH.</li>
+<li><strong>Where it stops:</strong> its README describes managing sessions. It does not describe shared memory or agents
+messaging each other, so deciding who does what stays with you.</li>
</ul>
-<p>Best for: Mac users who want a refined parallel-worktree workflow. If you’re cross-platform or want a
-coordination layer, see <a href="/blog/conductor-claude-code-alternative/">a Conductor alternative</a>.</p>
-<h2 id="crystal-the-open-source-desktop-gui" tabindex="-1">Crystal — the open-source desktop GUI <a class="anchor" href="#crystal-the-open-source-desktop-gui" aria-hidden="true">#</a></h2>
-<p><a href="https://github.com/stravu/crystal">Crystal</a> is an open-source desktop app that runs multiple Claude
-Code sessions in parallel, each in its own git worktree. You spin up sessions from prompts, run them
-concurrently, and review diffs and test results across worktrees in one window.</p>
+<p>Best for: terminal people running a handful of independent tasks. More in
+<a href="/blog/claude-squad-vs-munder-difflin/">Claude Squad vs Munder Difflin</a>.</p>
+<h2 id="conductor-the-polished-mac-app-now-with-a-cloud" tabindex="-1">Conductor: the polished Mac app, now with a cloud <a class="anchor" href="#conductor-the-polished-mac-app-now-with-a-cloud" aria-hidden="true">#</a></h2>
+<p><a href="https://conductor.build">Conductor</a> started as a Mac app for running coding agents in parallel, each in its
+own workspace. It now runs Claude Code, Codex, Cursor and OpenCode, and pitches itself as a way to run a team
+of coding agents in the cloud. The free plan covers running agents in parallel in local workspaces on your
+Mac. Cloud workspaces come with paid plans, and its Teams plan adds live collaboration on shared workspaces.</p>
<ul>
-<li><strong>Strengths:</strong> open source, a real GUI, and a tidy model for running several experiments in
-parallel and comparing the results.</li>
-<li><strong>Where it stops:</strong> like the others in this camp, it’s parallel isolated sessions — agents don’t
-pool memory or talk to each other, and there’s no orchestrator routing work between them.</li>
+<li><strong>Strengths:</strong> a refined interface, quick review, and cloud workspaces that can run for hours.</li>
+<li><strong>Where it stops:</strong> the desktop app is macOS only today, with iOS listed as coming soon, and the cloud and
+multiplayer parts are paid.</li>
</ul>
-<p>Best for: developers who want an open, visual, worktree-based parallel workflow without the terminal.
-More on the trade-offs in <a href="/blog/crystal-claude-code-alternative/">a Crystal alternative</a>.</p>
+<p>Best for: Mac users who want polish and are happy to pay for cloud runs. See
+<a href="/blog/conductor-claude-code-alternative/">a Conductor alternative</a>.</p>
+<h2 id="nimbalyst-what-crystal-became" tabindex="-1">Nimbalyst: what Crystal became <a class="anchor" href="#nimbalyst-what-crystal-became" aria-hidden="true">#</a></h2>
+<p><a href="https://github.com/stravu/crystal">Crystal</a> was deprecated in February 2026 and replaced by
+<a href="https://nimbalyst.com">Nimbalyst</a>, which describes itself as an open source visual workspace for building
+with Codex, Claude Code and more. It runs parallel sessions with optional git worktree isolation, adds visual
+editors for markdown, mockups, diagrams and data models, and organises sessions on a kanban. The desktop apps
+for macOS, Windows and Linux are MIT licensed.</p>
+<ul>
+<li><strong>Strengths:</strong> open source, cross platform, and good at reviewing what agents change, including non code files.</li>
+<li><strong>Worth checking:</strong> its pricing page, for the team and collaboration features.</li>
+</ul>
+<p>Best for: developers who want an open, visual, cross platform workspace. Our older
+<a href="/blog/crystal-claude-code-alternative/">Crystal comparison</a> still explains the worktree model it grew from.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/best-claude-code-multi-agent-tools/note-2.png" alt="Hand-drawn sketch from “The Best Tools to Run Multiple Claude Code Agents (2026)”" loading="lazy" decoding="async" /></figure>
-<h2 id="vibe-kanban-the-task-board-for-agents" tabindex="-1">vibe-kanban — the task board for agents <a class="anchor" href="#vibe-kanban-the-task-board-for-agents" aria-hidden="true">#</a></h2>
-<p><a href="https://github.com/BloopAI/vibe-kanban">vibe-kanban</a> takes a different angle: a kanban board where
-you create tasks as cards, assign them to coding agents (it’s agent-agnostic — Claude Code, Gemini,
-Codex, and more), and move work through columns from to-do to review.</p>
+<h2 id="emdash-the-open-source-newcomer" tabindex="-1">Emdash: the open source newcomer <a class="anchor" href="#emdash-the-open-source-newcomer" aria-hidden="true">#</a></h2>
+<p><a href="https://emdash.sh">Emdash</a> is an open source agentic development environment backed by Y Combinator. It runs
+coding agents in parallel with any provider, schedules repeat work, previews apps in a built in browser, pulls
+issues from Linear, Jira and GitHub, and manages prompts, skills and MCP tools. It runs on macOS, Windows and
+Linux under Apache 2.0.</p>
+<ul>
+<li><strong>Strengths:</strong> wide agent support, issue tracker integrations, and every desktop platform.</li>
+<li><strong>Worth checking:</strong> remote development is its newest area, so test it on your own setup first.</li>
+</ul>
+<p>Best for: teams that live in an issue tracker and want agents fed straight from it.</p>
+<h2 id="vibe-kanban-the-board-winding-down" tabindex="-1">Vibe Kanban: the board, winding down <a class="anchor" href="#vibe-kanban-the-board-winding-down" aria-hidden="true">#</a></h2>
+<p><a href="https://github.com/BloopAI/vibe-kanban">Vibe Kanban</a> turns agent work into cards you plan, prompt and review.
+It runs with <code>npx vibe-kanban</code> and works with Claude Code, Codex and other agents. Its site now says the project
+is sunsetting and will continue as open source, maintained by the community.</p>
<ul>
-<li><strong>Strengths:</strong> the board is a familiar, legible mental model; it’s agent-agnostic; and it’s great
-for managing a <em>queue</em> of work and reviewing each agent’s output before it lands.</li>
-<li><strong>Where it stops:</strong> a board is something <em>you</em> drive. Cards don’t carry shared long-term memory
-between agents, agents don’t message each other, and nothing auto-routes work — the orchestration
-is you, moving cards.</li>
+<li><strong>Strengths:</strong> a familiar board for planning work and reviewing agent output.</li>
+<li><strong>Where it stops:</strong> with its company stepping back, future fixes depend on the community.</li>
</ul>
-<p>Best for: people who think in tasks and want a board over their agents. When you’d rather the routing
-happen for you, see <a href="/blog/vibe-kanban-alternative/">a vibe-kanban alternative</a>.</p>
-<h2 id="munder-difflin-the-coordinated-hive" tabindex="-1">Munder Difflin — the coordinated hive <a class="anchor" href="#munder-difflin-the-coordinated-hive" aria-hidden="true">#</a></h2>
-<p><a href="/#what">Munder Difflin</a> is our own take, and it’s deliberately in a different camp. It turns the
-Claude Code terminals you already run into a <em>self-coordinating hive</em>: each agent gets a role and a
-mailbox, agents message each other directly, they share long-term <strong>semantic memory</strong> (MemPalace),
-and a <a href="/#how">GOD orchestrator</a> you talk to in plain language decomposes your intent and routes work
-across the team. The whole floor is visualized as avatars at their desks, so you can actually watch
-it run.</p>
+<p>Best for: people already on it who are comfortable with a community maintained tool. See
+<a href="/blog/vibe-kanban-alternative/">a Vibe Kanban alternative</a>.</p>
+<h2 id="munder-difflin-the-coordinated-office" tabindex="-1">Munder Difflin: the coordinated office <a class="anchor" href="#munder-difflin-the-coordinated-office" aria-hidden="true">#</a></h2>
+<p><a href="https://munderdiffl.in/">Munder Difflin</a> is ours, and it lives in the fourth camp. It wraps twelve terminal
+coding CLIs (Claude Code, Codex, Gemini CLI, Antigravity, Grok, Kimi Code, Qwen, OpenCode, Crush, Pi, Copilot
+and Cursor) and runs them as real processes on your machine. Each agent gets a desk on an office floor, a
+mailbox and long term memory. Michael, your clone, turns what you ask for into tasks, hires workers and
+routes messages between them. Questions that need you land on an ASK ME board.</p>
<ul>
-<li><strong>Strengths:</strong> it adds the coordination the other tools leave to you — <a href="/blog/give-claude-code-long-term-memory/">shared
-memory</a>, inter-agent messaging, an orchestrator, and
-visibility. Local-first and MIT-licensed, on macOS, Windows, and Linux.</li>
-<li><strong>Where it stops (honestly):</strong> it’s a younger project, and the office-floor visualization is
-heavier than a TUI. For one or two quick parallel tasks, a lean session manager is less to think
-about — a hive earns its keep once coordination is the real cost.</li>
+<li><strong>Strengths:</strong> memory and messaging that outlast a session, one orchestrator for the whole floor, engines
+from different vendors on the same team, and it runs on the subscriptions you already pay for. Free and MIT
+licensed on macOS, Windows and Linux.</li>
+<li><strong>Where it stops, honestly:</strong> an office floor is more app than a terminal manager, and your machine has to
+stay on for the agents to keep working. Hosted sandboxes are not available yet. For one or two quick
+parallel tasks, a lighter tool is less to think about.</li>
</ul>
-<p>Best for: people running enough agents that <em>coordination</em> — not just parallelism — is the problem.</p>
-<h2 id="how-to-choose" tabindex="-1">How to choose <a class="anchor" href="#how-to-choose" aria-hidden="true">#</a></h2>
+<p>Best for: anyone running enough agents that coordination, not parallelism, is the real cost. The free app does
+all of the above. Optional paid plans exist too: Pro puts the office in one window and adds Stapler, and Teams
+lets your clone work with your teammates’ clones. Details are on the <a href="https://munderdiffl.in/#pricing">pricing page</a>.</p>
+<h2 id="how-do-you-choose" tabindex="-1">How do you choose? <a class="anchor" href="#how-do-you-choose" aria-hidden="true">#</a></h2>
<p>Match the tool to the bottleneck:</p>
<ul>
-<li><strong>“I just want parallel agents, fast.”</strong> → Claude Squad (terminal) or Crystal (GUI).</li>
-<li><strong>“I’m on a Mac and want it to feel native.”</strong> → Conductor.</li>
-<li><strong>“I want to manage work as tasks.”</strong> → vibe-kanban.</li>
-<li><strong>“My agents need to share what they learn and stop colliding.”</strong> → Munder Difflin.</li>
+<li><strong>“I only use Claude Code and need a team for one job.”</strong> Try agent teams first.</li>
+<li><strong>“I want parallel agents in the terminal, now.”</strong> Claude Squad.</li>
+<li><strong>“I am on a Mac, I want polish, and I will pay for cloud runs.”</strong> Conductor.</li>
+<li><strong>“I want an open, visual, cross platform workspace.”</strong> Nimbalyst or Emdash.</li>
+<li><strong>“My agents need to remember, talk to each other and stop colliding, across more than one vendor.”</strong>
+Munder Difflin.</li>
</ul>
-<p>If you want a structured way to weigh these, we wrote a <a href="/blog/how-to-choose-a-multi-agent-tool/">buyer’s
-checklist</a> and a criteria-based <a href="/blog/claude-code-orchestration-tools-compared/">orchestration tools
-comparison</a>. And if you’ve decided coordination is
-your real problem, the fastest way to feel the difference is to
-<a href="/#install">download Munder Difflin</a> — it’s free and open source.</p>
+<p>For a more structured pass, there is a <a href="/blog/how-to-choose-a-multi-agent-tool/">buyer’s checklist</a> and a
+criteria based <a href="/blog/claude-code-orchestration-tools-compared/">orchestration tools comparison</a>.</p>
<hr>
-<p>The honest summary: there’s no universal “best.” The best tool is the one that solves <em>your</em>
-bottleneck — and the bottleneck shifts from parallelism to coordination the moment you’re running a
-real team of agents.</p>
+<p>There is no universal best. The right tool is the one that removes your bottleneck, and that bottleneck moves
+from parallelism to coordination the day you start running a real team of agents.</p>
@@ -350,17 +364,27 @@ real team of agents.</p>
<div class="qa">
<p class="q">What is the best tool to run multiple Claude Code agents?</p>
- <p class="a">There&#39;s no single best — it depends on what you want. Claude Squad is the leanest terminal option; Conductor is the most polished on macOS; Crystal is a good open-source desktop GUI; vibe-kanban gives you a task board; and Munder Difflin adds shared memory, inter-agent messaging, and a GOD orchestrator so the agents act as one coordinated team.</p>
+ <p class="a">It depends on your bottleneck. Claude Code agent teams are the built in option for a single session. Claude Squad is the leanest terminal manager. Conductor is the polished Mac app, with cloud workspaces on paid plans. Nimbalyst and Emdash are open source desktop apps. Munder Difflin adds long term memory, messaging and an orchestrator across twelve CLIs.</p>
+ </div>
+
+ <div class="qa">
+ <p class="q">Does Claude Code have a built in way to run multiple agents?</p>
+ <p class="a">Yes, two. Subagents handle short helper jobs inside one session. Agent teams, which are experimental and stay off until you set CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1, let a lead session spawn teammates that share a task list and message each other.</p>
+ </div>
+
+ <div class="qa">
+ <p class="q">What happened to Crystal?</p>
+ <p class="a">Its maker deprecated Crystal in February 2026 and replaced it with Nimbalyst, an MIT licensed visual workspace for Claude Code and Codex.</p>
</div>
<div class="qa">
- <p class="q">Are these Claude Code multi-agent tools free?</p>
- <p class="a">Most are free and several are open source (Claude Squad, Crystal, vibe-kanban, and Munder Difflin are open source; Munder Difflin is MIT-licensed). Conductor is a free, native macOS app. Always check each project&#39;s current license and pricing before you commit.</p>
+ <p class="q">Is Vibe Kanban still maintained?</p>
+ <p class="a">Its website says Vibe Kanban is sunsetting and will continue as an open source, community maintained project.</p>
</div>
<div class="qa">
- <p class="q">Do I need a multi-agent tool to use Claude Code?</p>
- <p class="a">No. One Claude Code session handles most tasks. You start wanting a multi-agent tool once you&#39;re running three or more sessions at once and the coordination overhead — who&#39;s doing what, who knows what, who edits which file — starts costing you time.</p>
+ <p class="q">Are these tools free?</p>
+ <p class="a">Most have a free option. Claude Squad (AGPL 3.0), Nimbalyst (MIT), Emdash (Apache 2.0), Vibe Kanban (Apache 2.0) and the Munder Difflin app (MIT) are open source. Conductor has a free plan for local workspaces and paid plans for cloud and team features. Check pricing before you commit, because this space moves fast.</p>
</div>
</section>
@@ -379,7 +403,7 @@ real team of agents.</p>
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#the-four-shapes-of-multi-agent-claude-code">The four shapes of “multi-agent Claude Code”</a></li><li class="lvl-2"><a href="#the-tools-at-a-glance">The tools, at a glance</a></li><li class="lvl-2"><a href="#claude-squad-the-lean-terminal-option">Claude Squad — the lean terminal option</a></li><li class="lvl-2"><a href="#conductor-the-polished-macos-experience">Conductor — the polished macOS experience</a></li><li class="lvl-2"><a href="#crystal-the-open-source-desktop-gui">Crystal — the open-source desktop GUI</a></li><li class="lvl-2"><a href="#vibe-kanban-the-task-board-for-agents">vibe-kanban — the task board for agents</a></li><li class="lvl-2"><a href="#munder-difflin-the-coordinated-hive">Munder Difflin — the coordinated hive</a></li><li class="lvl-2"><a href="#how-to-choose">How to choose</a></li>
+ <li class="lvl-2"><a href="#what-kinds-of-tools-run-multiple-claude-code-agents">What kinds of tools run multiple Claude Code agents?</a></li><li class="lvl-2"><a href="#the-tools-at-a-glance">The tools at a glance</a></li><li class="lvl-2"><a href="#claude-code-agent-teams-the-built-in-option">Claude Code agent teams: the built in option</a></li><li class="lvl-2"><a href="#claude-squad-the-lean-terminal-manager">Claude Squad: the lean terminal manager</a></li><li class="lvl-2"><a href="#conductor-the-polished-mac-app-now-with-a-cloud">Conductor: the polished Mac app, now with a cloud</a></li><li class="lvl-2"><a href="#nimbalyst-what-crystal-became">Nimbalyst: what Crystal became</a></li><li class="lvl-2"><a href="#emdash-the-open-source-newcomer">Emdash: the open source newcomer</a></li><li class="lvl-2"><a href="#vibe-kanban-the-board-winding-down">Vibe Kanban: the board, winding down</a></li><li class="lvl-2"><a href="#munder-difflin-the-coordinated-office">Munder Difflin: the coordinated office</a></li><li class="lvl-2"><a href="#how-do-you-choose">How do you choose?</a></li>
</ol>
</aside>
@@ -398,13 +422,13 @@ real team of agents.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/building-reliable-ai-agents/index.html b/docs/blog/building-reliable-ai-agents/index.html
index f3604cc1..67122386 100644
--- a/docs/blog/building-reliable-ai-agents/index.html
+++ b/docs/blog/building-reliable-ai-agents/index.html
@@ -344,13 +344,13 @@ by design; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -376,10 +376,10 @@ by design; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/claude-code-automation-while-you-sleep/index.html b/docs/blog/claude-code-automation-while-you-sleep/index.html
index 7903aee7..098e241e 100644
--- a/docs/blog/claude-code-automation-while-you-sleep/index.html
+++ b/docs/blog/claude-code-automation-while-you-sleep/index.html
@@ -332,13 +332,13 @@ while you sleep; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -364,10 +364,10 @@ while you sleep; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/claude-code-multi-agent-setup-tutorial/index.html b/docs/blog/claude-code-multi-agent-setup-tutorial/index.html
index baf27a3a..0267ddf5 100644
--- a/docs/blog/claude-code-multi-agent-setup-tutorial/index.html
+++ b/docs/blog/claude-code-multi-agent-setup-tutorial/index.html
@@ -321,13 +321,13 @@ coordinated team of Claude Code agents — it’s free, local, and open source.<
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -353,10 +353,10 @@ coordinated team of Claude Code agents — it’s free, local, and open source.<
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/claude-code-orchestration-tools-compared/index.html b/docs/blog/claude-code-orchestration-tools-compared/index.html
index 6393b5fb..6c656b50 100644
--- a/docs/blog/claude-code-orchestration-tools-compared/index.html
+++ b/docs/blog/claude-code-orchestration-tools-compared/index.html
@@ -320,13 +320,13 @@ Linux.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/claude-code-subagents-vs-multi-agent-harness/index.html b/docs/blog/claude-code-subagents-vs-multi-agent-harness/index.html
index 03b1b855..b629706d 100644
--- a/docs/blog/claude-code-subagents-vs-multi-agent-harness/index.html
+++ b/docs/blog/claude-code-subagents-vs-multi-agent-harness/index.html
@@ -351,7 +351,7 @@ open source.</p>
</div>
<footer class="post-foot wrap"><div class="prevnext">
- <a class="pn prev" href="/blog/what-is-a-multi-agent-harness/"><span class="k">← Older</span><span class="t">What Is a Multi-Agent Harness? (Plain-English Guide)</span></a>
+ <a class="pn prev" href="/blog/what-is-a-multi-agent-harness/"><span class="k">← Older</span><span class="t">What Is a Multi-Agent Harness? A Plain English Guide</span></a>
<a class="pn next" href="/blog/claude-code-multi-agent-setup-tutorial/"><span class="k">Newer →</span><span class="t">Set Up a Claude Code Multi-Agent Workflow in 10 Minutes</span></a>
</div>
<section class="related">
@@ -363,13 +363,13 @@ open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -395,10 +395,10 @@ open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/claude-squad-alternative/index.html b/docs/blog/claude-squad-alternative/index.html
index c0ddfce1..075493ba 100644
--- a/docs/blog/claude-squad-alternative/index.html
+++ b/docs/blog/claude-squad-alternative/index.html
@@ -256,13 +256,13 @@ on macOS, Windows, and Linux.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/claude-squad-vs-munder-difflin/index.html b/docs/blog/claude-squad-vs-munder-difflin/index.html
index 0d403a08..8184a222 100644
--- a/docs/blog/claude-squad-vs-munder-difflin/index.html
+++ b/docs/blog/claude-squad-vs-munder-difflin/index.html
@@ -345,13 +345,13 @@ local-first.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/cline-vs-munder-difflin/index.html b/docs/blog/cline-vs-munder-difflin/index.html
index 01242945..4f2ee69d 100644
--- a/docs/blog/cline-vs-munder-difflin/index.html
+++ b/docs/blog/cline-vs-munder-difflin/index.html
@@ -336,13 +336,13 @@ licensed.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/command-center-guide/index.html b/docs/blog/command-center-guide/index.html
index 3e74c8e1..3d05fd5a 100644
--- a/docs/blog/command-center-guide/index.html
+++ b/docs/blog/command-center-guide/index.html
@@ -3,8 +3,8 @@
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
-<title>The Command Center: Kanban, Fleet, and Budgets in One Place — Munder Difflin Blog</title>
-<meta name="description" content="A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor." />
+<title>The Command Center: Kanban, Fleet and Budgets in One Place — Munder Difflin Blog</title>
+<meta name="description" content="A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor." />
<link rel="canonical" href="https://munderdiffl.in/blog/command-center-guide/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -14,8 +14,8 @@
<!-- Open Graph -->
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
-<meta property="og:title" content="The Command Center: Kanban, Fleet, and Budgets in One Place — Munder Difflin Blog" />
-<meta property="og:description" content="A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor." />
+<meta property="og:title" content="The Command Center: Kanban, Fleet and Budgets in One Place — Munder Difflin Blog" />
+<meta property="og:description" content="A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor." />
<meta property="og:url" content="https://munderdiffl.in/blog/command-center-guide/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -30,8 +30,8 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
-<meta name="twitter:title" content="The Command Center: Kanban, Fleet, and Budgets in One Place — Munder Difflin Blog" />
-<meta name="twitter:description" content="A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor." />
+<meta name="twitter:title" content="The Command Center: Kanban, Fleet and Budgets in One Place — Munder Difflin Blog" />
+<meta name="twitter:description" content="A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -44,11 +44,11 @@
{
"@context": "https://schema.org",
"@type": "BlogPosting",
- "headline": "The Command Center: Kanban, Fleet, and Budgets in One Place",
- "description": "A guide to Munder Difflin's Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.",
+ "headline": "The Command Center: Kanban, Fleet and Budgets in One Place",
+ "description": "A guide to Munder Difflin's Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent's budget, and when to watch the board instead of the floor.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-07-03T00:00:00.000Z",
- "dateModified": "2026-07-03T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -69,7 +69,7 @@
"itemListElement": [
{ "@type": "ListItem", "position": 1, "name": "Home", "item": "https://munderdiffl.in/" },
{ "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://munderdiffl.in/blog/" },
- { "@type": "ListItem", "position": 3, "name": "The Command Center: Kanban, Fleet, and Budgets in One Place", "item": "https://munderdiffl.in/blog/command-center-guide/" }
+ { "@type": "ListItem", "position": 3, "name": "The Command Center: Kanban, Fleet and Budgets in One Place", "item": "https://munderdiffl.in/blog/command-center-guide/" }
]
}
</script>
@@ -79,12 +79,12 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "What is the Command Center in Munder Difflin?", "acceptedAnswer": { "@type": "Answer", "text": "It's Michael's control surface — the management view of the whole hive. It has tabs for the orchestrator Terminal, the Floor (roster, dispatch, per-agent model selector, live fleet monitoring), Memory (semantic search plus a memory graph), Activity (event log, board, real token telemetry, observability, and a CI watcher), Tasks (a dependency-aware kanban board), and Triggers (everything that starts work without you — schedules with last/next-fired times, context, webhooks, and organization)." } },
- { "@type": "Question", "name": "How does the task kanban work?", "acceptedAnswer": { "@type": "Answer", "text": "The Tasks tab is a dependency-aware kanban board. You assign tasks to agents and track them across todo, doing, blocked, and done. Because tasks can declare dependencies, downstream work waits until its prerequisites finish, so a chain of related tasks starts in the right order without you babysitting the handoffs." } },
- { "@type": "Question", "name": "How do per-agent budgets and cost tracking work?", "acceptedAnswer": { "@type": "Answer", "text": "You set a token budget per agent, and live fleet monitoring tracks consumption across the roster. Cost numbers are real, not estimates from vibes: the Activity tab reads the JSONL transcripts Claude Code writes and surfaces actual token counts plus estimated USD cost per agent per session, backed by a durable cost ledger that survives restarts. The budget pairs with a circuit breaker that steers, then constrains, then stops an agent that loops or blows its cap." } },
- { "@type": "Question", "name": "When should I watch the office floor versus the Command Center?", "acceptedAnswer": { "@type": "Answer", "text": "The floor is for ambient awareness and single-agent moments — seeing who's working, watching envelopes fly, dropping into one terminal. The Command Center is for management questions: what's queued and blocked, what each agent has spent, whether CI is green, and what fires next on the schedule. Roughly: floor for one agent right now, board for the whole fleet over time." } },
- { "@type": "Question", "name": "Does the Command Center include observability?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. The Activity tab includes a live OpenTelemetry collector with per-model cost attribution, a fleet grid, and a per-agent tool-span waterfall, so you can see exactly what every agent is doing and what it costs in real time. Each agent card also carries a context-window gauge showing how much of the model's context that agent has consumed." } },
- { "@type": "Question", "name": "Can the Command Center pull in work from GitHub?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. You can ingest open issues from any registered repo via the gh CLI and assign them to agents with one click, and a CI status watcher shows live pass/fail/in-progress state for GitHub Actions runs in the Activity tab. Work flows in from GitHub, gets tracked on the kanban, and its CI result lands back in the same view." } }
+ { "@type": "Question", "name": "What is the Command Center in Munder Difflin?", "acceptedAnswer": { "@type": "Answer", "text": "It is the management view of your office, next to the floor. Its tabs are terminal, monitor, tasks, ask me, triggers, history, memory, graph, activity, skills and workers. The floor shows you who is busy. The Command Center shows what is queued, what is blocked, what it costs and what runs next." } },
+ { "@type": "Question", "name": "How does the task kanban work?", "acceptedAnswer": { "@type": "Answer", "text": "Tasks move across todo, doing, blocked and done, and each one is assigned to an agent. A task can depend on other tasks, so downstream work waits for its prerequisites and you stop being the one who sequences handoffs." } },
+ { "@type": "Question", "name": "How do budgets and cost tracking work?", "acceptedAnswer": { "@type": "Answer", "text": "Each agent can have a token budget, and a circuit breaker steers, then constrains, then stops an agent that loops or runs past its cap. The activity tab shows live token use from each agent's telemetry against its own limit, or against the floor budget when an agent has none." } },
+ { "@type": "Question", "name": "Where do questions for me show up?", "acceptedAnswer": { "@type": "Answer", "text": "On the ask me tab. When the team blocks a task on your input, whether a question to answer or a to do only you can do, it lands there and on the ASK ME board on the floor. Answer it and the work carries on, or dismiss it and the history is kept." } },
+ { "@type": "Question", "name": "When should I watch the floor instead of the Command Center?", "acceptedAnswer": { "@type": "Answer", "text": "Watch the floor for one agent right now: you just sent work, you want to type into a session, or you want ambient awareness. Watch the Command Center for the whole fleet over time: what is queued or blocked, who is near budget, and what runs tonight." } },
+ { "@type": "Question", "name": "Can the Command Center take work from GitHub?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. Open issues from a registered GitHub repo can come onto the board as tasks and be assigned to agents." } }
]
}
@@ -153,15 +153,15 @@
<span>/</span>
<a href="/blog/topics/guides/">Guides</a>
</nav>
- <h1>The Command Center: Kanban, Fleet, and Budgets in One Place</h1>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h1>The Command Center: Kanban, Fleet and Budgets in One Place</h1>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true">·</span>
- <span>5 min read</span>
+ <span>4 min read</span>
<span class="tag kind">Guides</span>
</div>
</header>
@@ -170,48 +170,78 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p>The office floor is fun to watch, but you don't manage a team by staring at desks. Munder Difflin's <strong>Command Center</strong> is the management view: a <strong>dependency-aware task kanban</strong>, <strong>live fleet monitoring with per-agent token budgets</strong>, <strong>real cost telemetry</strong> backed by a durable ledger, <strong>OpenTelemetry observability</strong> with a per-agent tool-span waterfall, plus GitHub issue ingestion, a CI watcher, and a Triggers tab. Rule of thumb: <strong>watch the floor for one agent right now, watch the board for the whole fleet over time</strong>.</p></div>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p>The floor is fun to watch, but you do not manage a team
+by staring at desks. The <strong>Command Center</strong> is the management view: a <strong>kanban with task
+dependencies</strong>, an <strong>ask me</strong> tab for decisions only you can make, <strong>triggers</strong> for work
+that starts without you, <strong>memory</strong> and its <strong>graph</strong>, <strong>activity</strong> with live token use
+against each agent's budget, and a <strong>skills</strong> catalog. Rule of thumb: <strong>the floor for one agent right now,
+the Command Center for the whole fleet over time</strong>.</p></div>
<video controls preload="none" playsinline poster="/media/demo/features-poster.jpg" style="width:100%; border-radius:12px; margin:12px 0 24px;">
<source src="/media/demo/features.mp4" type="video/mp4" />
</video>
-<p>Munder Difflin has two ways of looking at the same hive. The <strong>office floor</strong> is the ambient view: avatars at desks, envelopes flying between them, speech bubbles when a tool fires. The <strong>Command Center</strong> is the management view — Michael’s control surface, the place you go when the question stops being “what is Dwight doing?” and becomes “what is <em>everyone</em> doing, what has it cost, and what’s next?”</p>
-<p>This guide walks through what’s in it and how the pieces fit together.</p>
-<h2 id="whats-in-the-command-center" tabindex="-1">What’s in the Command Center <a class="anchor" href="#whats-in-the-command-center" aria-hidden="true">#</a></h2>
-<p>Six tabs, one control surface:</p>
+<p>Munder Difflin gives you two views of the same office. The <strong>floor</strong> is the ambient one: characters at desks and envelopes
+flying between them. The <strong>Command Center</strong> is the management one, where you go when the question changes from “what is Dwight
+doing?” to “what is everyone doing, what has it cost, and what happens next?” This guide walks through it as of 0.5.2.</p>
+<h2 id="what-is-in-the-command-center" tabindex="-1">What is in the Command Center? <a class="anchor" href="#what-is-in-the-command-center" aria-hidden="true">#</a></h2>
+<p>Eleven tabs on one control surface: terminal, monitor, tasks, ask me, triggers, history, memory, graph, activity, skills and
+workers. The ones you will use most:</p>
<ul>
-<li><strong>Terminal</strong> — Michael’s own terminal, the <a href="/blog/how-the-god-orchestrator-works/">GOD orchestrator</a> you talk to directly.</li>
-<li><strong>Floor</strong> — the roster and dispatch controls: hire, dispatch work, pick a model per agent, and watch live fleet monitoring across everyone at once.</li>
-<li><strong>Memory</strong> — semantic search over the shared memory palace, plain text search, and a memory graph.</li>
-<li><strong>Activity</strong> — the append-only event log, the blackboard, real token telemetry, the observability view, and a CI watcher.</li>
-<li><strong>Tasks</strong> — the kanban board.</li>
-<li><strong>Triggers</strong> — everything that starts work without you, grouped by kind: schedules (recurring missions and the adaptive heartbeat, with last/next-fired times), context, webhooks, and organization.</li>
+<li><strong>terminal:</strong> Michael’s own terminal, where you talk to your clone.</li>
+<li><strong>monitor:</strong> the roster at a glance, with a context gauge on every agent.</li>
+<li><strong>tasks:</strong> the kanban board.</li>
+<li><strong>ask me:</strong> everything that is waiting on your answer.</li>
+<li><strong>triggers:</strong> everything that starts work without you.</li>
+<li><strong>memory</strong> and <strong>graph:</strong> search the office’s shared memory, and see how it connects.</li>
+<li><strong>activity:</strong> live token use and each agent’s tool calls.</li>
+<li><strong>skills:</strong> a catalog of skills to install for your agents.</li>
</ul>
-<p>You could run the hive from the floor alone. But once you have more than two or three agents, the Command Center is where the actual decisions happen.</p>
-<h2 id="the-task-kanban-work-as-a-board-not-a-chat-log" tabindex="-1">The task kanban: work as a board, not a chat log <a class="anchor" href="#the-task-kanban-work-as-a-board-not-a-chat-log" aria-hidden="true">#</a></h2>
-<p>The Tasks tab is a <strong>dependency-aware kanban board</strong>. Tasks move across four columns — todo, doing, blocked, done — and each task is assigned to a specific agent.</p>
-<p>The load-bearing word is <em>dependency-aware</em>. Real work has ordering: the refactor has to land before the tests get updated; the schema migration comes before the endpoint. On the board you wire those dependencies explicitly, and downstream tasks wait until their prerequisites finish. You’re not the sequencer anymore — the board is.</p>
-<p>Work gets onto the board a few ways: you create tasks yourself, Michael creates and assigns them as he routes requests, and you can <strong>pull open GitHub issues from any registered repo</strong> (via the <code>gh</code> CLI) and assign them to agents with one click. Issue in, task tracked, agent dispatched — and when the work ships, the <strong>CI status watcher</strong> in the Activity tab shows live pass/fail/in-progress for the repo’s GitHub Actions runs. The loop closes in the same window it opened in.</p>
+<p>You could run a small office from the floor alone. Past two or three agents, the Command Center is where the real decisions
+happen.</p>
+<h2 id="how-does-the-task-kanban-work" tabindex="-1">How does the task kanban work? <a class="anchor" href="#how-does-the-task-kanban-work" aria-hidden="true">#</a></h2>
+<p>It turns work into a board instead of a chat log. Tasks move across four columns, todo, doing, blocked and done, and every task
+is assigned to an agent.</p>
+<p>The part that matters is dependencies. A task can depend on other tasks, and it waits until they are done. The refactor lands
+before the tests are updated. The migration runs before the endpoint. You stop being the sequencer, because the board does it.</p>
+<p>Work reaches the board three ways: you add it, Michael creates and assigns it while he routes your requests, or it comes in from
+an open issue on a registered GitHub repo.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/command-center-guide/note-1.png" alt="Hand-drawn sketch from “The Command Center: Kanban, Fleet, and Budgets in One Place”" loading="lazy" decoding="async" /></figure>
-<h2 id="fleet-status-the-roster-at-a-glance" tabindex="-1">Fleet status: the roster at a glance <a class="anchor" href="#fleet-status-the-roster-at-a-glance" aria-hidden="true">#</a></h2>
-<p>The Floor tab is where the fleet stops being a set of individual terminals and becomes a roster. Each agent card shows what that agent is up to, and the card’s progress bar doubles as a <strong>context-window gauge</strong> — a glanceable read on how much of the model’s context each agent has burned. An agent near the top of its gauge is an agent about to compact or slow down; you can see it coming instead of discovering it.</p>
-<p>Dispatch also lives here: pick an agent, pick a model (the <strong>per-agent model selector</strong> means your cheap-tier worker and your frontier-tier reviewer are one dropdown apart), and send work. Routine tasks go to the cheap tier; the frontier tier is reserved for the reasoning that needs it.</p>
-<h2 id="budgets-and-cost-real-numbers-not-vibes" tabindex="-1">Budgets and cost: real numbers, not vibes <a class="anchor" href="#budgets-and-cost-real-numbers-not-vibes" aria-hidden="true">#</a></h2>
-<p>This is the part that makes a 24/7 fleet safe to leave running.</p>
-<p><strong>Per-agent token budgets.</strong> You set a budget per agent, and live fleet monitoring tracks consumption across the whole roster against it. Budgets pair with the <strong>circuit breaker</strong>, which handles the runaway case on a ladder: <em>steer</em> (nudge the agent), then <em>constrain</em>, then <em>stop</em>. An agent that loops, storms errors, or blows through its cap gets caught by machinery, not by you noticing the bill.</p>
-<p><strong>Real telemetry.</strong> The Activity tab doesn’t estimate from message counts — it reads the JSONL transcripts Claude Code writes to <code>~/.claude/projects/</code> and surfaces <strong>actual token counts and estimated USD cost per agent, per session</strong>, backed by a <strong>durable cost ledger</strong> that survives app restarts. Yesterday’s spend is still there this morning, attributed to the agent that spent it.</p>
-<p><strong>Observability.</strong> Also in Activity: a live <strong>OpenTelemetry collector</strong> with per-model cost attribution, a fleet grid, and a <strong>per-agent tool-span waterfall</strong> — the “what exactly is this agent doing, step by step, and what does each step cost” view. It’s the same discipline we covered in <a href="/blog/observability-for-agent-fleets/">observability for agent fleets</a>, built into the harness instead of bolted on.</p>
+<h2 id="what-does-the-ask-me-tab-do" tabindex="-1">What does the ask me tab do? <a class="anchor" href="#what-does-the-ask-me-tab-do" aria-hidden="true">#</a></h2>
+<p>It collects every decision that needs you. When the team blocks a task on your input, whether that is a question to answer or a
+to do only you can do, it shows up here and on the ASK ME board on the floor. Questions render as formatted text, so a list of
+options reads like a list. Answer it and the work carries on. Dismiss it and the history is kept.</p>
+<p>This is what keeps a busy office from hiding its questions in five different scrollbacks.</p>
+<h2 id="how-do-budgets-and-cost-tracking-work" tabindex="-1">How do budgets and cost tracking work? <a class="anchor" href="#how-do-budgets-and-cost-tracking-work" aria-hidden="true">#</a></h2>
+<p>This is the part that makes a floor safe to leave running.</p>
+<ul>
+<li><strong>Per agent token budgets.</strong> Set a budget for each agent in Settings, under Autonomy &amp; Budgets. The activity tab shows live
+token use from each agent’s telemetry as bars against its limit, or against the floor budget when an agent has no limit of its own.</li>
+<li><strong>A circuit breaker.</strong> An agent that loops or blows through its cap is steered, then constrained, then stopped, by the app
+rather than by you noticing a bill.</li>
+<li><strong>A tool waterfall.</strong> Activity also lays out each agent’s tool calls, so you can see exactly what it did, step by step.</li>
+<li><strong>A context gauge.</strong> On the monitor, every agent shows how much of its context window it has used, so you see a slowdown coming
+before it arrives.</li>
+</ul>
<figure class="fig fig-inline"><img src="/blog/assets/media/command-center-guide/note-2.png" alt="Hand-drawn sketch from “The Command Center: Kanban, Fleet, and Budgets in One Place”" loading="lazy" decoding="async" /></figure>
-<h2 id="triggers-the-boards-fourth-dimension" tabindex="-1">Triggers: the board’s fourth dimension <a class="anchor" href="#triggers-the-boards-fourth-dimension" aria-hidden="true">#</a></h2>
-<p>The kanban answers “what’s in flight.” The <strong>Triggers tab</strong> answers “what happens next without me” — every way the hive wakes up when you aren’t typing. <strong>Schedules</strong> are the first and oldest of them: recurring missions carry a label, an interval, a target agent, and a body, and the tab shows last-fired and next-fired times, so drift is visible. A scheduler <strong>heartbeat</strong> re-engages the floor when it goes quiet. Together with the board, it turns the hive from something you drive into something you supervise — the full pattern is in <a href="/blog/scheduling-autonomous-agent-missions/">scheduling autonomous agent missions</a>.</p>
-<h2 id="floor-or-board-a-simple-rule" tabindex="-1">Floor or board? A simple rule <a class="anchor" href="#floor-or-board-a-simple-rule" aria-hidden="true">#</a></h2>
-<p>Both views are the same hive, so this is about attention, not data:</p>
+<h2 id="what-starts-work-without-you" tabindex="-1">What starts work without you? <a class="anchor" href="#what-starts-work-without-you" aria-hidden="true">#</a></h2>
+<p>The triggers tab. <strong>Schedules</strong> run a prompt on an interval or on chosen weekdays at a set time, and the prompt is sent word for
+word on every run. <strong>Context</strong> rules decide what happens as an agent’s memory fills up. <strong>Webhooks</strong> let outside systems post work
+in, and <strong>organisation</strong> lets a teammate’s office send work to yours. Together with the board, triggers turn the office from
+something you drive into something you supervise. The full pattern is in
+<a href="/blog/scheduling-autonomous-agent-missions/">scheduling autonomous agent missions</a>.</p>
+<h2 id="should-you-watch-the-floor-or-the-command-center" tabindex="-1">Should you watch the floor or the Command Center? <a class="anchor" href="#should-you-watch-the-floor-or-the-command-center" aria-hidden="true">#</a></h2>
+<p>Both show the same office, so this is about attention, not data:</p>
<ul>
-<li><strong>Watch the floor</strong> when you care about <em>one agent right now</em> — you just dispatched something, you want to type into a session, or you want ambient awareness while doing other work. (Also: it’s charming.)</li>
-<li><strong>Watch the board</strong> when you care about <em>the fleet over time</em> — what’s queued, what’s blocked on what, who’s near budget, whether CI is green, what fires tonight.</li>
+<li><strong>Watch the floor</strong> for one agent right now: you just sent work, you want to type into a session, or you want ambient
+awareness while you do something else. It is also, admittedly, charming.</li>
+<li><strong>Watch the Command Center</strong> for the whole fleet over time: what is queued, what is blocked on what, who is close to budget, and
+what runs tonight.</li>
</ul>
-<p>In practice: floor open while you work, Command Center when you check in. Morning check-in is the board — done column, spend per agent, blocked tasks, next scheduled missions. That’s a two-minute read that would take twenty minutes of terminal-scrolling without it. And when something needs a human decision, escalations land in the <a href="/blog/human-in-the-loop-approving-ai-agents/">approvals queue</a> rather than hiding in a scrollback buffer.</p>
+<p>In practice, keep the floor open while you work and check the Command Center when you check in. A morning check in takes two
+minutes: the done column, the ask me tab, spend per agent, and what is scheduled next. When something needs a human decision, it is
+already waiting for you on <a href="/blog/human-in-the-loop-approving-ai-agents/">the ask me tab</a> rather than buried in a scrollback.</p>
<h2 id="try-it" tabindex="-1">Try it <a class="anchor" href="#try-it" aria-hidden="true">#</a></h2>
-<p>The Command Center ships in the current release, free and open source. <a href="https://github.com/chaitanyagiri/munder-difflin/releases/latest">Download Munder Difflin</a> — and if the board earns its place in your morning routine, a <a href="https://github.com/chaitanyagiri/munder-difflin">star on GitHub</a> helps more people find it.</p>
+<p>The Command Center is part of the free classic office. <a href="https://munderdiffl.in/">Download Munder Difflin</a>, and if the board earns a
+place in your morning routine, <a href="https://github.com/chaitanyagiri/munder-difflin">a star on GitHub</a> helps other people find it.</p>
@@ -220,32 +250,32 @@
<div class="qa">
<p class="q">What is the Command Center in Munder Difflin?</p>
- <p class="a">It&#39;s Michael&#39;s control surface — the management view of the whole hive. It has tabs for the orchestrator Terminal, the Floor (roster, dispatch, per-agent model selector, live fleet monitoring), Memory (semantic search plus a memory graph), Activity (event log, board, real token telemetry, observability, and a CI watcher), Tasks (a dependency-aware kanban board), and Triggers (everything that starts work without you — schedules with last/next-fired times, context, webhooks, and organization).</p>
+ <p class="a">It is the management view of your office, next to the floor. Its tabs are terminal, monitor, tasks, ask me, triggers, history, memory, graph, activity, skills and workers. The floor shows you who is busy. The Command Center shows what is queued, what is blocked, what it costs and what runs next.</p>
</div>
<div class="qa">
<p class="q">How does the task kanban work?</p>
- <p class="a">The Tasks tab is a dependency-aware kanban board. You assign tasks to agents and track them across todo, doing, blocked, and done. Because tasks can declare dependencies, downstream work waits until its prerequisites finish, so a chain of related tasks starts in the right order without you babysitting the handoffs.</p>
+ <p class="a">Tasks move across todo, doing, blocked and done, and each one is assigned to an agent. A task can depend on other tasks, so downstream work waits for its prerequisites and you stop being the one who sequences handoffs.</p>
</div>
<div class="qa">
- <p class="q">How do per-agent budgets and cost tracking work?</p>
- <p class="a">You set a token budget per agent, and live fleet monitoring tracks consumption across the roster. Cost numbers are real, not estimates from vibes: the Activity tab reads the JSONL transcripts Claude Code writes and surfaces actual token counts plus estimated USD cost per agent per session, backed by a durable cost ledger that survives restarts. The budget pairs with a circuit breaker that steers, then constrains, then stops an agent that loops or blows its cap.</p>
+ <p class="q">How do budgets and cost tracking work?</p>
+ <p class="a">Each agent can have a token budget, and a circuit breaker steers, then constrains, then stops an agent that loops or runs past its cap. The activity tab shows live token use from each agent&#39;s telemetry against its own limit, or against the floor budget when an agent has none.</p>
</div>
<div class="qa">
- <p class="q">When should I watch the office floor versus the Command Center?</p>
- <p class="a">The floor is for ambient awareness and single-agent moments — seeing who&#39;s working, watching envelopes fly, dropping into one terminal. The Command Center is for management questions: what&#39;s queued and blocked, what each agent has spent, whether CI is green, and what fires next on the schedule. Roughly: floor for one agent right now, board for the whole fleet over time.</p>
+ <p class="q">Where do questions for me show up?</p>
+ <p class="a">On the ask me tab. When the team blocks a task on your input, whether a question to answer or a to do only you can do, it lands there and on the ASK ME board on the floor. Answer it and the work carries on, or dismiss it and the history is kept.</p>
</div>
<div class="qa">
- <p class="q">Does the Command Center include observability?</p>
- <p class="a">Yes. The Activity tab includes a live OpenTelemetry collector with per-model cost attribution, a fleet grid, and a per-agent tool-span waterfall, so you can see exactly what every agent is doing and what it costs in real time. Each agent card also carries a context-window gauge showing how much of the model&#39;s context that agent has consumed.</p>
+ <p class="q">When should I watch the floor instead of the Command Center?</p>
+ <p class="a">Watch the floor for one agent right now: you just sent work, you want to type into a session, or you want ambient awareness. Watch the Command Center for the whole fleet over time: what is queued or blocked, who is near budget, and what runs tonight.</p>
</div>
<div class="qa">
- <p class="q">Can the Command Center pull in work from GitHub?</p>
- <p class="a">Yes. You can ingest open issues from any registered repo via the gh CLI and assign them to agents with one click, and a CI status watcher shows live pass/fail/in-progress state for GitHub Actions runs in the Activity tab. Work flows in from GitHub, gets tracked on the kanban, and its CI result lands back in the same view.</p>
+ <p class="q">Can the Command Center take work from GitHub?</p>
+ <p class="a">Yes. Open issues from a registered GitHub repo can come onto the board as tasks and be assigned to agents.</p>
</div>
</section>
@@ -264,7 +294,7 @@
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#whats-in-the-command-center">What’s in the Command Center</a></li><li class="lvl-2"><a href="#the-task-kanban-work-as-a-board-not-a-chat-log">The task kanban: work as a board, not a chat log</a></li><li class="lvl-2"><a href="#fleet-status-the-roster-at-a-glance">Fleet status: the roster at a glance</a></li><li class="lvl-2"><a href="#budgets-and-cost-real-numbers-not-vibes">Budgets and cost: real numbers, not vibes</a></li><li class="lvl-2"><a href="#triggers-the-boards-fourth-dimension">Triggers: the board’s fourth dimension</a></li><li class="lvl-2"><a href="#floor-or-board-a-simple-rule">Floor or board? A simple rule</a></li><li class="lvl-2"><a href="#try-it">Try it</a></li>
+ <li class="lvl-2"><a href="#what-is-in-the-command-center">What is in the Command Center?</a></li><li class="lvl-2"><a href="#how-does-the-task-kanban-work">How does the task kanban work?</a></li><li class="lvl-2"><a href="#what-does-the-ask-me-tab-do">What does the ask me tab do?</a></li><li class="lvl-2"><a href="#how-do-budgets-and-cost-tracking-work">How do budgets and cost tracking work?</a></li><li class="lvl-2"><a href="#what-starts-work-without-you">What starts work without you?</a></li><li class="lvl-2"><a href="#should-you-watch-the-floor-or-the-command-center">Should you watch the floor or the Command Center?</a></li><li class="lvl-2"><a href="#try-it">Try it</a></li>
</ol>
</aside>
@@ -272,7 +302,7 @@
<footer class="post-foot wrap"><div class="prevnext">
<a class="pn prev" href="/blog/crewai-autogen-vs-a-local-agent-harness/"><span class="k">← Older</span><span class="t">CrewAI and AutoGen vs a Local Agent Harness: Framework or App?</span></a>
- <a class="pn next" href="/blog/qm-vs-munder-difflin/"><span class="k">Newer →</span><span class="t">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</span></a>
+ <a class="pn next" href="/blog/qm-vs-munder-difflin/"><span class="k">Newer →</span><span class="t">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</span></a>
</div>
<section class="related">
<h2>Related in Guides</h2>
@@ -299,10 +329,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/conductor-claude-code-alternative/index.html b/docs/blog/conductor-claude-code-alternative/index.html
index 564bf37c..96a1e05d 100644
--- a/docs/blog/conductor-claude-code-alternative/index.html
+++ b/docs/blog/conductor-claude-code-alternative/index.html
@@ -257,13 +257,13 @@ coordinated team locally — it’s free and open source.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/context-engineering-for-ai-agents/index.html b/docs/blog/context-engineering-for-ai-agents/index.html
index d93b4a93..038c97b4 100644
--- a/docs/blog/context-engineering-for-ai-agents/index.html
+++ b/docs/blog/context-engineering-for-ai-agents/index.html
@@ -349,13 +349,13 @@ it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -381,10 +381,10 @@ it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/crewai-autogen-vs-a-local-agent-harness/index.html b/docs/blog/crewai-autogen-vs-a-local-agent-harness/index.html
index 0b277dda..3bf04101 100644
--- a/docs/blog/crewai-autogen-vs-a-local-agent-harness/index.html
+++ b/docs/blog/crewai-autogen-vs-a-local-agent-harness/index.html
@@ -271,7 +271,7 @@
<footer class="post-foot wrap"><div class="prevnext">
<a class="pn prev" href="/blog/github-copilot-cli-vs-claude-code-for-hive-work/"><span class="k">← Older</span><span class="t">GitHub Copilot CLI vs Claude Code as Hive Workers</span></a>
- <a class="pn next" href="/blog/command-center-guide/"><span class="k">Newer →</span><span class="t">The Command Center: Kanban, Fleet, and Budgets in One Place</span></a>
+ <a class="pn next" href="/blog/command-center-guide/"><span class="k">Newer →</span><span class="t">The Command Center: Kanban, Fleet and Budgets in One Place</span></a>
</div>
<section class="related">
<h2>Related in Comparisons</h2>
@@ -282,13 +282,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/crystal-claude-code-alternative/index.html b/docs/blog/crystal-claude-code-alternative/index.html
index ca3c6d2d..6bbcadd3 100644
--- a/docs/blog/crystal-claude-code-alternative/index.html
+++ b/docs/blog/crystal-claude-code-alternative/index.html
@@ -254,13 +254,13 @@ actually remember — free and open source.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/debugging-multi-agent-systems/index.html b/docs/blog/debugging-multi-agent-systems/index.html
index 34784601..f1297994 100644
--- a/docs/blog/debugging-multi-agent-systems/index.html
+++ b/docs/blog/debugging-multi-agent-systems/index.html
@@ -312,13 +312,13 @@ to run agents you can actually trace; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -344,10 +344,10 @@ to run agents you can actually trace; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/deploy-a-blog-writer-agent/index.html b/docs/blog/deploy-a-blog-writer-agent/index.html
index dc020b80..854951ea 100644
--- a/docs/blog/deploy-a-blog-writer-agent/index.html
+++ b/docs/blog/deploy-a-blog-writer-agent/index.html
@@ -3,8 +3,8 @@
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
-<title>Deploy a Blog-Writer Agent: The One That Wrote This Post — Munder Difflin Blog</title>
-<meta name="description" content="How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own." />
+<title>Deploy a Blog Writer Agent: The One That Wrote This Post — Munder Difflin Blog</title>
+<meta name="description" content="How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own." />
<link rel="canonical" href="https://munderdiffl.in/blog/deploy-a-blog-writer-agent/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -14,8 +14,8 @@
<!-- Open Graph -->
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
-<meta property="og:title" content="Deploy a Blog-Writer Agent: The One That Wrote This Post — Munder Difflin Blog" />
-<meta property="og:description" content="How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own." />
+<meta property="og:title" content="Deploy a Blog Writer Agent: The One That Wrote This Post — Munder Difflin Blog" />
+<meta property="og:description" content="How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own." />
<meta property="og:url" content="https://munderdiffl.in/blog/deploy-a-blog-writer-agent/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -30,8 +30,8 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
-<meta name="twitter:title" content="Deploy a Blog-Writer Agent: The One That Wrote This Post — Munder Difflin Blog" />
-<meta name="twitter:description" content="How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own." />
+<meta name="twitter:title" content="Deploy a Blog Writer Agent: The One That Wrote This Post — Munder Difflin Blog" />
+<meta name="twitter:description" content="How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -44,11 +44,11 @@
{
"@context": "https://schema.org",
"@type": "BlogPosting",
- "headline": "Deploy a Blog-Writer Agent: The One That Wrote This Post",
- "description": "How Munder Difflin's blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.",
+ "headline": "Deploy a Blog Writer Agent: The One That Wrote This Post",
+ "description": "How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-06-10T00:00:00.000Z",
- "dateModified": "2026-08-20T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -56,7 +56,7 @@
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/deploy-a-blog-writer-agent/" },
- "keywords": "automated blog writer agent, ai blog automation, content agent, multi-agent blogging, automated content pipeline",
+ "keywords": "automated blog writer agent, ai blog automation, content agent, multi-agent blogging, automated content pipeline, ai blog writing workflow",
"articleSection": "Use Cases",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/deploy-a-blog-writer-agent/"
@@ -69,7 +69,7 @@
"itemListElement": [
{ "@type": "ListItem", "position": 1, "name": "Home", "item": "https://munderdiffl.in/" },
{ "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://munderdiffl.in/blog/" },
- { "@type": "ListItem", "position": 3, "name": "Deploy a Blog-Writer Agent: The One That Wrote This Post", "item": "https://munderdiffl.in/blog/deploy-a-blog-writer-agent/" }
+ { "@type": "ListItem", "position": 3, "name": "Deploy a Blog Writer Agent: The One That Wrote This Post", "item": "https://munderdiffl.in/blog/deploy-a-blog-writer-agent/" }
]
}
</script>
@@ -79,10 +79,11 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "Did an AI actually write the Munder Difflin blog?", "acceptedAnswer": { "@type": "Answer", "text": "Yes — most of it. A writer agent in the hive drafts each post from a topic brief and a house-style reference, Michael integrates it as the single committer, Eleventy builds the markdown into the static site, and a human approves the final deploy to munderdiffl.in. This very post is an example of that pipeline running — and as of August 2026, the same office also draws every post's illustrations as code." } },
- { "@type": "Question", "name": "Is the blog-writer agent fully autonomous?", "acceptedAnswer": { "@type": "Answer", "text": "Almost — it's deliberately human-gated at one point: publish. The agent drafts and self-reviews against a style reference in an isolated worktree; the orchestrator integrates and builds; a person reviews the diff and approves the deploy. That keeps the volume high and hands-off without putting an unreviewed post on the live domain." } },
- { "@type": "Question", "name": "How do I build my own blog-writer agent?", "acceptedAnswer": { "@type": "Answer", "text": "Give one agent in the hive three things: a topic or brief, a house-style reference (a few of your best existing posts), and write access to an isolated worktree. Let it draft, have a reviewer agent or the orchestrator check it, then human-approve the build and deploy. Munder Difflin gives you the worktree isolation, single-committer git, skills, and approval queue out of the box." } },
- { "@type": "Question", "name": "Why use a multi-agent hive instead of one prompt to write blog posts?", "acceptedAnswer": { "@type": "Answer", "text": "One prompt writes one post. A hive runs a content function: a writer drafts, a reviewer checks, the orchestrator integrates and builds, and a scheduled mission can fire the loop on a cadence — so you get a steady, compounding stream of on-topic posts rather than a single output you have to re-prompt for each time." } }
+ { "@type": "Question", "name": "Did an AI actually write the Munder Difflin blog?", "acceptedAnswer": { "@type": "Answer", "text": "Mostly, yes. Agents in our own Munder Difflin office draft posts from a brief and a house style, and the illustrations are drawn in code. The change goes up as a pull request, a person reads it and merges it, and the merge publishes it. This post was rewritten that way in September 2026." } },
+ { "@type": "Question", "name": "Is the blog writer agent fully autonomous?", "acceptedAnswer": { "@type": "Answer", "text": "Everything before publishing is. Research, drafting, checks and the pull request run without a person. Publishing waits for a human merge on purpose, so nothing unreviewed reaches the live site." } },
+ { "@type": "Question", "name": "How do I build my own blog writer agent?", "acceptedAnswer": { "@type": "Answer", "text": "Give one agent a brief, a house style reference and its own workspace. Have a second agent or a checklist review the draft, then send it wherever a human approves changes, like a pull request. Put the loop on a schedule once a few posts come out right." } },
+ { "@type": "Question", "name": "What makes an AI written post worth reading?", "acceptedAnswer": { "@type": "Answer", "text": "Something only you can say: a command you ran, a screenshot of the real screen, a result you measured and dated, a mistake you made. Without that, a draft is a summary of whatever already ranks, and readers can tell." } },
+ { "@type": "Question", "name": "Why use a team of agents instead of one prompt?", "acceptedAnswer": { "@type": "Answer", "text": "One prompt writes one post. A team runs a content function: one agent researches, one drafts, one reviews, and a schedule starts the loop again. You stop prompting for posts and start reviewing them." } }
]
}
@@ -151,15 +152,15 @@
<span>/</span>
<a href="/blog/topics/use-cases/">Use Cases</a>
</nav>
- <h1>Deploy a Blog-Writer Agent: The One That Wrote This Post</h1>
- <p class="dek">How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</p>
+ <h1>Deploy a Blog Writer Agent: The One That Wrote This Post</h1>
+ <p class="dek">How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-06-10">Jun 10, 2026</time>
<span class="dot" aria-hidden="true">·</span>
- <span>6 min read</span>
+ <span>5 min read</span>
<span class="tag kind">Use Cases</span>
</div>
</header>
@@ -168,108 +169,83 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p>The Munder Difflin blog is written by an
-<strong>automated writer agent</strong> living in the hive. The loop: a writer agent drafts a post in an
-<strong>isolated worktree</strong> from a topic brief + a house-style reference; <strong>Michael</strong>
-integrates it as the single committer; <strong>Eleventy</strong> builds
-<code>blog/src/posts</code> → <code>docs/blog</code>; and a human approves the deploy to
-<strong>munderdiffl.in</strong>. The outcome is a steady, on-topic stream — <strong>well over a hundred
-posts and counting</strong>, every one now illustrated by code the same office wrote.
-<em>This post was made that way.</em></p></div>
-<p>Here’s a fact that’s either a confession or a flex, depending on how you read it: the blog you’re reading
-is mostly written by one of our own agents. Not “AI-assisted.” Not “drafted then heavily rewritten.” A
-<strong>writer agent</strong> in the Munder Difflin hive takes a brief, drafts a full post, and hands it down a pipeline
-that ends — after one human nod — on the live site.</p>
-<p>This very post is an instance of that system working. So let me do the most on-brand thing possible and use
-it as the worked example. Here’s how the blog-writer agent is automated, why the outcome compounds, and how
-to stand up your own.</p>
-<h2 id="the-outcome-first-a-library-not-a-launch-post" tabindex="-1">The outcome first: a library, not a launch post <a class="anchor" href="#the-outcome-first-a-library-not-a-launch-post" aria-hidden="true">#</a></h2>
-<p>Before the <em>how</em>, the <em>why it matters</em>. This blog now holds <strong>well over a hundred published posts</strong> —
-the topic chips at the top of <a href="/blog/">the index</a> show the live counts. They’re not filler — they
-cluster into a real content strategy: guides, internals deep-dives, concept explainers, comparisons,
-orchestration patterns, and a pointed story set, including <a href="/blog/why-we-built-munder-difflin/">why we built
-Munder Difflin</a> and the
-<a href="/blog/what-reddit-told-us-about-munder-difflin/">launch-week retros</a>. The thought-leadership thread
-does the heavy SEO lifting — the <a href="/blog/the-multi-agent-cost-playbook/">multi-agent cost playbook</a>,
-<a href="/blog/compressing-agent-memory/">compressing agent memory</a>,
-<a href="/blog/context-engineering-for-ai-agents/">context engineering for AI agents</a>, and more.</p>
-<p>That’s the whole point of automating the writer: <strong>volume that stays on-topic compounds</strong>. A hundred-plus
-internally-linked, keyword-targeted posts is a discoverability moat you cannot hand-write at a startup’s
-spare-time pace. When someone — or an AI answer engine — searches “single-committer git multi-agent” or
-“compressing agent memory,” there’s a post for that, and it links to five neighbors. The blog-writer agent
-is how a one-person project publishes like a content team.</p>
-<p>And crucially: it’s <em>largely hands-off</em>. The expensive part of blogging isn’t typing — it’s the discipline
-to keep shipping. An agent has no problem with discipline.</p>
-<figure class="fig fig-inline"><img src="/blog/assets/media/deploy-a-blog-writer-agent/note-1.png" alt="Hand-drawn sketch from “Deploy a Blog-Writer Agent: The One That Wrote This Post”" loading="lazy" decoding="async" /><figcaption>The content function: brief in, draft in a worktree, one committer, one build, one human gate — on a timer.</figcaption></figure>
-<h2 id="the-pipeline-end-to-end" tabindex="-1">The pipeline, end to end <a class="anchor" href="#the-pipeline-end-to-end" aria-hidden="true">#</a></h2>
-<p>The writer agent isn’t a magic monolith. It’s one role in a hive, and the post moves through the same
-machinery any work does. Four stages:</p>
-<h3 id="1-draft-in-an-isolated-worktree" tabindex="-1">1. Draft — in an isolated worktree <a class="anchor" href="#1-draft-in-an-isolated-worktree" aria-hidden="true">#</a></h3>
-<p>The writer agent gets a <strong>brief</strong> (a topic + intent, usually straight from our SEO backlog) and a
-<strong>house-style reference</strong>: a handful of existing posts to mirror for voice, front-matter, and structure. It
-works in its <strong>own git worktree</strong> — a separate working directory so its in-progress draft never collides
-with anyone else’s files. (We wrote about why that isolation matters in
-<a href="/blog/claude-code-git-worktrees-vs-hive/">git worktrees vs a hive</a>.) Since v0.4.4 the writer can also
-carry an installed <strong>skill</strong> — a house-style checklist it re-reads on every draft, instead of a prompt
-we hope it remembers.</p>
-<p>The agent reads the reference posts, picks internal links to neighbors, writes the front-matter (title,
-description, category, keywords, FAQ schema), and drafts ~1,000–1,400 words of body. It self-checks
-against the style reference before handing off. The output is a single <code>.md</code> file in
-<code>blog/src/posts/</code> — filename becomes the URL slug.</p>
-<h3 id="2-integrate-the-orchestrator-is-the-single-committer" tabindex="-1">2. Integrate — the orchestrator is the single committer <a class="anchor" href="#2-integrate-the-orchestrator-is-the-single-committer" aria-hidden="true">#</a></h3>
-<p>The writer <strong>never commits</strong>. It writes a plain file; <strong>Michael</strong> owns every commit. This is
-the <a href="/blog/single-committer-git-pattern/">single-committer pattern</a> — agents write files, one process
-serializes all the git, so parallel agents never race on <code>.git/index.lock</code> and the repo stays a clean audit
-log. The orchestrator picks up the finished draft, reviews routing, and commits it into the real tree.</p>
-<h3 id="3-build-eleventy-turns-markdown-into-a-site" tabindex="-1">3. Build — Eleventy turns markdown into a site <a class="anchor" href="#3-build-eleventy-turns-markdown-into-a-site" aria-hidden="true">#</a></h3>
-<p>A static <a href="https://www.11ty.dev/">Eleventy</a> build compiles <code>blog/src/posts</code> → <code>docs/blog</code>. Dropping one
-markdown file is the entire authoring action: the build auto-adds the post to the index, its topic page,
-each tag page, the sitemap, and the RSS feed — plus the SEO that’s already wired (canonical URLs, OpenGraph,
-<code>BlogPosting</code> + <code>FAQPage</code> JSON-LD). No other file gets touched. That’s deliberate: the agent’s job is <em>write
-one file correctly</em>, and the build does the rest.</p>
-<p><strong>And since August 2026, the pictures are part of the pipeline too.</strong> Every post’s hero and inline
-sketches are <a href="/blog/an-agent-redesigned-this-blog/">drawn as code by an agent</a> — an SVG parts library,
-composition archetypes, and one spec file, rendered through a headless browser for exactly $0 in image
-APIs. One <code>media.json</code> manifest drives every image on the blog; a post ships with designed placeholders
-until its drawings land, so nothing is ever broken.</p>
-<h3 id="4-deploy-human-gated-on-purpose" tabindex="-1">4. Deploy — human-gated, on purpose <a class="anchor" href="#4-deploy-human-gated-on-purpose" aria-hidden="true">#</a></h3>
-<p>This is the one stage that is <strong>not</strong> automated, by design. The build output under <code>docs/blog</code> is served by
-GitHub Pages at <strong><a href="http://munderdiffl.in/blog">munderdiffl.in/blog</a></strong>. Before that goes live, a person reviews the diff and approves the
-deploy. The orchestrator <a href="/blog/how-the-god-orchestrator-works/">escalates exactly this kind of “publish to the world”
-action</a> to the human-approval queue rather than shipping it itself.</p>
-<p>The split is the lesson: <strong>drafting and integration are autonomous; publishing is human-gated.</strong> You get the
-throughput of an agent and the safety of a final human read. Nobody wants a hallucinated claim on their
-front page — so that one gate stays manual while everything upstream runs hands-off.</p>
-<figure class="fig fig-inline"><img src="/blog/assets/media/deploy-a-blog-writer-agent/note-2.png" alt="Hand-drawn sketch from “Deploy a Blog-Writer Agent: The One That Wrote This Post”" loading="lazy" decoding="async" /><figcaption>One gate stays human on purpose: the deploy. Everything upstream of it runs itself.</figcaption></figure>
-<h2 id="build-your-own-blog-writer-agent" tabindex="-1">Build your own blog-writer agent <a class="anchor" href="#build-your-own-blog-writer-agent" aria-hidden="true">#</a></h2>
-<p>You don’t need our exact stack. The pattern transfers to any static site or CMS. Here’s the recipe.</p>
-<p><strong>1. Give it a brief.</strong> One topic, the search intent, and the angle. Pull it from a keyword backlog so the
-agent is always writing something discoverable, not random.</p>
-<p><strong>2. Give it a house-style reference.</strong> This is the highest-leverage input. Point the agent at three to five
-of your <em>best</em> existing posts and tell it to mirror their front-matter, voice, length, and link density —
-or install that guidance as a <strong>skill</strong>, so every draft re-reads it. A writer with a strong reference
-produces something publishable; a writer without one produces generic AI slop.</p>
-<p><strong>3. Isolate the draft.</strong> Let the agent write in its own worktree (or branch, or scratch directory) so an
-in-flight draft can’t clobber live files. In Munder Difflin this is a per-agent Git isolation toggle.</p>
-<p><strong>4. Review before publish.</strong> Have a reviewer agent or your orchestrator check the draft against the brief,
-then put the <strong>deploy</strong> behind a human approval. Draft and integrate automatically; publish on a click.</p>
-<p><strong>5. Make it recurring.</strong> The real unlock is a scheduled mission: fire the writer on a cadence, feeding it
-the next backlog item each time. One prompt to the orchestrator stood up <a href="/blog/one-prompt-automated-pr-review/">an hourly PR
-reviewer</a> for us the same way — automation that just keeps running.
-Point that same scheduling at content and the blog writes itself on a timer.</p>
-<h2 id="the-meta-point" tabindex="-1">The meta-point <a class="anchor" href="#the-meta-point" aria-hidden="true">#</a></h2>
-<p>A multi-agent hive isn’t only for code. Once you have a writer that drafts, an orchestrator that integrates
-and commits, a build that publishes, and one human gate, you have a <strong>content function</strong> — not a one-off
-prompt. The difference shows up as a library instead of a launch post.</p>
-<p>So consider this post Exhibit A. It was briefed, drafted in a worktree against a style reference, integrated
-single-committer, Eleventy-built, illustrated by code, and human-approved to the domain — the exact loop it
-describes. The system is, quite literally, writing about itself.</p>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p>The blog you are reading is <strong>mostly
+written by agents</strong> in our own office. An agent takes a <strong>brief</strong> and a <strong>house
+style</strong>, drafts in its <strong>own worktree</strong>, the draft gets checked, the pictures are
+<strong>drawn in code</strong>, and the change goes up as a <strong>pull request</strong>. A person merges
+it, and the merge publishes it. <em>This post was rewritten exactly that way.</em></p></div>
+<p>Here is a fact that is either a confession or a flex: most of this blog is written by agents. Not “AI
+assisted” in the vague marketing sense. Agents in a Munder Difflin office take a brief, write the post and
+open the pull request. A human does one thing: reads it and merges it.</p>
+<p>This post describes that pipeline, and in September 2026 the pipeline rewrote this post, so it doubles as
+the worked example. Here is how it runs and how to build your own.</p>
+<h2 id="what-does-a-blog-writer-agent-actually-do" tabindex="-1">What does a blog writer agent actually do? <a class="anchor" href="#what-does-a-blog-writer-agent-actually-do" aria-hidden="true">#</a></h2>
+<p>It turns a brief into a finished change that is ready for review. It reads the brief and the house style,
+researches the topic, drafts the post with its frontmatter and the questions a reader would type, checks its
+own work, and hands the result on. It does not publish.</p>
+<h2 id="how-does-the-pipeline-work-end-to-end" tabindex="-1">How does the pipeline work, end to end? <a class="anchor" href="#how-does-the-pipeline-work-end-to-end" aria-hidden="true">#</a></h2>
+<p>Five stages, and only the last one needs a person.</p>
+<h3 id="1-brief" tabindex="-1">1. Brief <a class="anchor" href="#1-brief" aria-hidden="true">#</a></h3>
+<p>Every post starts from a brief: the topic, the question a reader is trying to answer, and the angle. Briefs
+come from a backlog, so the agent is always writing something somebody is looking for, not whatever it
+happens to feel like.</p>
+<h3 id="2-draft-in-its-own-worktree" tabindex="-1">2. Draft, in its own worktree <a class="anchor" href="#2-draft-in-its-own-worktree" aria-hidden="true">#</a></h3>
+<p>The writer works in its <strong>own git worktree</strong>, a separate checkout, so a half written draft never collides
+with anyone else’s files. (<a href="/blog/claude-code-git-worktrees-vs-hive/">Why that isolation matters</a>.) It reads a
+handful of existing posts for structure, and it can carry the house style as an installed <strong>skill</strong>, a
+checklist it rereads on every draft instead of a prompt we hope it remembers.</p>
+<p>Our house style is short and blunt: no dashes, get to the point, one light joke at most, and nothing we cannot
+back up. The output is a single markdown file, and its filename becomes the URL.</p>
+<h3 id="3-check" tabindex="-1">3. Check <a class="anchor" href="#3-check" aria-hidden="true">#</a></h3>
+<p>A second pass holds the draft against the brief and the style. Are the facts sourced and dated? Does every
+heading answer a question a real person asks? Do the links work? A draft that fails goes back with the reason.</p>
+<h3 id="4-pictures-drawn-in-code" tabindex="-1">4. Pictures, drawn in code <a class="anchor" href="#4-pictures-drawn-in-code" aria-hidden="true">#</a></h3>
+<p>Every hero image and inline sketch on this blog is <a href="/blog/an-agent-redesigned-this-blog/">drawn as code by an agent</a>:
+a library of SVG parts, a set of scene layouts and one spec file, rendered in a headless browser. No image API
+bills. One manifest lists every image, and a post shows a designed placeholder until its drawings land, so a page
+never renders broken.</p>
+<h3 id="5-pull-request-then-one-human-merge" tabindex="-1">5. Pull request, then one human merge <a class="anchor" href="#5-pull-request-then-one-human-merge" aria-hidden="true">#</a></h3>
+<p>The agent opens a pull request with the new post and the rebuilt pages. A person reads it and merges it. Merging
+to main rebuilds the site with <a href="https://www.11ty.dev/">Eleventy</a>, and GitHub Pages serves it at <a href="http://munderdiffl.in/blog">munderdiffl.in/blog</a>.
+The build adds the post to the index, its topic and tag pages, the sitemap and the RSS feed, with the structured
+data already wired in.</p>
+<p>That split is the whole lesson: <strong>writing is automated, publishing is reviewed.</strong> You get an agent’s stamina and
+a human’s final read. Nobody wants a confidently wrong claim on their front page.</p>
+<figure class="fig fig-inline"><img src="/blog/assets/media/deploy-a-blog-writer-agent/note-1.png" alt="Hand-drawn sketch from “Deploy a Blog-Writer Agent: The One That Wrote This Post”" loading="lazy" decoding="async" /><figcaption>Brief in, draft in its own worktree, one human gate at the very end.</figcaption></figure>
+<h2 id="how-do-you-build-your-own-blog-writer-agent" tabindex="-1">How do you build your own blog writer agent? <a class="anchor" href="#how-do-you-build-your-own-blog-writer-agent" aria-hidden="true">#</a></h2>
+<p>Five steps, and none of them needs our exact stack.</p>
+<p><strong>1. Keep a backlog of briefs.</strong> One topic, one reader question and one angle per brief.</p>
+<p><strong>2. Write the house style down.</strong> Point the agent at three to five of your best posts and list the rules you
+actually care about. Install it as a skill so every draft rereads it. A writer with a sharp reference produces
+something publishable. A writer without one produces beige.</p>
+<p><strong>3. Give the draft its own workspace.</strong> A worktree, a branch or a scratch folder. In Munder Difflin this is the
+git isolation toggle on the agent.</p>
+<p><strong>4. Check, then gate publishing.</strong> A reviewer agent or a checklist first, then a pull request or an approval that
+only a human can complete.</p>
+<p><strong>5. Put it on a schedule.</strong> Once a few posts come out right, open the Triggers tab and create a schedule: a label,
+who it goes to, and the prompt that starts the loop. It is the same move that stood up
+<a href="/blog/one-prompt-automated-pr-review/">an hourly PR reviewer</a> for us.</p>
+<figure class="fig fig-inline"><img src="/blog/assets/media/deploy-a-blog-writer-agent/note-2.png" alt="Hand-drawn sketch from “Deploy a Blog-Writer Agent: The One That Wrote This Post”" loading="lazy" decoding="async" /><figcaption>One gate stays human on purpose: publishing. Everything before it runs on its own.</figcaption></figure>
+<h2 id="what-makes-an-ai-written-post-worth-reading" tabindex="-1">What makes an AI written post worth reading? <a class="anchor" href="#what-makes-an-ai-written-post-worth-reading" aria-hidden="true">#</a></h2>
+<p>Something only you can say. A draft built only from what already ranks is a summary of other people’s posts, and
+readers smell it quickly. Give the writer real material: a command you ran, a screenshot of the actual screen, a
+number you measured and dated, a mistake you made and fixed. The agent brings structure and stamina. You bring the
+proof.</p>
+<p>Two more rules we follow:</p>
+<ul>
+<li><strong>Write for a job someone is doing.</strong> “How do I run pi on a local model?” beats “Some thoughts on local AI”.</li>
+<li><strong>Refresh instead of duplicating.</strong> When the product changes, update the post that already exists rather than
+writing a second one that competes with it. This rewrite is that rule in action.</li>
+</ul>
+<h2 id="the-meta-point" tabindex="-1">The meta point <a class="anchor" href="#the-meta-point" aria-hidden="true">#</a></h2>
+<p>A multi agent office is not only for code. Once you have a writer that drafts, a check that catches mistakes, a build
+that publishes and one human gate, you have a content function instead of a one off prompt.</p>
+<p>So this post is exhibit A: briefed, rewritten in a worktree against a house style, checked, and handed to a human as
+a pull request.</p>
<hr>
-<p>Munder Difflin runs a <a href="https://munderdiffl.in/#how">hive of agents on ten engines</a> on
-your own machine — with isolated worktrees, single-committer git, skills, and a human-approval queue built
-in, so an agent can draft and integrate while you keep the one gate that matters.
-<a href="https://munderdiffl.in/#install">Download Munder Difflin</a> to put a blog-writer (or any worker) on your
-floor; it’s free and open source.</p>
+<p>Munder Difflin runs an office of agents on twelve terminal CLIs on your own machine, with worktrees, skills, schedules
+and an ASK ME board for the decisions that need you. <a href="https://munderdiffl.in/">Download it free</a> and put a writer, or
+any other worker, on your floor.</p>
@@ -278,22 +254,27 @@ floor; it’s free and open source.</p>
<div class="qa">
<p class="q">Did an AI actually write the Munder Difflin blog?</p>
- <p class="a">Yes — most of it. A writer agent in the hive drafts each post from a topic brief and a house-style reference, Michael integrates it as the single committer, Eleventy builds the markdown into the static site, and a human approves the final deploy to munderdiffl.in. This very post is an example of that pipeline running — and as of August 2026, the same office also draws every post&#39;s illustrations as code.</p>
+ <p class="a">Mostly, yes. Agents in our own Munder Difflin office draft posts from a brief and a house style, and the illustrations are drawn in code. The change goes up as a pull request, a person reads it and merges it, and the merge publishes it. This post was rewritten that way in September 2026.</p>
</div>
<div class="qa">
- <p class="q">Is the blog-writer agent fully autonomous?</p>
- <p class="a">Almost — it&#39;s deliberately human-gated at one point: publish. The agent drafts and self-reviews against a style reference in an isolated worktree; the orchestrator integrates and builds; a person reviews the diff and approves the deploy. That keeps the volume high and hands-off without putting an unreviewed post on the live domain.</p>
+ <p class="q">Is the blog writer agent fully autonomous?</p>
+ <p class="a">Everything before publishing is. Research, drafting, checks and the pull request run without a person. Publishing waits for a human merge on purpose, so nothing unreviewed reaches the live site.</p>
</div>
<div class="qa">
- <p class="q">How do I build my own blog-writer agent?</p>
- <p class="a">Give one agent in the hive three things: a topic or brief, a house-style reference (a few of your best existing posts), and write access to an isolated worktree. Let it draft, have a reviewer agent or the orchestrator check it, then human-approve the build and deploy. Munder Difflin gives you the worktree isolation, single-committer git, skills, and approval queue out of the box.</p>
+ <p class="q">How do I build my own blog writer agent?</p>
+ <p class="a">Give one agent a brief, a house style reference and its own workspace. Have a second agent or a checklist review the draft, then send it wherever a human approves changes, like a pull request. Put the loop on a schedule once a few posts come out right.</p>
</div>
<div class="qa">
- <p class="q">Why use a multi-agent hive instead of one prompt to write blog posts?</p>
- <p class="a">One prompt writes one post. A hive runs a content function: a writer drafts, a reviewer checks, the orchestrator integrates and builds, and a scheduled mission can fire the loop on a cadence — so you get a steady, compounding stream of on-topic posts rather than a single output you have to re-prompt for each time.</p>
+ <p class="q">What makes an AI written post worth reading?</p>
+ <p class="a">Something only you can say: a command you ran, a screenshot of the real screen, a result you measured and dated, a mistake you made. Without that, a draft is a summary of whatever already ranks, and readers can tell.</p>
+ </div>
+
+ <div class="qa">
+ <p class="q">Why use a team of agents instead of one prompt?</p>
+ <p class="a">One prompt writes one post. A team runs a content function: one agent researches, one drafts, one reviews, and a schedule starts the loop again. You stop prompting for posts and start reviewing them.</p>
</div>
</section>
@@ -312,7 +293,7 @@ floor; it’s free and open source.</p>
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#the-outcome-first-a-library-not-a-launch-post">The outcome first: a library, not a launch post</a></li><li class="lvl-2"><a href="#the-pipeline-end-to-end">The pipeline, end to end</a></li><li class="lvl-3"><a href="#1-draft-in-an-isolated-worktree">1. Draft — in an isolated worktree</a></li><li class="lvl-3"><a href="#2-integrate-the-orchestrator-is-the-single-committer">2. Integrate — the orchestrator is the single committer</a></li><li class="lvl-3"><a href="#3-build-eleventy-turns-markdown-into-a-site">3. Build — Eleventy turns markdown into a site</a></li><li class="lvl-3"><a href="#4-deploy-human-gated-on-purpose">4. Deploy — human-gated, on purpose</a></li><li class="lvl-2"><a href="#build-your-own-blog-writer-agent">Build your own blog-writer agent</a></li><li class="lvl-2"><a href="#the-meta-point">The meta-point</a></li>
+ <li class="lvl-2"><a href="#what-does-a-blog-writer-agent-actually-do">What does a blog writer agent actually do?</a></li><li class="lvl-2"><a href="#how-does-the-pipeline-work-end-to-end">How does the pipeline work, end to end?</a></li><li class="lvl-3"><a href="#1-brief">1. Brief</a></li><li class="lvl-3"><a href="#2-draft-in-its-own-worktree">2. Draft, in its own worktree</a></li><li class="lvl-3"><a href="#3-check">3. Check</a></li><li class="lvl-3"><a href="#4-pictures-drawn-in-code">4. Pictures, drawn in code</a></li><li class="lvl-3"><a href="#5-pull-request-then-one-human-merge">5. Pull request, then one human merge</a></li><li class="lvl-2"><a href="#how-do-you-build-your-own-blog-writer-agent">How do you build your own blog writer agent?</a></li><li class="lvl-2"><a href="#what-makes-an-ai-written-post-worth-reading">What makes an AI written post worth reading?</a></li><li class="lvl-2"><a href="#the-meta-point">The meta point</a></li>
</ol>
</aside>
diff --git a/docs/blog/deploy-automated-pr-reviewer-agent/index.html b/docs/blog/deploy-automated-pr-reviewer-agent/index.html
index 9dc67155..0c8f0b30 100644
--- a/docs/blog/deploy-automated-pr-reviewer-agent/index.html
+++ b/docs/blog/deploy-automated-pr-reviewer-agent/index.html
@@ -279,7 +279,7 @@
<footer class="post-foot wrap"><div class="prevnext">
<a class="pn prev" href="/blog/run-ai-agent-hive-from-slack-setup/"><span class="k">← Older</span><span class="t">Run Your AI Agent Hive from Slack: The Complete Setup</span></a>
- <a class="pn next" href="/blog/deploy-a-blog-writer-agent/"><span class="k">Newer →</span><span class="t">Deploy a Blog-Writer Agent: The One That Wrote This Post</span></a>
+ <a class="pn next" href="/blog/deploy-a-blog-writer-agent/"><span class="k">Newer →</span><span class="t">Deploy a Blog Writer Agent: The One That Wrote This Post</span></a>
</div>
<section class="related">
<h2>Related in Orchestration</h2>
diff --git a/docs/blog/does-the-june-2026-agent-sdk-change-affect-munder-difflin/index.html b/docs/blog/does-the-june-2026-agent-sdk-change-affect-munder-difflin/index.html
index 911c14d4..6e9b74dc 100644
--- a/docs/blog/does-the-june-2026-agent-sdk-change-affect-munder-difflin/index.html
+++ b/docs/blog/does-the-june-2026-agent-sdk-change-affect-munder-difflin/index.html
@@ -307,13 +307,13 @@ rides your existing Claude subscription. Local-first, on your own plan, no surpr
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -339,10 +339,10 @@ rides your existing Claude subscription. Local-first, on your own plan, no surpr
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/evaluating-ai-agent-reliability/index.html b/docs/blog/evaluating-ai-agent-reliability/index.html
index 9ac1c2d1..86ea5d59 100644
--- a/docs/blog/evaluating-ai-agent-reliability/index.html
+++ b/docs/blog/evaluating-ai-agent-reliability/index.html
@@ -315,13 +315,13 @@ your bar.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -347,10 +347,10 @@ your bar.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/feed.xml b/docs/blog/feed.xml
index 8636f09e..4eb84aa6 100644
--- a/docs/blog/feed.xml
+++ b/docs/blog/feed.xml
@@ -128,20 +128,20 @@
<summary>The v0.3.4 + v0.3.5 wave: voice orchestration with live floor context and full app control, markdown previews everywhere, commit history and branch compare in the built-in IDE, a six-tab Settings redesign, xAI Grok and Kimi Code engines, auto-update, and a queue that always has an escape hatch.</summary>
</entry>
<entry>
- <title>qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</title>
+ <title>qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</title>
<link href="https://munderdiffl.in/blog/qm-vs-munder-difflin/" />
<id>https://munderdiffl.in/blog/qm-vs-munder-difflin/</id>
<published>2026-08-06T00:00:00.000Z</published>
- <updated>2026-08-06T00:00:00.000Z</updated><category term="Comparisons" />
- <summary>An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</summary>
+ <updated>2026-09-10T00:00:00.000Z</updated><category term="Comparisons" />
+ <summary>An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</summary>
</entry>
<entry>
- <title>The Command Center: Kanban, Fleet, and Budgets in One Place</title>
+ <title>The Command Center: Kanban, Fleet and Budgets in One Place</title>
<link href="https://munderdiffl.in/blog/command-center-guide/" />
<id>https://munderdiffl.in/blog/command-center-guide/</id>
<published>2026-07-03T00:00:00.000Z</published>
- <updated>2026-07-03T00:00:00.000Z</updated><category term="Guides" />
- <summary>A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</summary>
+ <updated>2026-09-10T00:00:00.000Z</updated><category term="Guides" />
+ <summary>A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</summary>
</entry>
<entry>
<title>CrewAI and AutoGen vs a Local Agent Harness: Framework or App?</title>
@@ -172,8 +172,8 @@
<link href="https://munderdiffl.in/blog/how-to-hire-from-the-agent-gallery/" />
<id>https://munderdiffl.in/blog/how-to-hire-from-the-agent-gallery/</id>
<published>2026-07-03T00:00:00.000Z</published>
- <updated>2026-07-03T00:00:00.000Z</updated><category term="Guides" />
- <summary>A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</summary>
+ <updated>2026-09-10T00:00:00.000Z</updated><category term="Guides" />
+ <summary>A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</summary>
</entry>
<entry>
<title>How to Brief Your Orchestrator (So the Floor Actually Ships)</title>
@@ -268,8 +268,8 @@
<link href="https://munderdiffl.in/blog/your-first-hour-with-munder-difflin/" />
<id>https://munderdiffl.in/blog/your-first-hour-with-munder-difflin/</id>
<published>2026-07-03T00:00:00.000Z</published>
- <updated>2026-08-20T00:00:00.000Z</updated><category term="Guides" />
- <summary>A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</summary>
+ <updated>2026-09-10T00:00:00.000Z</updated><category term="Guides" />
+ <summary>A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</summary>
</entry>
<entry>
<title>Launching Munder Difflin v0.3.2: Talk to Michael, a Realtime Voice Orchestrator</title>
@@ -284,16 +284,16 @@
<link href="https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/" />
<id>https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/</id>
<published>2026-06-22T00:00:00.000Z</published>
- <updated>2026-06-22T00:00:00.000Z</updated><category term="Guides" />
- <summary>A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</summary>
+ <updated>2026-09-10T00:00:00.000Z</updated><category term="Guides" />
+ <summary>Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</summary>
</entry>
<entry>
- <title>Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</title>
+ <title>Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</title>
<link href="https://munderdiffl.in/blog/run-munder-difflin-on-open-models/" />
<id>https://munderdiffl.in/blog/run-munder-difflin-on-open-models/</id>
<published>2026-06-22T00:00:00.000Z</published>
- <updated>2026-08-20T00:00:00.000Z</updated><category term="Guides" />
- <summary>Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</summary>
+ <updated>2026-09-10T00:00:00.000Z</updated><category term="Guides" />
+ <summary>Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</summary>
</entry>
<entry>
<title>Launching Munder Difflin v0.3.0: Selectable Engines, Integrations &amp; Slack-Spawned Workers</title>
@@ -328,12 +328,12 @@
<summary>Why agent roles should be portable: what a hire manifest encodes, how one-click hiring stays safe, and how The Hiring Fair turns tacit setup into a shareable artifact.</summary>
</entry>
<entry>
- <title>Deploy a Blog-Writer Agent: The One That Wrote This Post</title>
+ <title>Deploy a Blog Writer Agent: The One That Wrote This Post</title>
<link href="https://munderdiffl.in/blog/deploy-a-blog-writer-agent/" />
<id>https://munderdiffl.in/blog/deploy-a-blog-writer-agent/</id>
<published>2026-06-10T00:00:00.000Z</published>
- <updated>2026-08-20T00:00:00.000Z</updated><category term="Use Cases" />
- <summary>How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</summary>
+ <updated>2026-09-10T00:00:00.000Z</updated><category term="Use Cases" />
+ <summary>How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</summary>
</entry>
<entry>
<title>Deploy a PR Reviewer That Never Sleeps — In About One Prompt</title>
@@ -452,8 +452,8 @@
<link href="https://munderdiffl.in/blog/how-to-install-and-use-munder-difflin/" />
<id>https://munderdiffl.in/blog/how-to-install-and-use-munder-difflin/</id>
<published>2026-06-05T00:00:00.000Z</published>
- <updated>2026-08-27T00:00:00.000Z</updated><category term="Guides" />
- <summary>A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</summary>
+ <updated>2026-09-10T00:00:00.000Z</updated><category term="Guides" />
+ <summary>A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</summary>
</entry>
<entry>
<title>MCP Tool Poisoning: How to Secure the Model Context Protocol</title>
diff --git a/docs/blog/github-copilot-cli-vs-claude-code-for-hive-work/index.html b/docs/blog/github-copilot-cli-vs-claude-code-for-hive-work/index.html
index 5b1584aa..5dc5f1cd 100644
--- a/docs/blog/github-copilot-cli-vs-claude-code-for-hive-work/index.html
+++ b/docs/blog/github-copilot-cli-vs-claude-code-for-hive-work/index.html
@@ -328,13 +328,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/how-ai-answer-engines-choose-sources/index.html b/docs/blog/how-ai-answer-engines-choose-sources/index.html
index 771995a9..2e63c7b5 100644
--- a/docs/blog/how-ai-answer-engines-choose-sources/index.html
+++ b/docs/blog/how-ai-answer-engines-choose-sources/index.html
@@ -308,13 +308,13 @@ broader map of the tooling AI engines cite, see our
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/how-to-add-a-github-copilot-cli-agent/index.html b/docs/blog/how-to-add-a-github-copilot-cli-agent/index.html
index c541c1b2..08cdcbb2 100644
--- a/docs/blog/how-to-add-a-github-copilot-cli-agent/index.html
+++ b/docs/blog/how-to-add-a-github-copilot-cli-agent/index.html
@@ -288,13 +288,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -304,10 +304,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/how-to-choose-a-multi-agent-tool/index.html b/docs/blog/how-to-choose-a-multi-agent-tool/index.html
index 35da0f70..d7b62ad5 100644
--- a/docs/blog/how-to-choose-a-multi-agent-tool/index.html
+++ b/docs/blog/how-to-choose-a-multi-agent-tool/index.html
@@ -323,13 +323,13 @@ local-first.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/how-to-hire-from-the-agent-gallery/index.html b/docs/blog/how-to-hire-from-the-agent-gallery/index.html
index 1bed249f..19fbf434 100644
--- a/docs/blog/how-to-hire-from-the-agent-gallery/index.html
+++ b/docs/blog/how-to-hire-from-the-agent-gallery/index.html
@@ -4,7 +4,7 @@
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>How to Hire From the Agent Gallery — Munder Difflin Blog</title>
-<meta name="description" content="A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself." />
+<meta name="description" content="A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself." />
<link rel="canonical" href="https://munderdiffl.in/blog/how-to-hire-from-the-agent-gallery/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -15,7 +15,7 @@
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
<meta property="og:title" content="How to Hire From the Agent Gallery — Munder Difflin Blog" />
-<meta property="og:description" content="A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself." />
+<meta property="og:description" content="A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself." />
<meta property="og:url" content="https://munderdiffl.in/blog/how-to-hire-from-the-agent-gallery/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -31,7 +31,7 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="How to Hire From the Agent Gallery — Munder Difflin Blog" />
-<meta name="twitter:description" content="A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself." />
+<meta name="twitter:description" content="A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -45,10 +45,10 @@
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "How to Hire From the Agent Gallery",
- "description": "A practical guide to Munder Difflin's Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.",
+ "description": "A practical guide to Munder Difflin's Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-07-03T00:00:00.000Z",
- "dateModified": "2026-07-03T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -56,7 +56,7 @@
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/how-to-hire-from-the-agent-gallery/" },
- "keywords": "agent gallery, hire an ai agent, off-the-shelf agent roles, import agent manifest, munder difflin hires, ready-made ai coworkers",
+ "keywords": "agent gallery, hire an ai agent, ready made ai agent roles, import agent manifest, munder difflin hires, ai coworker templates",
"articleSection": "Guides",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/how-to-hire-from-the-agent-gallery/"
@@ -79,12 +79,12 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "What is the Agent Gallery?", "acceptedAnswer": { "@type": "Answer", "text": "It's a free gallery of ready-made agent roles for Munder Difflin at munderdiffl.in/hires — no login, no trackers. Each role, called a hire, is a small JSON manifest that packages a configured agent: name, avatar, provider, model, flags, a goal, capability tags, and a token budget. It ships with six off-the-shelf hires you can import in one click and adapt to your project." } },
- { "@type": "Question", "name": "Does importing a hire from the gallery start an agent automatically?", "acceptedAnswer": { "@type": "Answer", "text": "No. Import only pre-fills the Add-Agent modal behind an explicit imported banner. You review every field, change anything you like, and the agent spawns only when you click spawn. There is no code path that runs an agent on import — the human is always the spawn gate." } },
- { "@type": "Question", "name": "Is it safe to import a hire manifest someone sent me?", "acceptedAnswer": { "@type": "Answer", "text": "The import pipeline is built on the assumption that it isn't — every manifest is validated as untrusted input. There is no executable field, so a manifest can never name a program to run; command flags pass a default-deny allowlist; and deep-link fetches are https-only and bounded. Even after all that, you still review the pre-filled form before anything spawns. Treat the goal text with the same skepticism you'd give any file from the internet." } },
- { "@type": "Question", "name": "Can I change the engine or model a gallery hire uses?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. The manifest is a starting point, not a lock. In the Add-Agent modal you can switch the hire onto any engine you have — Claude Code, Antigravity, Codex, OpenCode, Crush, pi.dev, or GitHub Copilot CLI — pick a different model, and on local-capable engines use the OSS-model quick-picks to run it on open models via Ollama or a BYOK provider." } },
- { "@type": "Question", "name": "What should I customize before spawning a hire?", "acceptedAnswer": { "@type": "Answer", "text": "Four things. Identity: the name and avatar you'll recognize on the floor. Workspace: the working directory it operates in, ideally with the git-isolation toggle on so it gets its own worktree. Engine: the CLI and model that match the job's difficulty and your budget. Briefing: rewrite the goal so it names your repo, your conventions, and what done means — a generic goal produces generic work." } },
- { "@type": "Question", "name": "How much does hiring from the gallery cost?", "acceptedAnswer": { "@type": "Answer", "text": "The gallery and the app are free and open source (MIT-licensed code). You pay only for the model usage of whatever CLI or key the agent runs on. Every hire carries a token budget field, and the harness pairs per-agent budgets with live fleet monitoring and a circuit breaker, so an imported role can't quietly outspend the ceiling you gave it." } }
+ { "@type": "Question", "name": "What is the Agent Gallery?", "acceptedAnswer": { "@type": "Answer", "text": "A free gallery of ready made agent roles for Munder Difflin at munderdiffl.in/hires. Each role, called a hire, is one JSON file describing a configured agent: its name and avatar, engine and model, flags, goal, skills and token budget. As of September 2026 it lists 80 roles." } },
+ { "@type": "Question", "name": "How do I import a hire?", "acceptedAnswer": { "@type": "Answer", "text": "Download the role's .json from its card, open Munder Difflin, click Add agent, then import hire, and pick the file. The form fills in with every field from the manifest." } },
+ { "@type": "Question", "name": "Does importing a hire start an agent?", "acceptedAnswer": { "@type": "Answer", "text": "No. Importing only fills in the Add agent form for you to review. Nothing spawns until you click spawn yourself." } },
+ { "@type": "Question", "name": "Is it safe to import a manifest someone sent me?", "acceptedAnswer": { "@type": "Answer", "text": "The import treats every manifest as untrusted. A manifest cannot name a program to run or carry shell syntax, and you review the final command before anything starts. Still read the goal text with the care you would give any file from the internet, because it is someone else's prompt." } },
+ { "@type": "Question", "name": "Can I change the engine or model a hire uses?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. The manifest is a starting point. In Add agent you can switch the role onto any of the twelve supported CLIs you have installed, pick a different model, and on OpenCode, Crush or Pi choose one of the open model quick picks." } },
+ { "@type": "Question", "name": "What does hiring from the gallery cost?", "acceptedAnswer": { "@type": "Answer", "text": "The gallery and the classic app are free. You pay only for the model usage of the CLI or key the agent runs on, and every agent's token budget is enforced by the app with a circuit breaker behind it." } }
]
}
@@ -154,14 +154,14 @@
<a href="/blog/topics/guides/">Guides</a>
</nav>
<h1>How to Hire From the Agent Gallery</h1>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true">·</span>
- <span>5 min read</span>
+ <span>4 min read</span>
<span class="tag kind">Guides</span>
</div>
</header>
@@ -170,40 +170,64 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p>The <strong>Agent Gallery</strong> (<a href="https://munderdiffl.in/hires/">munderdiffl.in/hires</a>) is where you hire ready-made AI coworkers for Munder Difflin — it ships with <strong>six off-the-shelf hires</strong>. A hire is a small JSON manifest: provider, model, flags, goal, capabilities, token budget. Importing one — by <code>munderdifflin://hire</code> link or a local file — <strong>never runs anything</strong>: it only pre-fills the Add-Agent modal behind an "imported" banner, and the manifest is <strong>validated as untrusted input</strong> along the way. You review, customize the <strong>identity, workspace, engine, and briefing</strong>, and <em>you</em> click spawn. Five minutes from browsing to a working agent on your floor.</p></div>
-<p>The slowest part of running a multi-agent office isn’t running the agents. It’s the blank Add-Agent form: which engine, which model, which flags, and — the one that actually stalls people — what is this thing’s <em>job</em>? The Agent Gallery exists so you never start from that blank box. You start from a role somebody already got right, and you edit.</p>
-<p>This is the practical walkthrough: browse, import, review, customize, spawn. (If you haven’t installed the app yet, do <a href="/blog/how-to-install-and-use-munder-difflin/">that first</a> — the gallery assumes you have a floor to hire onto.)</p>
-<h2 id="step-1-browse-the-gallery" tabindex="-1">Step 1: Browse the gallery <a class="anchor" href="#step-1-browse-the-gallery" aria-hidden="true">#</a></h2>
-<p>The gallery lives at <a href="https://munderdiffl.in/hires/">munderdiffl.in/hires</a> — a static page, no login, no trackers. It’s stocked with <strong>six off-the-shelf hires</strong>, each a complete role: a PR reviewer, a docs writer, a QA enforcer, a security auditor, a token-spend auditor, and a migrations grinder, each fronted by a pixel avatar from the app’s affectionate Office parody cast.</p>
-<p>Every card tells you what you’re getting before you click anything: the role’s goal, its capability tags, and which provider it’s written for. Don’t overthink the choice — you’re picking a starting point, not signing a contract. Everything on the card is editable after import.</p>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p>The <strong>Agent Gallery</strong> at
+<a href="https://munderdiffl.in/hires/">munderdiffl.in/hires</a> holds <strong>80 ready made roles</strong> for Munder Difflin.
+Download a role's <code>.json</code>, click <strong>Add agent</strong>, then <strong>import hire</strong>, and the form fills in
+for you. <strong>Nothing runs on import.</strong> You review every field, customise the <strong>identity, workspace, engine and
+briefing</strong>, and you click spawn. About five minutes from browsing to a working agent on your floor.</p></div>
+<p>The slowest part of running an office of agents is not the agents. It is the blank Add agent form: which engine, which model, which
+flags, and the question that really stalls people, what is this agent’s job? The gallery exists so you never start from a blank
+form. You start from a role somebody already got right, and you edit it.</p>
+<p>This is the walkthrough as of Munder Difflin 0.5.2: browse, import, review, customise, spawn. Not installed yet?
+<a href="/blog/how-to-install-and-use-munder-difflin/">Do that first</a>, because the gallery assumes you have a floor to hire onto.</p>
+<h2 id="where-is-the-agent-gallery-and-what-is-in-it" tabindex="-1">Where is the Agent Gallery, and what is in it? <a class="anchor" href="#where-is-the-agent-gallery-and-what-is-in-it" aria-hidden="true">#</a></h2>
+<p>At <a href="https://munderdiffl.in/hires/">munderdiffl.in/hires</a>, a static page with no login. It lists 80 roles, and they are not all
+engineers. Next to a PR reviewer, a QA enforcer and a security auditor you will find roles for research, data analysis, docs, design,
+customer support, sales outreach and marketing. Each card shows what the role is for, and you can search the page.</p>
+<p>Do not overthink the choice. You are picking a starting point, not signing a contract, and every field is editable after import.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/how-to-hire-from-the-agent-gallery/note-1.png" alt="Hand-drawn sketch from “How to Hire From the Agent Gallery”" loading="lazy" decoding="async" /></figure>
-<h2 id="step-2-import-the-hire" tabindex="-1">Step 2: Import the hire <a class="anchor" href="#step-2-import-the-hire" aria-hidden="true">#</a></h2>
-<p>Two ways in, both landing in the same place:</p>
+<h2 id="how-do-you-import-a-hire" tabindex="-1">How do you import a hire? <a class="anchor" href="#how-do-you-import-a-hire" aria-hidden="true">#</a></h2>
+<p>Three steps:</p>
+<ol>
+<li>On the role’s card, click <strong>download</strong> to save its <code>.json</code>.</li>
+<li>In Munder Difflin, click <strong>Add agent</strong>, then <strong>import hire</strong>.</li>
+<li>Pick the file you just downloaded.</li>
+</ol>
+<p>The Add agent form opens with every field filled in from the manifest. Because a hire is just a file, a teammate can also send you one
+directly, and you import it the same way.</p>
+<p>The one thing to be precise about: <strong>importing never spawns anything.</strong> It only fills in the form. The agent starts when you click spawn,
+and not a moment before.</p>
+<h2 id="why-read-the-manifest-before-you-spawn" tabindex="-1">Why read the manifest before you spawn? <a class="anchor" href="#why-read-the-manifest-before-you-spawn" aria-hidden="true">#</a></h2>
+<p>Because the goal text is somebody else’s prompt. The import treats every manifest as untrusted input: a manifest cannot name a program to
+run or carry shell syntax, so the program always comes from your own installed engine. What validation cannot vouch for is the goal,
+which is free text. So take thirty seconds:</p>
<ul>
-<li><strong>Deep link.</strong> Click the hire button on a card and the <code>munderdifflin://hire</code> link opens the app directly.</li>
-<li><strong>Local file.</strong> Save the manifest JSON (or receive one from a teammate) and import it from disk.</li>
-</ul>
-<p>Either way, one thing is worth being precise about: <strong>import never spawns anything.</strong> What opens is the app’s ordinary Add-Agent modal, pre-filled with the manifest’s values and marked with an explicit <strong>“imported” banner</strong> so you know these fields came from outside. The agent starts only when you click spawn. A hire you got from the internet is inert data until a human acts on it.</p>
-<p>That’s not an accident of UI — it’s a security posture. A manifest arrives by clicked link or downloaded file, which makes it attacker-controlled bytes, and the import pipeline treats it that way: there’s <strong>no executable field</strong> (the binary always comes from your own locally configured provider presets), command flags pass a <strong>default-deny allowlist</strong>, and the deep-link fetch is <strong>https-only and bounded</strong>. The full threat model is its own post: <a href="/blog/hire-manifest-untrusted-input/">treating a hire manifest as untrusted input</a>.</p>
-<h2 id="step-3-actually-read-the-manifest" tabindex="-1">Step 3: Actually read the manifest <a class="anchor" href="#step-3-actually-read-the-manifest" aria-hidden="true">#</a></h2>
-<p>The imported banner is your cue to slow down for thirty seconds. A hire manifest encodes the whole role — name, avatar, provider, model, command flags, goal, capability tags, and a token budget — and every one of those is now sitting in front of you as a form field. Read them like you’d read a job description before forwarding it to HR:</p>
-<ul>
-<li><strong>Goal text</strong> is the prompt this agent will work from. It’s free text from someone else, which makes it the one field validation can’t vouch for. If anything in it looks off, rewrite it.</li>
-<li><strong>Model and flags</strong> tell you what it costs to run and how it behaves. The allowlist keeps flags safe; it doesn’t make them right for you.</li>
-<li><strong>Token budget</strong> is the spend ceiling the author suggested. Set it to what <em>you</em> are comfortable with — the harness enforces per-agent budgets with live fleet monitoring and a circuit breaker behind it.</li>
+<li><strong>Goal.</strong> The prompt this agent will work from. If anything in it looks off, rewrite it.</li>
+<li><strong>Model and flags.</strong> What the agent costs to run and how it behaves.</li>
+<li><strong>Token budget.</strong> The ceiling the author suggested. Set the one you are comfortable with, because the app enforces it.</li>
+<li><strong>Skills.</strong> The skills this hire activates. Keep only what the job needs.</li>
</ul>
+<p>The full threat model is its own post: <a href="/blog/hire-manifest-untrusted-input/">treating a hire manifest as untrusted input</a>.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/how-to-hire-from-the-agent-gallery/note-2.png" alt="Hand-drawn sketch from “How to Hire From the Agent Gallery”" loading="lazy" decoding="async" /></figure>
-<h2 id="step-4-customize-the-four-things-that-matter" tabindex="-1">Step 4: Customize the four things that matter <a class="anchor" href="#step-4-customize-the-four-things-that-matter" aria-hidden="true">#</a></h2>
-<p>A gallery hire is deliberately generic. Four edits turn it into <em>your</em> hire:</p>
-<p><strong>Identity.</strong> Rename it and pick the avatar you want to see walking the floor. This sounds cosmetic; it isn’t. When six agents are working at once, you triage by face and name.</p>
-<p><strong>Workspace.</strong> Point the hire at the repo it should work in, and flip on the <strong>git isolation</strong> toggle so it auto-provisions its own worktree on spawn — agents that share a checkout collide on branches; agents with worktrees don’t (<a href="/blog/claude-code-git-worktrees-vs-hive/">why worktrees, in depth</a>).</p>
-<p><strong>Engine.</strong> The manifest names a provider, but the modal’s picker is yours: run the role on Claude Code, Antigravity, Codex, OpenCode, Crush, pi.dev, or GitHub Copilot CLI. Match the engine to the job — and if the role is routine enough, the local-capable engines surface <strong>OSS-model quick-picks</strong> so it can run on open models for pennies.</p>
-<p><strong>Briefing.</strong> Rewrite the goal until it names your project, your conventions, and what “done” looks like. This is the highest-leverage sixty seconds of the whole flow: a hire that arrives with “review pull requests” leaves with “review PRs on repo X, enforce the lint config, never approve without a green typecheck.”</p>
-<p>One more gate you’ll meet after spawning: hires carry a manifest of allowed <strong>skills and MCP servers</strong>, default-deny over a shared catalog. Anything the hire wants to use surfaces in a consent UI for you to grant — imported input is reviewed, never auto-granted.</p>
-<h2 id="step-5-spawn-and-put-it-to-work" tabindex="-1">Step 5: Spawn, and put it to work <a class="anchor" href="#step-5-spawn-and-put-it-to-work" aria-hidden="true">#</a></h2>
-<p>Click spawn. The agent appears at a desk, joins the hive with its own mailbox and long-term memory, and is immediately routable: assign it kanban tasks, or just tell the GOD orchestrator what you need and let <a href="/blog/how-the-god-orchestrator-works/">Michael route the work</a>.</p>
-<p>From browsing a card to a working coworker is genuinely a five-minute path — and every minute of it kept you in the loop: you read the manifest, you set the budget, you clicked spawn.</p>
-<p>Grab the latest build from the <a href="https://github.com/chaitanyagiri/munder-difflin/releases/latest">releases page</a>, and if the gallery saves you a blank-form afternoon, a <a href="https://github.com/chaitanyagiri/munder-difflin">GitHub star</a> is appreciated.</p>
+<h2 id="what-should-you-customise-before-spawning" tabindex="-1">What should you customise before spawning? <a class="anchor" href="#what-should-you-customise-before-spawning" aria-hidden="true">#</a></h2>
+<p>Four edits turn a generic role into your hire.</p>
+<p><strong>Identity.</strong> Rename it and pick the avatar you want to see walking the floor. When six agents are working at once, you triage by face and
+name, so this matters more than it sounds.</p>
+<p><strong>Workspace.</strong> Point it at the project it should work in and turn on git isolation, so it gets its own worktree. Agents that share one
+checkout collide on branches, and agents with worktrees do not. (<a href="/blog/claude-code-git-worktrees-vs-hive/">Why worktrees</a>.)</p>
+<p><strong>Engine.</strong> The manifest suggests an engine, but the picker is yours. Run the role on any of the twelve supported CLIs you have installed.
+If the job is routine, OpenCode, Crush and Pi offer open model quick picks, so it can run on a local or cheap model.</p>
+<p><strong>Briefing.</strong> Rewrite the goal until it names your project, your conventions and what done looks like. “Review pull requests” becomes “review
+PRs on this repo, enforce the lint config, and never approve without a green typecheck”. This is the most valuable minute of the whole flow.</p>
+<h2 id="what-happens-after-you-spawn" tabindex="-1">What happens after you spawn? <a class="anchor" href="#what-happens-after-you-spawn" aria-hidden="true">#</a></h2>
+<p>The agent takes a desk on the floor, gets its own mailbox and long term memory, and is ready for work straight away. Assign it tasks on the
+kanban, or tell Michael what you need and let him <a href="/blog/how-the-god-orchestrator-works/">route the work</a>.</p>
+<h2 id="can-you-add-your-own-hire" tabindex="-1">Can you add your own hire? <a class="anchor" href="#can-you-add-your-own-hire" aria-hidden="true">#</a></h2>
+<p>Yes. A hire is one JSON file that follows the hire manifest spec. Start from any card’s JSON, check your version with the validator on the
+gallery page, then import it with Add agent and import hire. Manifests are plain files, so you can share them anywhere.</p>
+<hr>
+<p>Grab the latest build from <a href="https://munderdiffl.in/">munderdiffl.in</a>, and if the gallery saves you an afternoon of blank forms,
+<a href="https://github.com/chaitanyagiri/munder-difflin">a GitHub star</a> is appreciated.</p>
@@ -212,32 +236,32 @@
<div class="qa">
<p class="q">What is the Agent Gallery?</p>
- <p class="a">It&#39;s a free gallery of ready-made agent roles for Munder Difflin at munderdiffl.in/hires — no login, no trackers. Each role, called a hire, is a small JSON manifest that packages a configured agent: name, avatar, provider, model, flags, a goal, capability tags, and a token budget. It ships with six off-the-shelf hires you can import in one click and adapt to your project.</p>
+ <p class="a">A free gallery of ready made agent roles for Munder Difflin at munderdiffl.in/hires. Each role, called a hire, is one JSON file describing a configured agent: its name and avatar, engine and model, flags, goal, skills and token budget. As of September 2026 it lists 80 roles.</p>
</div>
<div class="qa">
- <p class="q">Does importing a hire from the gallery start an agent automatically?</p>
- <p class="a">No. Import only pre-fills the Add-Agent modal behind an explicit imported banner. You review every field, change anything you like, and the agent spawns only when you click spawn. There is no code path that runs an agent on import — the human is always the spawn gate.</p>
+ <p class="q">How do I import a hire?</p>
+ <p class="a">Download the role&#39;s .json from its card, open Munder Difflin, click Add agent, then import hire, and pick the file. The form fills in with every field from the manifest.</p>
</div>
<div class="qa">
- <p class="q">Is it safe to import a hire manifest someone sent me?</p>
- <p class="a">The import pipeline is built on the assumption that it isn&#39;t — every manifest is validated as untrusted input. There is no executable field, so a manifest can never name a program to run; command flags pass a default-deny allowlist; and deep-link fetches are https-only and bounded. Even after all that, you still review the pre-filled form before anything spawns. Treat the goal text with the same skepticism you&#39;d give any file from the internet.</p>
+ <p class="q">Does importing a hire start an agent?</p>
+ <p class="a">No. Importing only fills in the Add agent form for you to review. Nothing spawns until you click spawn yourself.</p>
</div>
<div class="qa">
- <p class="q">Can I change the engine or model a gallery hire uses?</p>
- <p class="a">Yes. The manifest is a starting point, not a lock. In the Add-Agent modal you can switch the hire onto any engine you have — Claude Code, Antigravity, Codex, OpenCode, Crush, pi.dev, or GitHub Copilot CLI — pick a different model, and on local-capable engines use the OSS-model quick-picks to run it on open models via Ollama or a BYOK provider.</p>
+ <p class="q">Is it safe to import a manifest someone sent me?</p>
+ <p class="a">The import treats every manifest as untrusted. A manifest cannot name a program to run or carry shell syntax, and you review the final command before anything starts. Still read the goal text with the care you would give any file from the internet, because it is someone else&#39;s prompt.</p>
</div>
<div class="qa">
- <p class="q">What should I customize before spawning a hire?</p>
- <p class="a">Four things. Identity: the name and avatar you&#39;ll recognize on the floor. Workspace: the working directory it operates in, ideally with the git-isolation toggle on so it gets its own worktree. Engine: the CLI and model that match the job&#39;s difficulty and your budget. Briefing: rewrite the goal so it names your repo, your conventions, and what done means — a generic goal produces generic work.</p>
+ <p class="q">Can I change the engine or model a hire uses?</p>
+ <p class="a">Yes. The manifest is a starting point. In Add agent you can switch the role onto any of the twelve supported CLIs you have installed, pick a different model, and on OpenCode, Crush or Pi choose one of the open model quick picks.</p>
</div>
<div class="qa">
- <p class="q">How much does hiring from the gallery cost?</p>
- <p class="a">The gallery and the app are free and open source (MIT-licensed code). You pay only for the model usage of whatever CLI or key the agent runs on. Every hire carries a token budget field, and the harness pairs per-agent budgets with live fleet monitoring and a circuit breaker, so an imported role can&#39;t quietly outspend the ceiling you gave it.</p>
+ <p class="q">What does hiring from the gallery cost?</p>
+ <p class="a">The gallery and the classic app are free. You pay only for the model usage of the CLI or key the agent runs on, and every agent&#39;s token budget is enforced by the app with a circuit breaker behind it.</p>
</div>
</section>
@@ -256,7 +280,7 @@
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#step-1-browse-the-gallery">Step 1: Browse the gallery</a></li><li class="lvl-2"><a href="#step-2-import-the-hire">Step 2: Import the hire</a></li><li class="lvl-2"><a href="#step-3-actually-read-the-manifest">Step 3: Actually read the manifest</a></li><li class="lvl-2"><a href="#step-4-customize-the-four-things-that-matter">Step 4: Customize the four things that matter</a></li><li class="lvl-2"><a href="#step-5-spawn-and-put-it-to-work">Step 5: Spawn, and put it to work</a></li>
+ <li class="lvl-2"><a href="#where-is-the-agent-gallery-and-what-is-in-it">Where is the Agent Gallery, and what is in it?</a></li><li class="lvl-2"><a href="#how-do-you-import-a-hire">How do you import a hire?</a></li><li class="lvl-2"><a href="#why-read-the-manifest-before-you-spawn">Why read the manifest before you spawn?</a></li><li class="lvl-2"><a href="#what-should-you-customise-before-spawning">What should you customise before spawning?</a></li><li class="lvl-2"><a href="#what-happens-after-you-spawn">What happens after you spawn?</a></li><li class="lvl-2"><a href="#can-you-add-your-own-hire">Can you add your own hire?</a></li>
</ol>
</aside>
@@ -275,13 +299,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/how-to-install-and-use-munder-difflin/index.html b/docs/blog/how-to-install-and-use-munder-difflin/index.html
index 91902e13..5dce61a3 100644
--- a/docs/blog/how-to-install-and-use-munder-difflin/index.html
+++ b/docs/blog/how-to-install-and-use-munder-difflin/index.html
@@ -4,7 +4,7 @@
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>How to Install and Use Munder Difflin: A Beginner&#39;s Guide — Munder Difflin Blog</title>
-<meta name="description" content="A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux." />
+<meta name="description" content="A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux." />
<link rel="canonical" href="https://munderdiffl.in/blog/how-to-install-and-use-munder-difflin/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -15,7 +15,7 @@
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
<meta property="og:title" content="How to Install and Use Munder Difflin: A Beginner&#39;s Guide — Munder Difflin Blog" />
-<meta property="og:description" content="A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux." />
+<meta property="og:description" content="A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux." />
<meta property="og:url" content="https://munderdiffl.in/blog/how-to-install-and-use-munder-difflin/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -31,7 +31,7 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="How to Install and Use Munder Difflin: A Beginner&#39;s Guide — Munder Difflin Blog" />
-<meta name="twitter:description" content="A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux." />
+<meta name="twitter:description" content="A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -45,10 +45,10 @@
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "How to Install and Use Munder Difflin: A Beginner's Guide",
- "description": "A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you'll ever run, and how to install and set it up on macOS, Windows or Linux.",
+ "description": "A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you'll ever run, and how to install and set it up on macOS, Windows or Linux.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-06-05T00:00:00.000Z",
- "dateModified": "2026-08-27T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -80,9 +80,9 @@
"@type": "FAQPage",
"mainEntity": [
{ "@type": "Question", "name": "Do I need to know how to code to use Munder Difflin?", "acceptedAnswer": { "@type": "Answer", "text": "No. The onboarding asks whether you are technical or not on its very first screen, and the non-technical path replaces every piece of jargon with plain language. You will open a terminal exactly once, to install and log into your AI engine, and you can copy and paste those commands." } },
- { "@type": "Question", "name": "Does Munder Difflin cost anything?", "acceptedAnswer": { "@type": "Answer", "text": "The app is free and open source. What you may pay for is the AI engine behind it. Antigravity is available at no charge and OpenCode has a free path, so you can run the whole thing without paying anything. If you already pay for ChatGPT or Claude, you can use those subscriptions instead." } },
+ { "@type": "Question", "name": "Does Munder Difflin cost anything?", "acceptedAnswer": { "@type": "Answer", "text": "The classic app is free and open source, with unlimited local agents. Pro and Teams are optional paid plans with a 14 day trial. What you may pay for is the AI engine behind it. Antigravity is available at no charge and OpenCode has a free path, so you can run the whole thing without paying anything. If you already pay for ChatGPT or Claude, you can use those subscriptions instead." } },
{ "@type": "Question", "name": "Can these agents really change files on my computer?", "acceptedAnswer": { "@type": "Answer", "text": "Yes, and that is the point, so it is worth understanding before you start. An agent can read files, write files and run commands in the folders you give it. You choose how much freedom it has during setup, and you can set it to ask permission before every change." } },
- { "@type": "Question", "name": "Which operating systems does Munder Difflin run on?", "acceptedAnswer": { "@type": "Answer", "text": "macOS, Windows and Linux. macOS ships as a .dmg, Windows as an installer and a portable build, and Linux as an AppImage and a .deb." } },
+ { "@type": "Question", "name": "Which operating systems does Munder Difflin run on?", "acceptedAnswer": { "@type": "Answer", "text": "macOS, Windows and Linux. macOS ships as one universal .dmg for Apple Silicon and Intel, Windows 10 and 11 as a setup installer or a portable build, and Linux as an AppImage." } },
{ "@type": "Question", "name": "Is my code or data sent anywhere?", "acceptedAnswer": { "@type": "Answer", "text": "The app runs on your machine and stores its files there. Your prompts and the files an agent reads do go to whichever AI engine you picked, the same as if you used that tool directly. Anonymous usage telemetry is opt-out and every event is listed publicly in TELEMETRY.md." } }
]
@@ -153,7 +153,7 @@
<a href="/blog/topics/guides/">Guides</a>
</nav>
<h1>How to Install and Use Munder Difflin: A Beginner&#39;s Guide</h1>
- <p class="dek">A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
+ <p class="dek">A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
@@ -182,7 +182,8 @@ not. It starts with what these things actually are, then walks the whole way to
working on something you asked for. One of them is your clone, the boss of the floor, who takes what
you ask for, breaks it into jobs, and hands those jobs to the others. You talk to your clone, and
your clone manages everyone else.</p>
-<p>The app is free and open source. It runs on your machine, and it keeps its files on your machine.</p>
+<p>The app is free and open source. It runs on your machine, and it keeps its files on your machine. There are
+optional paid plans, Pro and Teams, but nothing in this guide needs them.</p>
<p>The thing it is not: a chatbot. You are not sitting there typing and waiting for replies. You give
your clone a job, close the laptop, and come back to work that was done while you were gone.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/how-to-install-and-use-munder-difflin/note-1.png" alt="Hand-drawn sketch from “How to Install and Use Munder Difflin”" loading="lazy" decoding="async" /></figure>
@@ -421,14 +422,16 @@ and every agent will still fail the moment it starts.</p></div>
straight from the
<a href="https://github.com/chaitanyagiri/munder-difflin/releases/latest">latest release on GitHub</a>.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/how-to-install-and-use-munder-difflin/shot-download.png" alt="The Munder Difflin download page showing builds for macOS, Windows and Linux" loading="lazy" decoding="async" /><figcaption>The download page. Pick the file that matches your computer.</figcaption></figure>
-<p><strong>macOS.</strong> Download the <code>.dmg</code>. Open it, drag Munder Difflin into Applications. The first time you
-open it, macOS may say it cannot verify the developer. Right-click the app icon, choose Open, then
-click Open in the dialog. You only do this once.</p>
-<p><strong>Windows.</strong> Download the <code>.exe</code> installer and run it. Windows SmartScreen will likely warn you,
-because the installer is not yet EV code signed. Click More info, then Run anyway. If the installer
-fails to start at all, download the portable build instead, which needs no installation.</p>
-<p><strong>Linux.</strong> Download the <code>AppImage</code>, make it executable (<code>chmod +x</code> on the file) and run it. There is
-a <code>.deb</code> if you prefer to install it properly.</p>
+<p><strong>macOS.</strong> Download the universal <code>.dmg</code>, which runs on Apple Silicon and Intel. Open it and drag
+Munder Difflin into Applications. The app is signed and notarized by Apple, so the first launch only
+shows the usual question about opening an app downloaded from the internet. Click Open.</p>
+<p><strong>Windows.</strong> Download the setup <code>.exe</code> and run it. If Windows SmartScreen shows a warning, click
+More info, then Run anyway. If the installer will not start at all, download the portable <code>.exe</code>
+from the release instead, which needs no installation.</p>
+<p><strong>Linux.</strong> Download the <code>.AppImage</code>, make it executable (<code>chmod +x</code> on the file) and run it.</p>
+<p><strong>Want to be sure the file is genuine?</strong> Every release carries a <code>SHA256SUMS.txt</code>. Run
+<code>shasum -a 256</code> on the file you downloaded (or <code>Get-FileHash</code> in Windows PowerShell) and check that
+the value matches that file’s line.</p>
<h2 id="step-4-onboarding-six-screens" tabindex="-1">Step 4: Onboarding, six screens <a class="anchor" href="#step-4-onboarding-six-screens" aria-hidden="true">#</a></h2>
<p>Open the app. It walks you through setup. There are two intro screens and then four numbered steps.</p>
<h3 id="screen-1-who-are-you" tabindex="-1">Screen 1: Who are you? <a class="anchor" href="#screen-1-who-are-you" aria-hidden="true">#</a></h3>
@@ -518,10 +521,11 @@ it alone. The default is sensible and most people never touch it.</p>
<p>Seven sections. These are the ones that matter early:</p>
<ul>
<li><strong>General.</strong> Change your home folder, switch between technical and plain language, pick your
-language. As of v0.4.6 the app runs in English, Chinese and Arabic.</li>
+language. The app runs in English, Simplified Chinese and Arabic.</li>
<li><strong>Prerequisites.</strong> Checks whether your engines are actually installed and logged in. <strong>This is the
first place to look when something is not working.</strong></li>
-<li><strong>Agents &amp; Models.</strong> Which engines are available and which models each uses.</li>
+<li><strong>Agents &amp; Models.</strong> Which engines are available and which models each uses. New models show up
+without an app update, because the model pickers read the latest list when the app launches.</li>
<li><strong>Autonomy &amp; Budgets.</strong> The permission choice from onboarding, plus spending limits per agent.
Worth setting a cap before leaving anything running overnight.</li>
<li><strong>Connections.</strong> Slack, webhooks, MCP and the REST API.</li>
@@ -540,9 +544,10 @@ Open Settings, then Prerequisites. Or run the engine’s command once in a termi
sign-in.</p>
<p><strong>“Engine not installed” during onboarding, but you installed it.</strong> The app looks for the command on
your system path. Close the app completely and reopen it, since it reads your environment at launch.</p>
-<p><strong>Windows blocks the installer.</strong> SmartScreen flags it because the installer is not EV code signed.
-Choose More info, then Run anyway, or use the portable build.</p>
-<p><strong>macOS says the developer cannot be verified.</strong> Right-click the app, choose Open, then Open again.</p>
+<p><strong>Windows blocks the installer.</strong> If SmartScreen flags it, choose More info, then Run anyway, or use
+the portable build.</p>
+<p><strong>macOS asks whether to open an app downloaded from the internet.</strong> That is normal for anything you
+download. Click Open.</p>
<p><strong>Nothing is happening.</strong> Check whether message delivery is paused in the Command Center. Paused
means queued work is being held for every agent, and nothing is lost when you switch it back on.</p>
<h2 id="the-short-version" tabindex="-1">The short version <a class="anchor" href="#the-short-version" aria-hidden="true">#</a></h2>
@@ -571,7 +576,7 @@ stuck, the <a href="https://munderdiffl.in">Discord</a> is the fastest place to
<div class="qa">
<p class="q">Does Munder Difflin cost anything?</p>
- <p class="a">The app is free and open source. What you may pay for is the AI engine behind it. Antigravity is available at no charge and OpenCode has a free path, so you can run the whole thing without paying anything. If you already pay for ChatGPT or Claude, you can use those subscriptions instead.</p>
+ <p class="a">The classic app is free and open source, with unlimited local agents. Pro and Teams are optional paid plans with a 14 day trial. What you may pay for is the AI engine behind it. Antigravity is available at no charge and OpenCode has a free path, so you can run the whole thing without paying anything. If you already pay for ChatGPT or Claude, you can use those subscriptions instead.</p>
</div>
<div class="qa">
@@ -581,7 +586,7 @@ stuck, the <a href="https://munderdiffl.in">Discord</a> is the fastest place to
<div class="qa">
<p class="q">Which operating systems does Munder Difflin run on?</p>
- <p class="a">macOS, Windows and Linux. macOS ships as a .dmg, Windows as an installer and a portable build, and Linux as an AppImage and a .deb.</p>
+ <p class="a">macOS, Windows and Linux. macOS ships as one universal .dmg for Apple Silicon and Intel, Windows 10 and 11 as a setup installer or a portable build, and Linux as an AppImage.</p>
</div>
<div class="qa">
@@ -624,13 +629,13 @@ stuck, the <a href="https://munderdiffl.in">Discord</a> is the fastest place to
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -656,10 +661,10 @@ stuck, the <a href="https://munderdiffl.in">Discord</a> is the fastest place to
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/how-to-run-multiple-claude-code-agents/index.html b/docs/blog/how-to-run-multiple-claude-code-agents/index.html
index a32052ca..9b75eca6 100644
--- a/docs/blog/how-to-run-multiple-claude-code-agents/index.html
+++ b/docs/blog/how-to-run-multiple-claude-code-agents/index.html
@@ -256,13 +256,13 @@ free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -288,10 +288,10 @@ free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/how-to-talk-to-your-orchestrator/index.html b/docs/blog/how-to-talk-to-your-orchestrator/index.html
index cb88bd8e..fa8451d2 100644
--- a/docs/blog/how-to-talk-to-your-orchestrator/index.html
+++ b/docs/blog/how-to-talk-to-your-orchestrator/index.html
@@ -279,13 +279,13 @@ Three goals in a trench coat. Split them — the task board handles dependencies
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -311,10 +311,10 @@ Three goals in a trench coat. Split them — the task board handles dependencies
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/how-to-use-the-built-in-monaco-ide/index.html b/docs/blog/how-to-use-the-built-in-monaco-ide/index.html
index 7d9589c6..f0d50a3f 100644
--- a/docs/blog/how-to-use-the-built-in-monaco-ide/index.html
+++ b/docs/blog/how-to-use-the-built-in-monaco-ide/index.html
@@ -277,13 +277,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -309,10 +309,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/human-in-the-loop-approving-ai-agents/index.html b/docs/blog/human-in-the-loop-approving-ai-agents/index.html
index f408a635..c923b280 100644
--- a/docs/blog/human-in-the-loop-approving-ai-agents/index.html
+++ b/docs/blog/human-in-the-loop-approving-ai-agents/index.html
@@ -312,13 +312,13 @@ you talk to? You can <a href="/#install">download Munder Difflin</a> free; it’
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -344,10 +344,10 @@ you talk to? You can <a href="/#install">download Munder Difflin</a> free; it’
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/index.html b/docs/blog/index.html
index be499ea4..541b93fb 100644
--- a/docs/blog/index.html
+++ b/docs/blog/index.html
@@ -391,13 +391,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
@@ -407,13 +407,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -471,10 +471,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
@@ -663,10 +663,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/your-first-hour-with-munder-difflin/">Your First Hour With Munder Difflin</a></h3>
- <p class="dek">A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</p>
+ <p class="dek">A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/your-first-hour-with-munder-difflin/" aria-label="Read: Your First Hour With Munder Difflin">Read →</a>
@@ -695,10 +695,10 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/run-munder-difflin-on-a-mac-mini/">Run Munder Difflin Locally on a Mac Mini</a></h3>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/run-munder-difflin-on-a-mac-mini/" aria-label="Read: Run Munder Difflin Locally on a Mac Mini">Read →</a>
@@ -711,13 +711,13 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>6 min</span>
</div>
- <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</a></h3>
- <p class="dek">Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</p>
+ <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</a></h3>
+ <p class="dek">Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider">Read →</a>
+ <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open Source Models: Fully Local or Through a Provider">Read →</a>
</div>
</div>
</article>
@@ -791,13 +791,13 @@
<div class="meta">
<time datetime="2026-06-10">Jun 10, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
- <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog-Writer Agent: The One That Wrote This Post</a></h3>
- <p class="dek">How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</p>
+ <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog Writer Agent: The One That Wrote This Post</a></h3>
+ <p class="dek">How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</p>
<div class="foot">
<span class="tag kind">Use Cases</span>
- <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog-Writer Agent: The One That Wrote This Post">Read →</a>
+ <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog Writer Agent: The One That Wrote This Post">Read →</a>
</div>
</div>
</article>
@@ -1034,7 +1034,7 @@
<span>14 min</span>
</div>
<h3><a href="/blog/how-to-install-and-use-munder-difflin/">How to Install and Use Munder Difflin: A Beginner&#39;s Guide</a></h3>
- <p class="dek">A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
+ <p class="dek">A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-install-and-use-munder-difflin/" aria-label="Read: How to Install and Use Munder Difflin: A Beginner&#39;s Guide">Read →</a>
@@ -1146,7 +1146,7 @@
<span>6 min</span>
</div>
<h3><a href="/blog/best-claude-code-multi-agent-tools/">The Best Tools to Run Multiple Claude Code Agents (2026)</a></h3>
- <p class="dek">An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared.</p>
+ <p class="dek">An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
<a class="more" href="/blog/best-claude-code-multi-agent-tools/" aria-label="Read: The Best Tools to Run Multiple Claude Code Agents (2026)">Read →</a>
@@ -2231,13 +2231,13 @@
<div class="meta">
<time datetime="2026-05-22">May 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>2 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? (Plain-English Guide)</a></h3>
- <p class="dek">A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework.</p>
+ <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? A Plain English Guide</a></h3>
+ <p class="dek">A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one.</p>
<div class="foot">
<span class="tag kind">Concepts</span>
- <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? (Plain-English Guide)">Read →</a>
+ <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? A Plain English Guide">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/launching-munder-difflin-v0-3-0/index.html b/docs/blog/launching-munder-difflin-v0-3-0/index.html
index 234b9a86..f10f966d 100644
--- a/docs/blog/launching-munder-difflin-v0-3-0/index.html
+++ b/docs/blog/launching-munder-difflin-v0-3-0/index.html
@@ -299,7 +299,7 @@
<footer class="post-foot wrap"><div class="prevnext">
<a class="pn prev" href="/blog/hire-manifest-untrusted-input/"><span class="k">← Older</span><span class="t">Treating a Hire Manifest as Untrusted Input</span></a>
- <a class="pn next" href="/blog/run-munder-difflin-on-open-models/"><span class="k">Newer →</span><span class="t">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</span></a>
+ <a class="pn next" href="/blog/run-munder-difflin-on-open-models/"><span class="k">Newer →</span><span class="t">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</span></a>
</div>
<section class="related">
<h2>Related in Story</h2>
diff --git a/docs/blog/launching-munder-difflin-v0-3-5/index.html b/docs/blog/launching-munder-difflin-v0-3-5/index.html
index c551eaf8..c942d654 100644
--- a/docs/blog/launching-munder-difflin-v0-3-5/index.html
+++ b/docs/blog/launching-munder-difflin-v0-3-5/index.html
@@ -310,7 +310,7 @@ to help it reach more people.</p>
</div>
<footer class="post-foot wrap"><div class="prevnext">
- <a class="pn prev" href="/blog/qm-vs-munder-difflin/"><span class="k">← Older</span><span class="t">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</span></a>
+ <a class="pn prev" href="/blog/qm-vs-munder-difflin/"><span class="k">← Older</span><span class="t">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</span></a>
<a class="pn next" href="/blog/why-our-auto-update-never-ran/"><span class="k">Newer →</span><span class="t">Our Auto-Update Never Ran Once: A CommonJS Export That Vanished Into an ESM Namespace</span></a>
</div>
<section class="related">
diff --git a/docs/blog/local-first-vs-cloud-agent-sdks/index.html b/docs/blog/local-first-vs-cloud-agent-sdks/index.html
index 7bf39a69..d92e549c 100644
--- a/docs/blog/local-first-vs-cloud-agent-sdks/index.html
+++ b/docs/blog/local-first-vs-cloud-agent-sdks/index.html
@@ -354,13 +354,13 @@ to run agents on your own terms — free and open source.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/manage-multiple-claude-code-sessions/index.html b/docs/blog/manage-multiple-claude-code-sessions/index.html
index 9c2044f5..3d3e5ac3 100644
--- a/docs/blog/manage-multiple-claude-code-sessions/index.html
+++ b/docs/blog/manage-multiple-claude-code-sessions/index.html
@@ -309,13 +309,13 @@ isolated agents, shared memory, and a single committer, all on your own machine.
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -341,10 +341,10 @@ isolated agents, shared memory, and a single committer, all on your own machine.
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/mcp-and-skills-in-a-hive/index.html b/docs/blog/mcp-and-skills-in-a-hive/index.html
index 7af3e8c0..e8c376ea 100644
--- a/docs/blog/mcp-and-skills-in-a-hive/index.html
+++ b/docs/blog/mcp-and-skills-in-a-hive/index.html
@@ -325,13 +325,13 @@ existing setup into a coordinated team of agents; it’s free and open source.</
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -357,10 +357,10 @@ existing setup into a coordinated team of agents; it’s free and open source.</
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/mcp-security-tool-poisoning/index.html b/docs/blog/mcp-security-tool-poisoning/index.html
index e4a37bf5..c96467d7 100644
--- a/docs/blog/mcp-security-tool-poisoning/index.html
+++ b/docs/blog/mcp-security-tool-poisoning/index.html
@@ -314,13 +314,13 @@ execution. Do that and MCP is what it should be: a powerful, <em>bounded</em> wa
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -346,10 +346,10 @@ execution. Do that and MCP is what it should be: a powerful, <em>bounded</em> wa
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/munder-difflin-v0-2-4-feature-walkthrough/index.html b/docs/blog/munder-difflin-v0-2-4-feature-walkthrough/index.html
index 88d07e03..aea7e025 100644
--- a/docs/blog/munder-difflin-v0-2-4-feature-walkthrough/index.html
+++ b/docs/blog/munder-difflin-v0-2-4-feature-walkthrough/index.html
@@ -290,13 +290,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -322,10 +322,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/orca-vs-munder-difflin/index.html b/docs/blog/orca-vs-munder-difflin/index.html
index a12842f7..ce28b712 100644
--- a/docs/blog/orca-vs-munder-difflin/index.html
+++ b/docs/blog/orca-vs-munder-difflin/index.html
@@ -332,13 +332,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/prompt-caching-for-ai-agents/index.html b/docs/blog/prompt-caching-for-ai-agents/index.html
index c26ceb5e..42f2b54d 100644
--- a/docs/blog/prompt-caching-for-ai-agents/index.html
+++ b/docs/blog/prompt-caching-for-ai-agents/index.html
@@ -344,13 +344,13 @@ that doesn’t waste your tokens; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -376,10 +376,10 @@ that doesn’t waste your tokens; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/qm-vs-munder-difflin/index.html b/docs/blog/qm-vs-munder-difflin/index.html
index 52e3c3ac..ca03fa60 100644
--- a/docs/blog/qm-vs-munder-difflin/index.html
+++ b/docs/blog/qm-vs-munder-difflin/index.html
@@ -3,8 +3,8 @@
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
-<title>qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office? — Munder Difflin Blog</title>
-<meta name="description" content="An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE." />
+<title>qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones? — Munder Difflin Blog</title>
+<meta name="description" content="An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them." />
<link rel="canonical" href="https://munderdiffl.in/blog/qm-vs-munder-difflin/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -14,8 +14,8 @@
<!-- Open Graph -->
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
-<meta property="og:title" content="qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office? — Munder Difflin Blog" />
-<meta property="og:description" content="An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE." />
+<meta property="og:title" content="qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones? — Munder Difflin Blog" />
+<meta property="og:description" content="An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them." />
<meta property="og:url" content="https://munderdiffl.in/blog/qm-vs-munder-difflin/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -30,8 +30,8 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
-<meta name="twitter:title" content="qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office? — Munder Difflin Blog" />
-<meta name="twitter:description" content="An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE." />
+<meta name="twitter:title" content="qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones? — Munder Difflin Blog" />
+<meta name="twitter:description" content="An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -44,11 +44,11 @@
{
"@context": "https://schema.org",
"@type": "BlogPosting",
- "headline": "qm vs Munder Difflin: YC's Multiplayer Harness or a Local-First Agent Office?",
- "description": "An honest comparison of YC's qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.",
+ "headline": "qm vs Munder Difflin: YC's Multiplayer Harness or an Office of Your Clones?",
+ "description": "An honest September 2026 comparison of YC's qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-08-06T00:00:00.000Z",
- "dateModified": "2026-08-06T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -56,7 +56,7 @@
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/qm-vs-munder-difflin/" },
- "keywords": "qm vs munder difflin, yc qm alternative, qm claude code, qm agent harness, multiplayer agent harness, local first agent orchestration, qm y combinator agents",
+ "keywords": "qm vs munder difflin, yc qm alternative, qm agent harness, multiplayer agent harness, local first agent orchestration, qm y combinator agents, self hosted agent harness",
"articleSection": "Comparisons",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/qm-vs-munder-difflin/"
@@ -69,7 +69,7 @@
"itemListElement": [
{ "@type": "ListItem", "position": 1, "name": "Home", "item": "https://munderdiffl.in/" },
{ "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://munderdiffl.in/blog/" },
- { "@type": "ListItem", "position": 3, "name": "qm vs Munder Difflin: YC's Multiplayer Harness or a Local-First Agent Office?", "item": "https://munderdiffl.in/blog/qm-vs-munder-difflin/" }
+ { "@type": "ListItem", "position": 3, "name": "qm vs Munder Difflin: YC's Multiplayer Harness or an Office of Your Clones?", "item": "https://munderdiffl.in/blog/qm-vs-munder-difflin/" }
]
}
</script>
@@ -79,12 +79,11 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "What is qm?", "acceptedAnswer": { "@type": "Answer", "text": "qm (github.com/yc-software/qm) is Y Combinator's open-source 'multiplayer agent harness for work' — a headless Node.js core with a web UI and a Slack plugin. Team members get agent 'employees' with isolated workspaces, shared skills with scope-based permissions, background crons and watches, and three security postures (Strict, Auto, Dangerous). It's MIT-licensed and installs via npm." } },
- { "@type": "Question", "name": "What's the main difference between qm and Munder Difflin?", "acceptedAnswer": { "@type": "Answer", "text": "Where the harness lives and what you can see. qm is multiplayer-first: a shared server your whole team reaches through Slack and the browser. Munder Difflin is local-first: a desktop app on your machine where every agent is a real CLI process in a real terminal you can watch, with a visual office floor, voice orchestration, and a built-in IDE. They cover the same core jobs — postures, scheduled work, Slack, skills — from opposite ends." } },
- { "@type": "Question", "name": "Does Munder Difflin do what qm's security postures do?", "acceptedAnswer": { "@type": "Answer", "text": "Yes, the same job with different controls: a floor-wide auto-mode toggle (the equivalent of switching between approval-gated and autonomous), per-agent pause/halt, voice-level confirm words for destructive actions, tool gating, a circuit breaker for runaway agents, and spend/scope escalation to a human. qm's Strict/Auto/Dangerous is org-level configuration; Munder Difflin's controls are per-floor and per-agent." } },
- { "@type": "Question", "name": "Does qm have voice control, an IDE, or git visualization?", "acceptedAnswer": { "@type": "Answer", "text": "Not as of this writing. qm's interfaces are Slack and a web UI. Munder Difflin ships realtime voice orchestration (a talking Michael with live floor context), a built-in Monaco IDE with commit history, branch compare and side-by-side diffs, markdown previews, real watchable terminals, and auto-update." } },
- { "@type": "Question", "name": "Which one should I use?", "acceptedAnswer": { "@type": "Answer", "text": "Use qm if your whole team wants one shared harness in Slack with org-level admin control. Use Munder Difflin if you want your agent office on your own machine — watchable, local-first, with voice and an IDE — riding the CLI subscriptions you already pay for. Both are MIT-licensed; some teams will genuinely want both: qm as the shared org layer, Munder Difflin as the personal floor." } },
- { "@type": "Question", "name": "Is Munder Difflin older than qm?", "acceptedAnswer": { "@type": "Answer", "text": "Munder Difflin has been shipping public releases since v0.2.0 in mid-2026 and is on v0.3.5 with a stable auto-updating release train, while qm is a newer project that grew popular quickly. Popularity and maturity are different axes — evaluate both against your workflow." } }
+ { "@type": "Question", "name": "What is qm?", "acceptedAnswer": { "@type": "Answer", "text": "qm (github.com/yc-software/qm) is Y Combinator's open source multiplayer agent harness for work, used in Slack and on the web. A company deploys it from its own deployment repository. Each person and each room gets scoped memory, files, permissions, crons and a durable sandbox, and Pi, OpenCode, Codex and Claude Code can all drive the same core. It is MIT licensed." } },
+ { "@type": "Question", "name": "What is the main difference between qm and Munder Difflin?", "acceptedAnswer": { "@type": "Answer", "text": "Where the agents run and who they belong to. qm is a shared service a company deploys, and everyone reaches it through Slack and a browser. Munder Difflin is a desktop app where your agents run as real terminal processes on your own machine, coordinated by your clone, Michael." } },
+ { "@type": "Question", "name": "Can Munder Difflin be multiplayer like qm?", "acceptedAnswer": { "@type": "Answer", "text": "Yes, in a different shape. The Munder Difflin Teams plan lets clones message each other, and every message is sealed for the device that opens it. Each person's clone still runs on their own machine under their own keys, and the relay carries messages it cannot read." } },
+ { "@type": "Question", "name": "How do their safety controls compare?", "acceptedAnswer": { "@type": "Answer", "text": "qm has one security posture for the whole org: Strict pauses harness tool calls for human approval, Auto screens external data with a classifier, and Dangerous does neither. Munder Difflin sets autonomy on your floor: an ask first mode where agents pause for tool approval, per agent token budgets, a circuit breaker that steers, constrains and then stops a runaway agent, and an ASK ME board for decisions that need you." } },
+ { "@type": "Question", "name": "Which should I use?", "acceptedAnswer": { "@type": "Answer", "text": "qm if your company wants one shared agent service in Slack, run by an admin on your own infrastructure. Munder Difflin if you want agents you can watch in real terminals on your own machine, across twelve CLIs, on the subscriptions you already pay for. Both are MIT licensed, so trying both costs you an afternoon." } }
]
}
@@ -153,15 +152,15 @@
<span>/</span>
<a href="/blog/topics/comparisons/">Comparisons</a>
</nav>
- <h1>qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</h1>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h1>qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</h1>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true">·</span>
- <span>3 min read</span>
+ <span>4 min read</span>
<span class="tag kind">Comparisons</span>
</div>
</header>
@@ -170,32 +169,28 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p><strong>qm</strong> is YC's
-multiplayer agent harness — headless, org-shaped, living in <strong>Slack and a web
-UI</strong>, great when a whole team shares one fleet. <strong>Munder Difflin</strong> is
-the same idea grown from the other end: a <strong>local-first desktop office</strong> where
-every agent is a real CLI process in a terminal you can watch, with <strong>voice
-orchestration, a built-in IDE with git history, and auto-update</strong> — things a
-headless harness doesn't have. Same jobs, opposite ends of the wire. Both MIT-licensed.</p></div>
-<p><a href="https://github.com/yc-software/qm">qm</a> earned its stars fast — “a multiplayer agent
-harness for work, in Slack and on the web” is a great pitch, and Y Combinator shipping
-open-source agent tooling is good for everyone. We get asked about it a lot, so here’s the
-honest comparison.</p>
-<h2 id="the-same-problem-from-opposite-ends" tabindex="-1">The same problem, from opposite ends <a class="anchor" href="#the-same-problem-from-opposite-ends" aria-hidden="true">#</a></h2>
-<p>Both tools exist because one CLI agent in one terminal doesn’t scale: you want several
-agents working in parallel, safely, with scheduled background work and a way to reach them
-from where you already are.</p>
-<p><strong>qm answers it org-first.</strong> A headless Node core (installed via <code>npm exec qm init</code>) that
-your team reaches through Slack and a web UI. Each “employee” gets an isolated workspace;
-skills are shared with scope-based permissions and can be promoted org-wide; admins set the
-security posture — <strong>Strict</strong> (approval-gated), <strong>Auto</strong> (default screening), or
-<strong>Dangerous</strong> — for everyone at once. Background work runs as <strong>crons and watches</strong>.</p>
-<p><strong>Munder Difflin answers it desk-first.</strong> A desktop app on your machine where every agent
-is a <strong>real CLI process in a real terminal</strong> — Claude Code, Codex, Antigravity, Copilot,
-Grok, Kimi, and more, riding the subscriptions you already pay for. The floor is visual
-(yes, it looks like The Office), a GOD orchestrator routes work, agents share long-term
-memory, and background work runs as <strong>scheduled missions</strong> with Slack and webhook triggers.</p>
-<p>If you map the concepts across, most of qm’s vocabulary has a Munder Difflin counterpart:</p>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p><strong>qm</strong> is Y Combinator's
+multiplayer agent harness. A company deploys it, and everyone works with agents through <strong>Slack and a
+web app</strong>, each person and room with its own scoped memory, crons and sandbox. <strong>Munder
+Difflin</strong> runs an <strong>office of your clones on your own machine</strong>: real terminal CLIs you
+can watch, coordinated by Michael, with <strong>Teams</strong> so your clone can work with your teammates'
+clones. Same destination, opposite starting points. Both MIT licensed.</p></div>
+<p><a href="https://github.com/yc-software/qm">qm</a> keeps picking up stars, and people keep asking how it compares. We
+wrote the first version of this in August. Both projects have shipped a lot since, so this is the updated
+comparison, checked against qm’s README and our own 0.5.2 release on 10 September 2026.</p>
+<h2 id="what-problem-do-both-tools-solve" tabindex="-1">What problem do both tools solve? <a class="anchor" href="#what-problem-do-both-tools-solve" aria-hidden="true">#</a></h2>
+<p>One agent in one terminal does not scale to real work. Both tools let several agents work in parallel, safely,
+on schedules, and reachable from where you already are. They just start from opposite ends.</p>
+<p><strong>qm starts from the company.</strong> You create a deployment repository that depends on <code>@yc-software/qm</code>, and the
+qm CLI validates and deploys it. A headless core backed by Postgres runs the sessions, and people reach it
+through Slack and a web app. Each person and each room gets its own scoped memory, files, keychain view,
+permissions, crons, web apps and durable sandbox. Admins choose which harnesses and models are available, and
+Pi, OpenCode, Codex and Claude Code can all drive the core.</p>
+<p><strong>Munder Difflin starts from your desk.</strong> It is a desktop app for macOS, Windows and Linux. Every agent is a
+real CLI process in a real terminal on your machine, running on the subscriptions you already pay for, and
+twelve CLIs are supported. The office floor shows who is doing what. Michael, your clone, turns requests into
+tasks and routes the work. Agents share long term memory and message each other through mailboxes.</p>
+<h2 id="how-do-the-concepts-map-across" tabindex="-1">How do the concepts map across? <a class="anchor" href="#how-do-the-concepts-map-across" aria-hidden="true">#</a></h2>
<div class="table-scroll">
<table>
<thead>
@@ -207,29 +202,39 @@ memory, and background work runs as <strong>scheduled missions</strong> with Sla
</thead>
<tbody>
<tr>
-<td>Safety levels</td>
-<td>Strict / Auto / Dangerous postures</td>
-<td>Auto-mode toggle, per-agent pause/halt, tool gating, circuit breaker, confirm-word voice safety</td>
+<td>Safety</td>
+<td>One org posture: Strict, Auto or Dangerous</td>
+<td>Ask first or autonomous agents, per agent token budgets, a circuit breaker, an ASK ME board</td>
</tr>
<tr>
<td>Background work</td>
-<td>Crons and watches</td>
-<td>Scheduled missions + Slack/webhook triggers</td>
+<td>Crons, watches, inbound webhooks</td>
+<td>Scheduled triggers on an interval or on chosen weekdays, webhooks, Slack</td>
</tr>
<tr>
-<td>Reusable behaviors</td>
-<td>Shared skills with scoped permissions</td>
-<td>Agent Gallery roles + shareable agent recipes</td>
+<td>Reusable behaviour</td>
+<td>Skills owned by a scope, shared by grant, skill packs from git</td>
+<td>A skills catalog, plus ready made hires from the Agent Gallery</td>
</tr>
<tr>
-<td>Reach it from anywhere</td>
-<td>Slack + web UI</td>
-<td>Slack + Remote Control from your phone</td>
+<td>Isolation</td>
+<td>A durable sandbox per person and room</td>
+<td>A git worktree per agent when isolation is on</td>
</tr>
<tr>
-<td>Isolation</td>
-<td>Isolated workspaces per employee</td>
-<td>Isolated git worktrees per agent</td>
+<td>Where you use it</td>
+<td>Slack and a web app</td>
+<td>A desktop app, plus Slack</td>
+</tr>
+<tr>
+<td>Multiplayer</td>
+<td>Built in: one deployment for the company</td>
+<td>Teams plan: clones message each other, sealed per device</td>
+</tr>
+<tr>
+<td>Where it runs</td>
+<td>Your company’s infrastructure</td>
+<td>Your own machine</td>
</tr>
<tr>
<td>License</td>
@@ -240,46 +245,38 @@ memory, and background work runs as <strong>scheduled missions</strong> with Sla
</table>
</div>
<figure class="fig fig-inline"><img src="/blog/assets/media/qm-vs-munder-difflin/note-1.png" alt="Hand-drawn sketch from “qm vs Munder Difflin: YC's Multiplayer Harness or a Local-First Agent Office?”" loading="lazy" decoding="async" /></figure>
-<h2 id="what-each-has-that-the-other-doesnt" tabindex="-1">What each has that the other doesn’t <a class="anchor" href="#what-each-has-that-the-other-doesnt" aria-hidden="true">#</a></h2>
-<p><strong>qm’s edge is multiplayer.</strong> One shared harness, unified identity, org-level admin,
-skills promotion across scopes. If your team wants a <em>single</em> fleet everyone shares from
-Slack, that’s qm’s home turf, and Munder Difflin doesn’t try to be that.</p>
-<p><strong>Munder Difflin’s edge is everything you can see and touch:</strong></p>
+<h2 id="what-does-qm-do-that-munder-difflin-does-not" tabindex="-1">What does qm do that Munder Difflin does not? <a class="anchor" href="#what-does-qm-do-that-munder-difflin-does-not" aria-hidden="true">#</a></h2>
+<p>Company wide sharing, run by an admin. That is qm’s home turf. One deployment serves everyone, the admin
+controls which harnesses and models people can use, the whole company inherits one security posture, and
+people can build internal web apps and publish them to each other. A sandbox per scope also means the tools
+someone installs stay installed. If your company wants a single agent service living in Slack, qm was built
+for exactly that.</p>
+<h2 id="what-does-munder-difflin-do-differently" tabindex="-1">What does Munder Difflin do differently? <a class="anchor" href="#what-does-munder-difflin-do-differently" aria-hidden="true">#</a></h2>
+<p>It puts everything on your own machine, where you can watch and touch it:</p>
<ul>
-<li><strong>Real terminals.</strong> Every agent is a watchable CLI process — when something goes
-sideways, you read the actual session, not a log abstraction.</li>
-<li><strong>Voice orchestration.</strong> Flip on talk mode and Michael greets you already knowing every
-agent’s status, steers work, changes settings from an allowlist, and waits for a spoken
-confirm word before anything destructive.</li>
-<li><strong>A built-in IDE.</strong> Monaco (the VS Code engine) one click over the floor: side-by-side
-diffs of agent changes, clickable commit history, branch compare, guarded checkout, live
-markdown previews.</li>
-<li><strong>Local-first by construction.</strong> No shared server, no code upload — agents, state, and
-memory live on your machine.</li>
-<li><strong>Auto-update.</strong> The app ships its own updates from GitHub releases; you just click
-“restart to update.”</li>
-<li><strong>A longer release train.</strong> Public releases since v0.2.0, now at v0.3.5, with the
-reliability work battle-tested by a growing contributor community.</li>
+<li><strong>Real terminals.</strong> Every agent is a CLI process you can open and type into. When something goes sideways,
+you read the actual session.</li>
+<li><strong>Twelve CLIs on your own subscriptions.</strong> Claude Code, Codex, Gemini CLI, Copilot, Cursor and seven more,
+mixed on one floor, within the usage limits you already pay for.</li>
+<li><strong>Your clone in charge.</strong> Michael takes your request, hires workers and routes messages, and you can talk
+to him by voice.</li>
+<li><strong>An IDE over the floor.</strong> Browse and edit an agent’s workspace and see its uncommitted changes as a diff.</li>
+<li><strong>Nothing to deploy.</strong> Download it, answer a short setup, give Michael a job. No server, no database.</li>
+<li><strong>Local first.</strong> Code, keys and personal context stay on your machine. With Teams, the only thing that
+travels is clone to clone messages, sealed so the relay in the middle cannot read them.</li>
</ul>
<figure class="fig fig-inline"><img src="/blog/assets/media/qm-vs-munder-difflin/note-2.png" alt="Hand-drawn sketch from “qm vs Munder Difflin: YC's Multiplayer Harness or a Local-First Agent Office?”" loading="lazy" decoding="async" /></figure>
-<h2 id="which-should-you-use" tabindex="-1">Which should you use? <a class="anchor" href="#which-should-you-use" aria-hidden="true">#</a></h2>
+<h2 id="which-one-should-you-use" tabindex="-1">Which one should you use? <a class="anchor" href="#which-one-should-you-use" aria-hidden="true">#</a></h2>
<ul>
-<li><strong>Your whole team wants one shared fleet with org-level control → qm.</strong> That’s what
-multiplayer-first buys you.</li>
-<li><strong>You want your own agent office, on your own machine, with everything visible → Munder
-Difflin.</strong> Simpler to adopt (download, open, hire), nothing leaves your computer, and
-the interfaces — floor, voice, IDE — are things a headless harness structurally can’t
-offer.</li>
-<li><strong>Honestly? Some teams will run both.</strong> qm as the shared org layer, Munder Difflin as
-the personal floor where you watch and steer your own agents. They don’t compete for
-the same seat.</li>
+<li><strong>Your company wants one shared agent service in Slack, with admin control:</strong> qm.</li>
+<li><strong>You want your own office of agents on your machine, visible and vendor neutral:</strong> Munder Difflin. Add
+Teams when you want your clone working with your teammates’ clones.</li>
+<li><strong>Some companies will run both.</strong> qm as the company layer in Slack, and Munder Difflin as the personal floor
+where each person watches and steers their own agents. They do not compete for the same seat.</li>
</ul>
-<p>Facts about qm above come from its public README as of August 2026 — check
-<a href="https://github.com/yc-software/qm">the repo</a> for the current state, because both projects
-move fast.</p>
-<p><strong><a href="https://munderdiffl.in/">Try Munder Difflin free</a></strong> — macOS, Windows, Linux, MIT-licensed.
-And if this comparison helped, a <a href="https://github.com/chaitanyagiri/munder-difflin">GitHub star</a>
-helps more people find it.</p>
+<p>Facts about qm above come from its public README as of 10 September 2026. Check
+<a href="https://github.com/yc-software/qm">the repo</a> for the current state, because both projects ship often.</p>
+<p><strong><a href="https://munderdiffl.in/">Download Munder Difflin free</a></strong> for macOS, Windows or Linux. MIT licensed.</p>
@@ -288,32 +285,27 @@ helps more people find it.</p>
<div class="qa">
<p class="q">What is qm?</p>
- <p class="a">qm (github.com/yc-software/qm) is Y Combinator&#39;s open-source &#39;multiplayer agent harness for work&#39; — a headless Node.js core with a web UI and a Slack plugin. Team members get agent &#39;employees&#39; with isolated workspaces, shared skills with scope-based permissions, background crons and watches, and three security postures (Strict, Auto, Dangerous). It&#39;s MIT-licensed and installs via npm.</p>
- </div>
-
- <div class="qa">
- <p class="q">What&#39;s the main difference between qm and Munder Difflin?</p>
- <p class="a">Where the harness lives and what you can see. qm is multiplayer-first: a shared server your whole team reaches through Slack and the browser. Munder Difflin is local-first: a desktop app on your machine where every agent is a real CLI process in a real terminal you can watch, with a visual office floor, voice orchestration, and a built-in IDE. They cover the same core jobs — postures, scheduled work, Slack, skills — from opposite ends.</p>
+ <p class="a">qm (github.com/yc-software/qm) is Y Combinator&#39;s open source multiplayer agent harness for work, used in Slack and on the web. A company deploys it from its own deployment repository. Each person and each room gets scoped memory, files, permissions, crons and a durable sandbox, and Pi, OpenCode, Codex and Claude Code can all drive the same core. It is MIT licensed.</p>
</div>
<div class="qa">
- <p class="q">Does Munder Difflin do what qm&#39;s security postures do?</p>
- <p class="a">Yes, the same job with different controls: a floor-wide auto-mode toggle (the equivalent of switching between approval-gated and autonomous), per-agent pause/halt, voice-level confirm words for destructive actions, tool gating, a circuit breaker for runaway agents, and spend/scope escalation to a human. qm&#39;s Strict/Auto/Dangerous is org-level configuration; Munder Difflin&#39;s controls are per-floor and per-agent.</p>
+ <p class="q">What is the main difference between qm and Munder Difflin?</p>
+ <p class="a">Where the agents run and who they belong to. qm is a shared service a company deploys, and everyone reaches it through Slack and a browser. Munder Difflin is a desktop app where your agents run as real terminal processes on your own machine, coordinated by your clone, Michael.</p>
</div>
<div class="qa">
- <p class="q">Does qm have voice control, an IDE, or git visualization?</p>
- <p class="a">Not as of this writing. qm&#39;s interfaces are Slack and a web UI. Munder Difflin ships realtime voice orchestration (a talking Michael with live floor context), a built-in Monaco IDE with commit history, branch compare and side-by-side diffs, markdown previews, real watchable terminals, and auto-update.</p>
+ <p class="q">Can Munder Difflin be multiplayer like qm?</p>
+ <p class="a">Yes, in a different shape. The Munder Difflin Teams plan lets clones message each other, and every message is sealed for the device that opens it. Each person&#39;s clone still runs on their own machine under their own keys, and the relay carries messages it cannot read.</p>
</div>
<div class="qa">
- <p class="q">Which one should I use?</p>
- <p class="a">Use qm if your whole team wants one shared harness in Slack with org-level admin control. Use Munder Difflin if you want your agent office on your own machine — watchable, local-first, with voice and an IDE — riding the CLI subscriptions you already pay for. Both are MIT-licensed; some teams will genuinely want both: qm as the shared org layer, Munder Difflin as the personal floor.</p>
+ <p class="q">How do their safety controls compare?</p>
+ <p class="a">qm has one security posture for the whole org: Strict pauses harness tool calls for human approval, Auto screens external data with a classifier, and Dangerous does neither. Munder Difflin sets autonomy on your floor: an ask first mode where agents pause for tool approval, per agent token budgets, a circuit breaker that steers, constrains and then stops a runaway agent, and an ASK ME board for decisions that need you.</p>
</div>
<div class="qa">
- <p class="q">Is Munder Difflin older than qm?</p>
- <p class="a">Munder Difflin has been shipping public releases since v0.2.0 in mid-2026 and is on v0.3.5 with a stable auto-updating release train, while qm is a newer project that grew popular quickly. Popularity and maturity are different axes — evaluate both against your workflow.</p>
+ <p class="q">Which should I use?</p>
+ <p class="a">qm if your company wants one shared agent service in Slack, run by an admin on your own infrastructure. Munder Difflin if you want agents you can watch in real terminals on your own machine, across twelve CLIs, on the subscriptions you already pay for. Both are MIT licensed, so trying both costs you an afternoon.</p>
</div>
</section>
@@ -332,14 +324,14 @@ helps more people find it.</p>
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#the-same-problem-from-opposite-ends">The same problem, from opposite ends</a></li><li class="lvl-2"><a href="#what-each-has-that-the-other-doesnt">What each has that the other doesn’t</a></li><li class="lvl-2"><a href="#which-should-you-use">Which should you use?</a></li>
+ <li class="lvl-2"><a href="#what-problem-do-both-tools-solve">What problem do both tools solve?</a></li><li class="lvl-2"><a href="#how-do-the-concepts-map-across">How do the concepts map across?</a></li><li class="lvl-2"><a href="#what-does-qm-do-that-munder-difflin-does-not">What does qm do that Munder Difflin does not?</a></li><li class="lvl-2"><a href="#what-does-munder-difflin-do-differently">What does Munder Difflin do differently?</a></li><li class="lvl-2"><a href="#which-one-should-you-use">Which one should you use?</a></li>
</ol>
</aside>
</div>
<footer class="post-foot wrap"><div class="prevnext">
- <a class="pn prev" href="/blog/command-center-guide/"><span class="k">← Older</span><span class="t">The Command Center: Kanban, Fleet, and Budgets in One Place</span></a>
+ <a class="pn prev" href="/blog/command-center-guide/"><span class="k">← Older</span><span class="t">The Command Center: Kanban, Fleet and Budgets in One Place</span></a>
<a class="pn next" href="/blog/launching-munder-difflin-v0-3-5/"><span class="k">Newer →</span><span class="t">Launching Munder Difflin v0.3.5: A Michael Who Knows the Floor, a Git Time-Machine, and an App That Updates Itself</span></a>
</div>
<section class="related">
diff --git a/docs/blog/run-a-mixed-engine-office/index.html b/docs/blog/run-a-mixed-engine-office/index.html
index 341674d8..8244dd06 100644
--- a/docs/blog/run-a-mixed-engine-office/index.html
+++ b/docs/blog/run-a-mixed-engine-office/index.html
@@ -272,13 +272,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -304,10 +304,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/run-ai-agent-hive-from-slack-setup/index.html b/docs/blog/run-ai-agent-hive-from-slack-setup/index.html
index 64825e2c..9f3ea4e5 100644
--- a/docs/blog/run-ai-agent-hive-from-slack-setup/index.html
+++ b/docs/blog/run-ai-agent-hive-from-slack-setup/index.html
@@ -370,13 +370,13 @@ it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -402,10 +402,10 @@ it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/run-munder-difflin-on-a-mac-mini/index.html b/docs/blog/run-munder-difflin-on-a-mac-mini/index.html
index a8b85c2a..3f8abb36 100644
--- a/docs/blog/run-munder-difflin-on-a-mac-mini/index.html
+++ b/docs/blog/run-munder-difflin-on-a-mac-mini/index.html
@@ -4,7 +4,7 @@
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Run Munder Difflin Locally on a Mac Mini — Munder Difflin Blog</title>
-<meta name="description" content="A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint." />
+<meta name="description" content="Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini." />
<link rel="canonical" href="https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -15,7 +15,7 @@
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
<meta property="og:title" content="Run Munder Difflin Locally on a Mac Mini — Munder Difflin Blog" />
-<meta property="og:description" content="A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint." />
+<meta property="og:description" content="Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini." />
<meta property="og:url" content="https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -32,7 +32,7 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="Run Munder Difflin Locally on a Mac Mini — Munder Difflin Blog" />
-<meta name="twitter:description" content="A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint." />
+<meta name="twitter:description" content="Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -46,10 +46,10 @@
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "Run Munder Difflin Locally on a Mac Mini",
- "description": "A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.",
+ "description": "Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-06-22T00:00:00.000Z",
- "dateModified": "2026-06-22T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -57,7 +57,7 @@
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/" },
- "keywords": "run munder difflin on a mac mini, local llm mac mini, ollama mac mini, apple silicon unified memory llm, offline ai agents mac, lm studio mac mini agents",
+ "keywords": "run munder difflin on a mac mini, local llm mac mini, ollama mac mini, apple silicon unified memory llm, offline ai agents mac, lm studio mac mini agents, m6 mac mini local llm",
"articleSection": "Guides",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/"
@@ -80,10 +80,11 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "Can a Mac Mini run a Munder Difflin hive offline?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. An Apple Silicon Mac Mini runs the model locally through Ollama or LM Studio, and Munder Difflin points OpenCode and Crush at that local OpenAI-compatible endpoint (in v0.3.1 pi reaches models through a third-party provider key rather than a local server). The hive's control plane (router, scheduler, mailboxes, audit log) is already local-first, so once the model is local there is no cloud dependency at all." } },
- { "@type": "Question", "name": "How much RAM does the Mac Mini need for local AI agents?", "acceptedAnswer": { "@type": "Answer", "text": "It depends on the model size, not the agent count — every worker shares one local model server. As a rule of thumb on Apple Silicon's unified memory: 16GB comfortably runs 7–8B models, 24GB runs up to ~14B, 32GB up to ~32B (tight), and 48–64GB (M4 Pro) runs a 70B-class model quantized to 4-bit. See the RAM-tier table below." } },
- { "@type": "Question", "name": "Ollama or LM Studio — which should I use on a Mac Mini?", "acceptedAnswer": { "@type": "Answer", "text": "Either works; both expose an OpenAI-compatible server that Munder Difflin's engines can target. Ollama is a lightweight CLI/daemon (great for headless, always-on hives); LM Studio is a GUI with a model browser and a one-click local server. You can even run both and point different engines at different ports." } },
- { "@type": "Question", "name": "Which open-source models run best on each engine?", "acceptedAnswer": { "@type": "Answer", "text": "All three engines — OpenCode, Crush, and pi — select models in `provider/model` form. For the fully-local path, OpenCode and Crush point at a local OpenAI-compatible endpoint, so the same locally-served model works across both; in v0.3.1 pi runs through a third-party provider key instead. For the exact, version-verified model slugs and per-RAM-tier picks, see the open-source model catalog linked below — that's the single source of truth we keep current." } }
+ { "@type": "Question", "name": "Can a Mac mini run a Munder Difflin office offline?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. The Mac mini serves an open model through Ollama or LM Studio, and OpenCode, Crush, Qwen and Pi agents point at that local server. Munder Difflin's routing, mailboxes, memory and schedules already run on your machine, so once the model is local, the agents' work loop does not need the internet." } },
+ { "@type": "Question", "name": "How much memory does a Mac mini need for local AI agents?", "acceptedAnswer": { "@type": "Answer", "text": "It depends on the size of the model, not the number of agents, because every agent shares one model server. As a rule of thumb for 4 bit models: 16 GB runs about 8B comfortably, 24 GB up to about 14B, 32 GB up to about 32B, and 64 GB runs a 70B class model." } },
+ { "@type": "Question", "name": "Which Mac mini should I buy for local models?", "acceptedAnswer": { "@type": "Answer", "text": "Buy for the biggest model you want to run, because the memory cannot be upgraded later. As of September 2026 the M6 Mac mini goes up to 32 GB and the M5 Pro model goes up to 64 GB. For 30B class models, 32 GB is the sweet spot. For 70B class models, get the M5 Pro with 64 GB." } },
+ { "@type": "Question", "name": "Ollama or LM Studio on a Mac mini?", "acceptedAnswer": { "@type": "Answer", "text": "Either works. Both serve an OpenAI compatible endpoint the engines can use. Ollama is a light background service that suits an always on box, and LM Studio is an app with a model browser and a one click local server. You can run both on different ports." } },
+ { "@type": "Question", "name": "How does Pi use a local model?", "acceptedAnswer": { "@type": "Answer", "text": "Through Pi's own config. Add the local server to ~/.pi/agent/models.json, and Munder Difflin copies that file into every Pi agent it starts." } }
]
}
@@ -153,14 +154,14 @@
<a href="/blog/topics/guides/">Guides</a>
</nav>
<h1>Run Munder Difflin Locally on a Mac Mini</h1>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true">·</span>
- <span>8 min read</span>
+ <span>5 min read</span>
<span class="tag kind">Guides</span>
</div>
</header>
@@ -169,249 +170,174 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p>A <strong>Mac Mini</strong> with Apple
-Silicon can run an entire <strong>Munder Difflin</strong> hive <em>offline</em>. Serve an open-source
-model locally with <strong>Ollama</strong> or <strong>LM Studio</strong>, size the model to your
-<strong>unified memory</strong> (16GB → ~7–8B, 24GB → ~14B, 32GB → ~32B, 48–64GB → 70B-class at 4-bit),
-then point <a href="/blog/why-cli-agents-are-powerful/">OpenCode and Crush</a> at the local endpoint via
-<strong>Settings → AI Engines</strong> (pi uses a third-party provider key in v0.3.1). The hive's control plane is
-<a href="/blog/local-first-ai-agent-orchestration/">already local-first</a>, so nothing leaves the box.</p></div>
-<p>It’s one thing to run a single local model in a chat window. It’s another to put a whole <em>team</em> of
-agents to work on your own hardware, with no API bill and no data leaving the room. That’s the promise
-of running Munder Difflin on a Mac Mini: the <a href="/blog/local-first-ai-agent-orchestration/">orchestration is already
-local-first</a> — the message router, scheduler, mailboxes, and
-git audit log are all processes on your machine — so the only remaining cloud dependency is the model
-call itself. Move that local, and the hive is fully offline.</p>
-<p>This guide walks the whole setup: picking a Mac Mini tier, understanding how Apple Silicon’s unified
-memory bounds what you can run, installing a local model server, and wiring each CLI engine to it.</p>
-<h2 id="why-the-mac-mini-is-a-good-hive-box" tabindex="-1">Why the Mac Mini is a good hive box <a class="anchor" href="#why-the-mac-mini-is-a-good-hive-box" aria-hidden="true">#</a></h2>
-<p>The Mac Mini is the cheapest way into Apple Silicon’s <strong>unified memory</strong> architecture, and unified
-memory is exactly what local LLMs want. On a Mac, the CPU, GPU, and Neural Engine all share one pool of
-high-bandwidth RAM — so the model’s weights live in the same memory the GPU computes against, with no
-copy across a PCIe bus. A 32GB Mac Mini can hold a model that would need a 32GB <em>discrete</em> GPU on a PC,
-at a fraction of the price and power draw. And because the Mini is small, quiet, and sips power, it’s a
-natural always-on box for a hive you leave running for hours.</p>
-<p>One catch worth knowing up front: <strong>Apple Silicon RAM is soldered to the chip and cannot be upgraded
-after purchase.</strong> Buy the memory you’ll want for the largest model you intend to run — it’s the one
-spec you can’t change later.</p>
-<h2 id="step-1-pick-a-mac-mini-tier-for-the-model-you-want-to-run" tabindex="-1">Step 1 — Pick a Mac Mini tier for the model you want to run <a class="anchor" href="#step-1-pick-a-mac-mini-tier-for-the-model-you-want-to-run" aria-hidden="true">#</a></h2>
-<p>Model count doesn’t drive your RAM — <em>model size</em> does. Every worker in the hive shares one local model
-server, so a ten-agent hive and a two-agent hive on the same model need roughly the same memory. What
-you’re really choosing is <strong>how large a model the box can hold.</strong></p>
-<p>A 4-bit-quantized model needs roughly <strong>0.6 GB of memory per billion parameters</strong> for its weights, plus
-headroom for the KV cache (grows with context length), the model server, and macOS itself. macOS also
-caps how much unified memory the GPU may pin — by default around 65–75% — though you can raise that
-ceiling on a dedicated box with <code>sudo sysctl iogpu.wired_limit_mb=&lt;MB&gt;</code> when you need to fit a large
-model. A practical rule: <strong>plan for the model to use about two-thirds of your unified memory</strong> and leave
-the rest for context and the OS.</p>
-<p>Current Mac Mini lineup and what each tier comfortably runs (4-bit quant):</p>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p>An Apple silicon <strong>Mac mini</strong> can run
+a whole <strong>Munder Difflin</strong> office offline. Serve an open model with <strong>Ollama</strong> or
+<strong>LM Studio</strong>, size it to your <strong>unified memory</strong> (16 GB about 8B, 24 GB about 14B, 32 GB
+about 32B, 64 GB a 70B class model at 4 bit), then point <strong>OpenCode</strong>, <strong>Crush</strong> and
+<strong>Qwen</strong> at it in <strong>Settings → AI Engines</strong> and add it to <strong>Pi</strong>'s own models file.
+The office's routing, mailboxes and schedules are <a href="/blog/local-first-ai-agent-orchestration/">already local
+first</a>.</p></div>
+<p>Running one local model in a chat window is a hobby. Putting a whole team of agents to work on your own hardware, with no
+API bill and nothing leaving the room, is a different project. Munder Difflin’s plumbing already runs on your machine: message
+routing, mailboxes, memory, schedules and git history. The one piece that usually calls out to the internet is the model. Move
+that onto the Mac mini and the agents’ whole loop stays in the room.</p>
+<p>This guide covers picking a Mac mini, sizing a model to its memory, installing a model server and wiring each engine, checked
+against Munder Difflin 0.5.2 and Apple’s Mac mini lineup on 10 September 2026.</p>
+<h2 id="why-is-a-mac-mini-a-good-box-for-local-agents" tabindex="-1">Why is a Mac mini a good box for local agents? <a class="anchor" href="#why-is-a-mac-mini-a-good-box-for-local-agents" aria-hidden="true">#</a></h2>
+<p>Unified memory. On Apple silicon the CPU and GPU share one pool of fast memory, so the model’s weights sit right where the GPU
+works on them, with no copying across a bus. A 32 GB Mac mini holds a model that would need a 32 GB graphics card on a PC, for
+a fraction of the power. It is also small and quiet, which makes it a natural always on box.</p>
+<p>One catch up front: <strong>the memory is part of the chip and cannot be upgraded later.</strong> Buy for the largest model you plan to run.</p>
+<h2 id="how-much-memory-do-you-need" tabindex="-1">How much memory do you need? <a class="anchor" href="#how-much-memory-do-you-need" aria-hidden="true">#</a></h2>
+<p>Enough for the model, not for the agents. Every agent shares one model server, so ten agents and two agents on the same model
+need about the same memory.</p>
+<p>A 4 bit model needs roughly <strong>0.6 GB per billion parameters</strong>, plus room for context, the model server and macOS. macOS also
+caps how much memory the GPU can hold by default, somewhere around two thirds to three quarters of it. On a dedicated box you can
+raise that ceiling with <code>sudo sysctl iogpu.wired_limit_mb=&lt;MB&gt;</code>. A practical rule: <strong>plan for the model to use about two thirds of
+your memory.</strong></p>
+<p>Here is the Mac mini range as of September 2026, and what each memory size runs comfortably at 4 bit:</p>
<div class="table-scroll">
<table>
<thead>
<tr>
-<th>Mac Mini</th>
-<th>Unified memory</th>
-<th>Memory bandwidth</th>
-<th>Comfortable model size (Q4)</th>
+<th>Mac mini</th>
+<th>Memory</th>
+<th>Comfortable model size</th>
</tr>
</thead>
<tbody>
<tr>
-<td>M4 (base)</td>
-<td>16GB</td>
-<td>120 GB/s</td>
-<td>~7–8B (e.g. an 8B-class coder)</td>
+<td>M6</td>
+<td>16 GB</td>
+<td>About 8B</td>
</tr>
<tr>
-<td>M4</td>
-<td>24GB</td>
-<td>120 GB/s</td>
-<td>up to ~14B</td>
+<td>M6</td>
+<td>24 GB</td>
+<td>Up to about 14B</td>
</tr>
<tr>
-<td>M4</td>
-<td>32GB</td>
-<td>120 GB/s</td>
-<td>up to ~32B (tight; short context)</td>
+<td>M6</td>
+<td>32 GB</td>
+<td>Up to about 32B, with shorter context</td>
</tr>
<tr>
-<td>M4 Pro</td>
-<td>48GB</td>
-<td>273 GB/s</td>
-<td>~32B with room, or a 70B at heavy quant</td>
+<td>M5 Pro</td>
+<td>24 GB</td>
+<td>Up to about 14B, generated faster</td>
</tr>
<tr>
-<td>M4 Pro</td>
-<td>64GB</td>
-<td>273 GB/s</td>
-<td>70B-class at 4-bit</td>
+<td>M5 Pro</td>
+<td>48 GB</td>
+<td>32B with room, or 70B at heavy quantization</td>
+</tr>
+<tr>
+<td>M5 Pro</td>
+<td>64 GB</td>
+<td>70B class at 4 bit</td>
</tr>
</tbody>
</table>
</div>
-<p><em>(Newer Mac Mini chips raise bandwidth and the memory ceiling, but the same “model ≈ two-thirds of
-unified memory” sizing rule carries over — pick the tier by the model size you want.)</em></p>
-<p>Here’s a concrete pick for each tier. Pull these with Ollama (LM Studio carries the same models as
-GGUF/MLX): <code>ollama pull &lt;tag&gt;</code>, then reference the model as <code>local/&lt;tag&gt;</code> in OpenCode or <code>ollama/&lt;tag&gt;</code> in
-Crush (more on that in Step 3).</p>
+<p>The M5 Pro also has much faster memory, 307 GB/s against up to 170 GB/s on the M6, and that shows up as tokens per second.</p>
+<p>A concrete pick for each memory size, pulled with <code>ollama pull &lt;tag&gt;</code>:</p>
<div class="table-scroll">
<table>
<thead>
<tr>
-<th>Mac Mini RAM</th>
-<th>Recommended pick (Ollama tag)</th>
-<th>Weights (Q4)</th>
+<th>Memory</th>
+<th>Pick (Ollama tag)</th>
<th>Good for</th>
</tr>
</thead>
<tbody>
<tr>
-<td>16GB</td>
-<td><code>gpt-oss:20b</code> (tight) — or lighter <code>qwen3:8b</code> / <code>deepseek-r1:8b</code></td>
-<td>~14GB / ~5GB</td>
-<td>Smallest genuinely capable default; the 8B picks leave more context headroom</td>
-</tr>
-<tr>
-<td>24GB</td>
-<td><code>qwen3:14b</code> · <code>mistral-small:24b</code></td>
-<td>9–14GB</td>
-<td>Roomy generalist with context to spare</td>
+<td>16 GB</td>
+<td><code>gpt-oss:20b</code> (tight), or <code>qwen3:8b</code> or <code>deepseek-r1:8b</code> for more context room</td>
+<td>The smallest capable default</td>
</tr>
<tr>
-<td>32GB</td>
-<td><code>qwen3:30b-a3b</code> (MoE) · <code>qwen3-coder:30b</code> · <code>deepseek-r1:32b</code></td>
-<td>~19–20GB</td>
-<td>The sweet spot — fast MoE generalist, coding, or reasoning</td>
+<td>24 GB</td>
+<td><code>qwen3:14b</code> or <code>mistral-small:24b</code></td>
+<td>A roomy generalist</td>
</tr>
<tr>
-<td>48GB (M4 Pro)</td>
-<td><code>glm-4.7-flash</code> (q8) · <code>mixtral:8x7b</code></td>
-<td>26–32GB</td>
-<td>Bigger context, plus the only Mac-viable GLM</td>
+<td>32 GB</td>
+<td><code>qwen3:30b-a3b</code>, <code>qwen3-coder:30b</code> or <code>deepseek-r1:32b</code></td>
+<td>The sweet spot: generalist, coding or reasoning</td>
</tr>
<tr>
-<td>64GB (M4 Pro)</td>
-<td><code>llama3.3:70b</code> · <code>deepseek-r1:70b</code></td>
-<td>~43GB</td>
-<td>A 70B-class generalist or reasoner at 4-bit</td>
+<td>48 GB</td>
+<td><code>glm-4.7-flash</code> at 8 bit, or <code>mixtral:8x7b</code></td>
+<td>Bigger context</td>
</tr>
<tr>
-<td>96GB+ (Studio-class)</td>
-<td><code>gpt-oss:120b</code> · <code>llama4:scout</code></td>
-<td>65–67GB</td>
-<td>Top local models; need ~80GB resident</td>
+<td>64 GB</td>
+<td><code>llama3.3:70b</code> or <code>deepseek-r1:70b</code></td>
+<td>A 70B class generalist or reasoner</td>
</tr>
</tbody>
</table>
</div>
-<p>These picks are drawn from the model families our open-source model catalog tracks — the single source of
-truth we keep version-current.
-The companion guide, <a href="/blog/run-munder-difflin-on-open-models/">running Munder Difflin on open
-models</a>, pairs each tier with the same picks and adds the third-party-provider route.</p>
-<div class="callout"><span class="ic">⚠️</span><p><strong>What a Mac Mini <em>can't</em> run
-locally.</strong> The frontier open-weight flagships — <strong>Kimi K2.x</strong> (~1&nbsp;trillion
-parameters) and <strong>GLM-5.2</strong> (744B) — are server-class: their 4-bit weights run to hundreds of
-gigabytes, far past any Mac Mini. To use those, skip the local server and point an engine at a third-party
-open-source-model provider (OpenRouter, Groq, DeepInfra, and friends) with an API key — covered in the
-<a href="/blog/run-munder-difflin-on-open-models/">open-models guide</a>. For a genuinely local GLM,
-<code>glm-4.7-flash</code> (~30B) is the Mac-friendly member of the family.</p></div>
+<div class="callout note"><span class="ic">Heads up</span><p><strong>What a Mac mini cannot run.</strong> The frontier open
+flagships, like Kimi K2.6 and Qwen3 235B, are server class and far beyond 64 GB. Use a provider for those, which the
+<a href="/blog/run-munder-difflin-on-open-models/">open models guide</a> covers. And <code>gpt-oss:120b</code> needs about 96 GB,
+which means a Mac Studio rather than a mini.</p></div>
<figure class="fig fig-inline"><img src="/blog/assets/media/run-munder-difflin-on-a-mac-mini/note-1.png" alt="Hand-drawn sketch from “Run Munder Difflin Locally on a Mac Mini”" loading="lazy" decoding="async" /></figure>
-<h2 id="step-2-install-a-local-model-server" tabindex="-1">Step 2 — Install a local model server <a class="anchor" href="#step-2-install-a-local-model-server" aria-hidden="true">#</a></h2>
-<p>You have two solid options. Both expose an <strong>OpenAI-compatible HTTP endpoint</strong>, which is exactly what
-Munder Difflin’s engines target.</p>
-<h3 id="option-a-ollama-lightweight-headless-friendly" tabindex="-1">Option A — Ollama (lightweight, headless-friendly) <a class="anchor" href="#option-a-ollama-lightweight-headless-friendly" aria-hidden="true">#</a></h3>
-<pre class="language-bash"><code class="language-bash"><span class="token comment"># Install (Homebrew) and start the daemon</span>
-brew <span class="token function">install</span> ollama
-ollama serve <span class="token comment"># runs the API on http://localhost:11434</span>
-
-<span class="token comment"># Pull a model (pick the tag for your RAM tier — see Step 1)</span>
-ollama pull gpt-oss:20b <span class="token comment"># 16GB-friendly default; swap for your tier's tag</span></code></pre>
-<p>Ollama’s OpenAI-compatible API is served at <strong><code>http://localhost:11434/v1</code></strong>. It runs as a background
-daemon, which makes it the natural choice for an always-on hive box.</p>
-<h3 id="option-b-lm-studio-gui-model-browser-one-click-server" tabindex="-1">Option B — LM Studio (GUI, model browser, one-click server) <a class="anchor" href="#option-b-lm-studio-gui-model-browser-one-click-server" aria-hidden="true">#</a></h3>
-<p>Download LM Studio from <a href="https://lmstudio.ai">lmstudio.ai</a>, search and download a model in its UI, then
-start the <strong>Local Server</strong> (the “Developer” / server tab). LM Studio serves an OpenAI-compatible API at
-<strong><code>http://localhost:1234/v1</code></strong>.</p>
-<div class="callout"><span class="ic">💡</span><p>You can run <em>both</em> — Ollama on :11434 and LM
-Studio on :1234 — and point different engines at different ports if you want to compare models side by
-side across your workers.</p></div>
-<h2 id="step-3-wire-each-cli-engine-to-the-local-endpoint" tabindex="-1">Step 3 — Wire each CLI engine to the local endpoint <a class="anchor" href="#step-3-wire-each-cli-engine-to-the-local-endpoint" aria-hidden="true">#</a></h2>
-<p>Munder Difflin runs each agent on a pluggable <a href="/blog/why-cli-agents-are-powerful/">CLI engine</a>. All three engines select models in <strong><code>provider/model</code></strong> form, but for the fully-local path it’s <strong>two</strong> —
-OpenCode and Crush — that wire a local OpenAI-compatible endpoint in v0.3.1, so the same locally-served
-model works across both. (pi reaches models through a third-party provider key in this release — see its
-note below.) The fastest path
-is <strong>Settings → AI Engines</strong>: set the per-engine base URL to your local server and pick the model; the
-harness writes the right per-agent config for you. Here’s what that maps to under the hood for each
-engine.</p>
+<h2 id="step-1-install-a-model-server" tabindex="-1">Step 1: Install a model server <a class="anchor" href="#step-1-install-a-model-server" aria-hidden="true">#</a></h2>
+<p>Ollama or LM Studio. Both serve an OpenAI compatible endpoint, which is what the engines talk to.</p>
+<h3 id="ollama-light-good-for-an-always-on-box" tabindex="-1">Ollama: light, good for an always on box <a class="anchor" href="#ollama-light-good-for-an-always-on-box" aria-hidden="true">#</a></h3>
+<pre class="language-bash"><code class="language-bash">brew <span class="token function">install</span> ollama
+ollama serve <span class="token comment"># serves the API on http://localhost:11434</span>
+ollama pull gpt-oss:20b <span class="token comment"># swap in the tag for your memory size</span></code></pre>
+<p>Its OpenAI compatible API lives at <strong><code>http://localhost:11434/v1</code></strong>. It runs as a background service, which suits a box you leave on.</p>
+<h3 id="lm-studio-an-app-with-a-model-browser" tabindex="-1">LM Studio: an app with a model browser <a class="anchor" href="#lm-studio-an-app-with-a-model-browser" aria-hidden="true">#</a></h3>
+<p>Download it from <a href="https://lmstudio.ai">lmstudio.ai</a>, find and download a model inside the app, then start its local server from the
+Developer section. It serves an OpenAI compatible API at <strong><code>http://localhost:1234/v1</code></strong>.</p>
+<p>You can run both, Ollama on 11434 and LM Studio on 1234, and point different engines at different ports to compare models.</p>
+<h2 id="step-2-wire-each-engine-to-the-local-server" tabindex="-1">Step 2: Wire each engine to the local server <a class="anchor" href="#step-2-wire-each-engine-to-the-local-server" aria-hidden="true">#</a></h2>
+<p>Four engines can use a local model on your Mac mini floor.</p>
+<h3 id="opencode" tabindex="-1">OpenCode <a class="anchor" href="#opencode" aria-hidden="true">#</a></h3>
+<p>Open <strong>Settings → AI Engines</strong> and set OpenCode’s local base URL to <code>http://localhost:11434/v1</code>. The app injects it as a local
+provider when the agent starts. Pick the model as <code>local/&lt;tag&gt;</code>, for example <code>local/gpt-oss:20b</code>.</p>
<h3 id="crush" tabindex="-1">Crush <a class="anchor" href="#crush" aria-hidden="true">#</a></h3>
-<p>Crush reads a JSON config. Define the local server as an <strong><code>openai-compat</code></strong> provider (this keeps it on
-the OpenAI wire, which is what the harness proxy expects — don’t use Crush’s native <code>ollama</code>/<code>lmstudio</code>
-discovery types for the hive path):</p>
+<p>Set Crush’s local base URL in the same panel. Crush runs through a small local proxy inside the app, and your server becomes that
+proxy’s upstream, so there is no Crush config to edit by hand. Pick the model as <code>ollama/&lt;tag&gt;</code>, for example <code>ollama/gpt-oss:20b</code>.</p>
+<h3 id="qwen" tabindex="-1">Qwen <a class="anchor" href="#qwen" aria-hidden="true">#</a></h3>
+<p>Set Qwen’s local base URL and default model in the same panel. Qwen uses the same proxy approach as Crush.</p>
+<h3 id="pi" tabindex="-1">Pi <a class="anchor" href="#pi" aria-hidden="true">#</a></h3>
+<p>Pi reads local models from its own file, <code>~/.pi/agent/models.json</code>, and Munder Difflin copies that file into every Pi agent it
+starts. The app’s base URL field for Pi stays reserved. Add Ollama like this, then pick <code>ollama/gpt-oss:20b</code>:</p>
<pre class="language-json"><code class="language-json"><span class="token punctuation">{</span>
<span class="token property">"providers"</span><span class="token operator">:</span> <span class="token punctuation">{</span>
<span class="token property">"ollama"</span><span class="token operator">:</span> <span class="token punctuation">{</span>
- <span class="token property">"name"</span><span class="token operator">:</span> <span class="token string">"Ollama"</span><span class="token punctuation">,</span>
- <span class="token property">"type"</span><span class="token operator">:</span> <span class="token string">"openai-compat"</span><span class="token punctuation">,</span>
- <span class="token property">"base_url"</span><span class="token operator">:</span> <span class="token string">"http://localhost:11434/v1"</span><span class="token punctuation">,</span>
- <span class="token property">"api_key"</span><span class="token operator">:</span> <span class="token string">"ollama"</span><span class="token punctuation">,</span>
- <span class="token property">"models"</span><span class="token operator">:</span> <span class="token punctuation">[</span>
- <span class="token punctuation">{</span> <span class="token property">"name"</span><span class="token operator">:</span> <span class="token string">"gpt-oss 20B"</span><span class="token punctuation">,</span> <span class="token property">"id"</span><span class="token operator">:</span> <span class="token string">"gpt-oss:20b"</span><span class="token punctuation">,</span> <span class="token property">"context_window"</span><span class="token operator">:</span> <span class="token number">131072</span><span class="token punctuation">,</span> <span class="token property">"default_max_tokens"</span><span class="token operator">:</span> <span class="token number">8192</span> <span class="token punctuation">}</span>
- <span class="token punctuation">]</span>
+ <span class="token property">"baseUrl"</span><span class="token operator">:</span> <span class="token string">"http://localhost:11434/v1"</span><span class="token punctuation">,</span>
+ <span class="token property">"api"</span><span class="token operator">:</span> <span class="token string">"openai-completions"</span><span class="token punctuation">,</span>
+ <span class="token property">"apiKey"</span><span class="token operator">:</span> <span class="token string">"ollama"</span><span class="token punctuation">,</span>
+ <span class="token property">"models"</span><span class="token operator">:</span> <span class="token punctuation">[</span><span class="token punctuation">{</span> <span class="token property">"id"</span><span class="token operator">:</span> <span class="token string">"gpt-oss:20b"</span> <span class="token punctuation">}</span><span class="token punctuation">]</span>
<span class="token punctuation">}</span>
<span class="token punctuation">}</span>
<span class="token punctuation">}</span></code></pre>
-<p>Then select the model as <code>ollama/gpt-oss:20b</code> (swap in the tag for your RAM tier from Step 1). For LM
-Studio, use <code>base_url: &quot;http://localhost:1234/v1&quot;</code>, <code>api_key: &quot;lm-studio&quot;</code>, and a <code>lmstudio</code> provider key,
-then select <code>lmstudio/&lt;model-id&gt;</code>.</p>
-<h3 id="opencode" tabindex="-1">OpenCode <a class="anchor" href="#opencode" aria-hidden="true">#</a></h3>
-<p>OpenCode reaches a local model through a custom OpenAI-compatible provider (the <code>@ai-sdk/openai-compatible</code>
-adapter with <code>options.baseURL</code>). In Munder Difflin, set the OpenCode base URL in Settings → AI Engines to
-your local server and choose the model in <code>provider/model</code> form; the harness supplies the engine config.</p>
-<h3 id="pi" tabindex="-1">pi <a class="anchor" href="#pi" aria-hidden="true">#</a></h3>
-<p>A heads-up specific to <strong>v0.3.1</strong>: pi’s local-endpoint field is <strong>reserved.</strong> This release wires a local
-base URL for OpenCode and Crush only — it does not yet write a local provider into pi’s <code>models.json</code> — so
-the fully-offline path today runs on <strong>OpenCode or Crush.</strong> To put pi to work in this release, point it at
-a third-party open-source-model provider instead of a local server: set a provider key in <strong>Settings → AI
-Engines</strong> (e.g. an OpenRouter or Groq key) and pick an open-model slug such as
-<code>openrouter/openai/gpt-oss-120b</code> or <code>groq/llama-3.3-70b-versatile</code>. pi joins the local path once its
-local-endpoint field ships; until then, build your offline hive on OpenCode and Crush.</p>
-<div class="callout"><span class="ic">📋</span><p><strong>The same local model, three prefixes.</strong>
-Once a model is served locally, each engine names it slightly differently: OpenCode as
-<code>local/&lt;tag&gt;</code>, Crush as <code>ollama/&lt;tag&gt;</code>. For example, gpt-oss 20B is
-<code>local/gpt-oss:20b</code> in OpenCode and <code>ollama/gpt-oss:20b</code> in Crush. (In v0.3.1, pi
-points at a third-party <code>provider/model</code> slug instead, per the note above.) The exact slugs for
-every tier live in the <a href="/blog/run-munder-difflin-on-open-models/">open-source model catalog</a>,
-kept in one place so they don't drift.</p></div>
+<p>For LM Studio, use <code>http://localhost:1234/v1</code> and the model id LM Studio shows you.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/run-munder-difflin-on-a-mac-mini/note-2.png" alt="Hand-drawn sketch from “Run Munder Difflin Locally on a Mac Mini”" loading="lazy" decoding="async" /></figure>
-<h2 id="step-4-run-the-hive-offline" tabindex="-1">Step 4 — Run the hive offline <a class="anchor" href="#step-4-run-the-hive-offline" aria-hidden="true">#</a></h2>
-<p>With a model served locally and the engines pointed at it, the rest of Munder Difflin is already
-local-first. Spin up your workers from the <a href="/blog/how-to-install-and-use-munder-difflin/">Agent Gallery or Add Agent</a>,
-give the hive a goal, and let it run. The <a href="/blog/atomic-file-mailboxes-for-agents/">message router</a> moves
-work between agents through file mailboxes, the scheduler fires tasks on local timers, and git records
-the audit trail — none of it touches a network. Pull the Ethernet cable and the hive keeps working.</p>
-<p>A few operational notes for an offline Mac Mini hive:</p>
+<h2 id="step-3-run-the-office" tabindex="-1">Step 3: Run the office <a class="anchor" href="#step-3-run-the-office" aria-hidden="true">#</a></h2>
+<p>With a model served and the engines pointed at it, hire workers from the
+<a href="/blog/how-to-hire-from-the-agent-gallery/">Agent Gallery or Add agent</a>, give Michael a goal, and let it run. Messages move between
+agents through <a href="/blog/atomic-file-mailboxes-for-agents/">file mailboxes</a>, schedules fire on local timers, and git keeps the history.</p>
+<p>A few notes for an always on Mac mini:</p>
<ul>
-<li><strong>One model, many workers.</strong> Every agent shares the local server, so concurrency is bounded by the
-Mac Mini’s compute, not by per-agent memory. Watch tokens/sec under load and keep contexts trim on the
-16/24GB tiers.</li>
-<li><strong>Keep it awake.</strong> A headless always-on hive wants the Mini to not sleep — set Energy settings to
-prevent sleep (or <code>caffeinate</code> the session) so the router and scheduler keep ticking.</li>
-<li><strong>Quantization is your lever.</strong> If a model won’t fit, drop to a heavier quant before dropping a tier —
-4-bit is the usual sweet spot for coder models on these boxes.</li>
-<li><strong>Fully offline means OpenCode or Crush.</strong> In v0.3.1 those are the two engines that wire a local base
-URL, so a hive built on them runs with the cable pulled. pi currently reaches models through a
-third-party provider key (its local field is reserved), so pi workers need a network — mix engines
-accordingly if total isolation is the goal.</li>
+<li><strong>One model, many workers.</strong> Every agent shares the model server, so the limit is the chip’s speed, not memory per agent. Watch
+tokens per second under load, and keep contexts trim on 16 and 24 GB.</li>
+<li><strong>Keep it awake.</strong> Set the Mac mini not to sleep, or run <code>caffeinate</code>, so schedules keep firing. Munder Difflin also holds the
+machine awake while at least one agent is running.</li>
+<li><strong>Quantization is your lever.</strong> If a model does not fit, try a smaller quantization before you try a smaller model.</li>
+<li><strong>Mix engines freely.</strong> OpenCode, Crush, Qwen and Pi can all share the same local server, so choose per agent.</li>
</ul>
<h2 id="where-to-go-next" tabindex="-1">Where to go next <a class="anchor" href="#where-to-go-next" aria-hidden="true">#</a></h2>
<ul>
-<li>The companion guide, <a href="/blog/run-munder-difflin-on-open-models/">run Munder Difflin on open models</a>,
-covers running fully local <em>or</em> through a third-party open-source provider, and pairs each RAM tier
-with a concrete model pick.</li>
-<li>For the <em>why</em> behind keeping it all on your machine, see <a href="/blog/why-local-first-matters-for-ai-agents/">why local-first matters for AI
-agents</a>.</li>
-<li>New to the app? Start with <a href="/blog/how-to-install-and-use-munder-difflin/">how to install and use Munder
-Difflin</a>.</li>
+<li><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on open models</a> adds the provider route and the full list of quick picks.</li>
+<li><a href="/blog/why-local-first-matters-for-ai-agents/">Why local first matters for AI agents</a>, for the reasoning behind all of this.</li>
+<li>New to the app? Start with <a href="/blog/how-to-install-and-use-munder-difflin/">how to install and use Munder Difflin</a>.</li>
</ul>
@@ -420,23 +346,28 @@ Difflin</a>.</li>
<h2 id="faq">FAQ</h2>
<div class="qa">
- <p class="q">Can a Mac Mini run a Munder Difflin hive offline?</p>
- <p class="a">Yes. An Apple Silicon Mac Mini runs the model locally through Ollama or LM Studio, and Munder Difflin points OpenCode and Crush at that local OpenAI-compatible endpoint (in v0.3.1 pi reaches models through a third-party provider key rather than a local server). The hive&#39;s control plane (router, scheduler, mailboxes, audit log) is already local-first, so once the model is local there is no cloud dependency at all.</p>
+ <p class="q">Can a Mac mini run a Munder Difflin office offline?</p>
+ <p class="a">Yes. The Mac mini serves an open model through Ollama or LM Studio, and OpenCode, Crush, Qwen and Pi agents point at that local server. Munder Difflin&#39;s routing, mailboxes, memory and schedules already run on your machine, so once the model is local, the agents&#39; work loop does not need the internet.</p>
+ </div>
+
+ <div class="qa">
+ <p class="q">How much memory does a Mac mini need for local AI agents?</p>
+ <p class="a">It depends on the size of the model, not the number of agents, because every agent shares one model server. As a rule of thumb for 4 bit models: 16 GB runs about 8B comfortably, 24 GB up to about 14B, 32 GB up to about 32B, and 64 GB runs a 70B class model.</p>
</div>
<div class="qa">
- <p class="q">How much RAM does the Mac Mini need for local AI agents?</p>
- <p class="a">It depends on the model size, not the agent count — every worker shares one local model server. As a rule of thumb on Apple Silicon&#39;s unified memory: 16GB comfortably runs 7–8B models, 24GB runs up to ~14B, 32GB up to ~32B (tight), and 48–64GB (M4 Pro) runs a 70B-class model quantized to 4-bit. See the RAM-tier table below.</p>
+ <p class="q">Which Mac mini should I buy for local models?</p>
+ <p class="a">Buy for the biggest model you want to run, because the memory cannot be upgraded later. As of September 2026 the M6 Mac mini goes up to 32 GB and the M5 Pro model goes up to 64 GB. For 30B class models, 32 GB is the sweet spot. For 70B class models, get the M5 Pro with 64 GB.</p>
</div>
<div class="qa">
- <p class="q">Ollama or LM Studio — which should I use on a Mac Mini?</p>
- <p class="a">Either works; both expose an OpenAI-compatible server that Munder Difflin&#39;s engines can target. Ollama is a lightweight CLI/daemon (great for headless, always-on hives); LM Studio is a GUI with a model browser and a one-click local server. You can even run both and point different engines at different ports.</p>
+ <p class="q">Ollama or LM Studio on a Mac mini?</p>
+ <p class="a">Either works. Both serve an OpenAI compatible endpoint the engines can use. Ollama is a light background service that suits an always on box, and LM Studio is an app with a model browser and a one click local server. You can run both on different ports.</p>
</div>
<div class="qa">
- <p class="q">Which open-source models run best on each engine?</p>
- <p class="a">All three engines — OpenCode, Crush, and pi — select models in `provider/model` form. For the fully-local path, OpenCode and Crush point at a local OpenAI-compatible endpoint, so the same locally-served model works across both; in v0.3.1 pi runs through a third-party provider key instead. For the exact, version-verified model slugs and per-RAM-tier picks, see the open-source model catalog linked below — that&#39;s the single source of truth we keep current.</p>
+ <p class="q">How does Pi use a local model?</p>
+ <p class="a">Through Pi&#39;s own config. Add the local server to ~/.pi/agent/models.json, and Munder Difflin copies that file into every Pi agent it starts.</p>
</div>
</section>
@@ -455,14 +386,14 @@ Difflin</a>.</li>
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#why-the-mac-mini-is-a-good-hive-box">Why the Mac Mini is a good hive box</a></li><li class="lvl-2"><a href="#step-1-pick-a-mac-mini-tier-for-the-model-you-want-to-run">Step 1 — Pick a Mac Mini tier for the model you want to run</a></li><li class="lvl-2"><a href="#step-2-install-a-local-model-server">Step 2 — Install a local model server</a></li><li class="lvl-3"><a href="#option-a-ollama-lightweight-headless-friendly">Option A — Ollama (lightweight, headless-friendly)</a></li><li class="lvl-3"><a href="#option-b-lm-studio-gui-model-browser-one-click-server">Option B — LM Studio (GUI, model browser, one-click server)</a></li><li class="lvl-2"><a href="#step-3-wire-each-cli-engine-to-the-local-endpoint">Step 3 — Wire each CLI engine to the local endpoint</a></li><li class="lvl-3"><a href="#crush">Crush</a></li><li class="lvl-3"><a href="#opencode">OpenCode</a></li><li class="lvl-3"><a href="#pi">pi</a></li><li class="lvl-2"><a href="#step-4-run-the-hive-offline">Step 4 — Run the hive offline</a></li><li class="lvl-2"><a href="#where-to-go-next">Where to go next</a></li>
+ <li class="lvl-2"><a href="#why-is-a-mac-mini-a-good-box-for-local-agents">Why is a Mac mini a good box for local agents?</a></li><li class="lvl-2"><a href="#how-much-memory-do-you-need">How much memory do you need?</a></li><li class="lvl-2"><a href="#step-1-install-a-model-server">Step 1: Install a model server</a></li><li class="lvl-3"><a href="#ollama-light-good-for-an-always-on-box">Ollama: light, good for an always on box</a></li><li class="lvl-3"><a href="#lm-studio-an-app-with-a-model-browser">LM Studio: an app with a model browser</a></li><li class="lvl-2"><a href="#step-2-wire-each-engine-to-the-local-server">Step 2: Wire each engine to the local server</a></li><li class="lvl-3"><a href="#opencode">OpenCode</a></li><li class="lvl-3"><a href="#crush">Crush</a></li><li class="lvl-3"><a href="#qwen">Qwen</a></li><li class="lvl-3"><a href="#pi">Pi</a></li><li class="lvl-2"><a href="#step-3-run-the-office">Step 3: Run the office</a></li><li class="lvl-2"><a href="#where-to-go-next">Where to go next</a></li>
</ol>
</aside>
</div>
<footer class="post-foot wrap"><div class="prevnext">
- <a class="pn prev" href="/blog/run-munder-difflin-on-open-models/"><span class="k">← Older</span><span class="t">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</span></a>
+ <a class="pn prev" href="/blog/run-munder-difflin-on-open-models/"><span class="k">← Older</span><span class="t">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</span></a>
<a class="pn next" href="/blog/launching-munder-difflin-v0-3-2/"><span class="k">Newer →</span><span class="t">Launching Munder Difflin v0.3.2: Talk to Michael, a Realtime Voice Orchestrator</span></a>
</div>
<section class="related">
@@ -474,13 +405,13 @@ Difflin</a>.</li>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -506,10 +437,10 @@ Difflin</a>.</li>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/run-munder-difflin-on-open-models/index.html b/docs/blog/run-munder-difflin-on-open-models/index.html
index 35f25405..34945d3a 100644
--- a/docs/blog/run-munder-difflin-on-open-models/index.html
+++ b/docs/blog/run-munder-difflin-on-open-models/index.html
@@ -3,8 +3,8 @@
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
-<title>Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider — Munder Difflin Blog</title>
-<meta name="description" content="Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4." />
+<title>Run Munder Difflin on Open Source Models: Fully Local or Through a Provider — Munder Difflin Blog</title>
+<meta name="description" content="Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2." />
<link rel="canonical" href="https://munderdiffl.in/blog/run-munder-difflin-on-open-models/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -14,8 +14,8 @@
<!-- Open Graph -->
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
-<meta property="og:title" content="Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider — Munder Difflin Blog" />
-<meta property="og:description" content="Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4." />
+<meta property="og:title" content="Run Munder Difflin on Open Source Models: Fully Local or Through a Provider — Munder Difflin Blog" />
+<meta property="og:description" content="Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2." />
<meta property="og:url" content="https://munderdiffl.in/blog/run-munder-difflin-on-open-models/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -30,8 +30,8 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
-<meta name="twitter:title" content="Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider — Munder Difflin Blog" />
-<meta name="twitter:description" content="Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4." />
+<meta name="twitter:title" content="Run Munder Difflin on Open Source Models: Fully Local or Through a Provider — Munder Difflin Blog" />
+<meta name="twitter:description" content="Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -44,11 +44,11 @@
{
"@context": "https://schema.org",
"@type": "BlogPosting",
- "headline": "Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider",
- "description": "Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker's own CLI. Here's the wiring, current as of v0.4.4.",
+ "headline": "Run Munder Difflin on Open Source Models: Fully Local or Through a Provider",
+ "description": "Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-06-22T00:00:00.000Z",
- "dateModified": "2026-08-20T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -56,7 +56,7 @@
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/run-munder-difflin-on-open-models/" },
- "keywords": "run ai agents on open source models, local llm coding agent, ollama coding agent, openrouter coding agent, gpt-oss, byok open models, opencode crush pi, qwen cli agent",
+ "keywords": "run ai agents on open source models, local llm coding agent, ollama coding agent, openrouter coding agent, gpt-oss, byok open models, opencode crush pi, pi models.json ollama",
"articleSection": "Guides",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/run-munder-difflin-on-open-models/"
@@ -69,7 +69,7 @@
"itemListElement": [
{ "@type": "ListItem", "position": 1, "name": "Home", "item": "https://munderdiffl.in/" },
{ "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://munderdiffl.in/blog/" },
- { "@type": "ListItem", "position": 3, "name": "Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider", "item": "https://munderdiffl.in/blog/run-munder-difflin-on-open-models/" }
+ { "@type": "ListItem", "position": 3, "name": "Run Munder Difflin on Open Source Models: Fully Local or Through a Provider", "item": "https://munderdiffl.in/blog/run-munder-difflin-on-open-models/" }
]
}
</script>
@@ -79,10 +79,11 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "Can Munder Difflin run entirely on open-source models?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. The OpenCode, Crush, and pi engines all support bring-your-own-key (BYOK) and local models, and the Qwen and Kimi CLIs are first-class engines whose flagship models are open-weight. You can run every agent — workers and Michael himself — on open models like gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, or Kimi: fully local on your own hardware, or through a third-party OSS provider with your own API key." } },
- { "@type": "Question", "name": "What's the difference between running local and using a third-party provider?", "acceptedAnswer": { "@type": "Answer", "text": "Local (Ollama, LM Studio, vLLM) runs the weights on your own machine — fully private, no per-token bill, but bounded by your RAM and GPU. A third-party OSS provider (OpenRouter, Groq, Together, Fireworks, DeepInfra) hosts the same open weights on their hardware and you pay per token with your own key — no local hardware limit, so you can reach the 100B–1T-parameter frontier models a laptop can't hold." } },
- { "@type": "Question", "name": "Which open model should I pick for the orchestrator seat?", "acceptedAnswer": { "@type": "Answer", "text": "Michael does the reasoning and long-context coordination, so give him a strong model: locally, gpt-oss-120b or Llama 3.3 70B on a 64–96 GB machine; via a provider, DeepSeek-V4-Flash, GLM-4.6, or Kimi-K2.6 on OpenRouter. Small local models (8B and under) are fine for routine workers but underpowered for orchestration." } },
- { "@type": "Question", "name": "Do I need a different model id for each engine?", "acceptedAnswer": { "@type": "Answer", "text": "No — the upstream model id is the same. All three BYOK engines use a provider/model slug; only the provider prefix and how the key or base-URL is wired differ. For local models, OpenCode uses local/<id> while Crush uses ollama/<id>." } }
+ { "@type": "Question", "name": "Can Munder Difflin run entirely on open source models?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. OpenCode, Crush, Qwen and Pi can all run open weight models, either on your own machine or through a provider with your own API key. You can put open models in every seat on the floor, including Michael's." } },
+ { "@type": "Question", "name": "What is the difference between running local and using a provider?", "acceptedAnswer": { "@type": "Answer", "text": "Local means Ollama, LM Studio or vLLM runs the weights on your machine: private, no per token bill, and limited by your memory. A provider such as OpenRouter or Groq hosts the same open weights on its hardware and bills your own key per token, so you can reach models far too big for a laptop." } },
+ { "@type": "Question", "name": "Which open model should I give the orchestrator?", "acceptedAnswer": { "@type": "Answer", "text": "A strong one, because Michael does the reasoning and the long context coordination. Locally that means gpt-oss 120B or Llama 3.3 70B on a machine with 64 to 96 GB of memory. Through a provider, DeepSeek V4 Flash or Kimi K2.6 on OpenRouter. Models of 8B and under make fine workers and thin orchestrators." } },
+ { "@type": "Question", "name": "Does each engine need a different model name?", "acceptedAnswer": { "@type": "Answer", "text": "The model id stays the same and only the prefix changes. A local model is local/<tag> on OpenCode and ollama/<tag> on Crush and Pi, and a provider model carries the provider first, like openrouter/openai/gpt-oss-120b." } },
+ { "@type": "Question", "name": "How do I run Pi on a local model?", "acceptedAnswer": { "@type": "Answer", "text": "Add your local server to Pi's own config file, ~/.pi/agent/models.json. Munder Difflin copies that file into every Pi agent it starts, so the models you define there are available to your Pi agents. The Pi base URL field in the app stays reserved." } }
]
}
@@ -151,15 +152,15 @@
<span>/</span>
<a href="/blog/topics/guides/">Guides</a>
</nav>
- <h1>Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</h1>
- <p class="dek">Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</p>
+ <h1>Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</h1>
+ <p class="dek">Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true">·</span>
- <span>8 min read</span>
+ <span>6 min read</span>
<span class="tag kind">Guides</span>
</div>
</header>
@@ -168,111 +169,142 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p><strong>Munder Difflin runs entirely on open models.</strong> Three routes: <strong>fully local</strong> (Ollama / LM Studio / vLLM on your own machine — private, no per-token bill, bounded by RAM), a <strong>third-party OSS provider</strong> (OpenRouter, Groq, Together, Fireworks, DeepInfra — their hardware, your key, reaching frontier 100B–1T models a laptop can't hold), or the <strong>model-maker's own CLI</strong> — the Qwen and Kimi engines are first-class hires whose flagship models are open-weight. The BYOK engines (<strong>OpenCode</strong>, <strong>Crush</strong>, <strong>pi</strong>) all use a <code>provider/model</code> slug; keys and local base-URLs live in <strong>Settings → AI Engines</strong>, and <strong>Settings → Prerequisites</strong> tells you which engine binaries the app can actually see.</p></div>
-<p>Munder Difflin started as a harness for the closed frontier CLIs — Claude Code, Codex, Antigravity. Useful, but it tied your agent floor to a handful of vendors and their pricing. That’s long since broken open: of the <strong>ten engines</strong> the app now ships — Claude Code, Antigravity, Codex, Grok, Kimi, Qwen, OpenCode, Crush, pi, and GitHub Copilot CLI — three (<a href="https://opencode.ai">OpenCode</a>, <a href="https://charm.land">Crush</a>, <a href="https://pi.dev">pi</a>) were built from the start to point at <em>any</em> model, and two more (Qwen, Kimi) are the model-makers’ own CLIs for families whose weights are public.</p>
-<p>That means you can run an entire office of agents on models anyone can download. There are two honest ways to do it, and they trade off differently. This guide walks both, then shows the exact wiring for each engine — current as of <strong>v0.4.4</strong>. (For the why-bother, see <a href="/blog/why-local-first-matters-for-ai-agents/">why local-first matters for AI agents</a> and <a href="/blog/why-cli-agents-are-powerful/">why CLI agents are so powerful</a>.)</p>
-<h2 id="two-routes-your-hardware-or-someone-elses" tabindex="-1">Two routes: your hardware, or someone else’s <a class="anchor" href="#two-routes-your-hardware-or-someone-elses" aria-hidden="true">#</a></h2>
-<p>“Open source models” is one phrase covering two very different setups. Pick by what you’re optimizing for.</p>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p><strong>Munder Difflin can run on open models
+end to end.</strong> Two routes: <strong>fully local</strong> with Ollama, LM Studio or vLLM (private, no per
+token bill, limited by your memory), or a <strong>provider</strong> such as OpenRouter or Groq (their hardware,
+your key, much bigger models). Four engines do the wiring: <strong>OpenCode</strong>, <strong>Crush</strong> and
+<strong>Qwen</strong> take a local base URL in <strong>Settings → AI Engines</strong>, and <strong>Pi</strong> reads
+your own <code>~/.pi/agent/models.json</code>. Keys go in the same panel and are stored write only.</p></div>
+<p>Munder Difflin started out wrapping the closed frontier CLIs. Today it supports twelve, and several of them will
+point at any model you like. That means a whole office of agents can run on models whose weights anyone can download.</p>
+<p>There are two honest ways to do it, and they trade off differently. This guide walks through both, then gives the exact
+wiring for each engine, checked against Munder Difflin 0.5.2 on 10 September 2026. (For the why, see
+<a href="/blog/why-local-first-matters-for-ai-agents/">why local first matters for AI agents</a>.)</p>
+<h2 id="should-you-run-open-models-locally-or-through-a-provider" tabindex="-1">Should you run open models locally or through a provider? <a class="anchor" href="#should-you-run-open-models-locally-or-through-a-provider" aria-hidden="true">#</a></h2>
+<p>Local if privacy and a fixed cost matter and your machine can hold the model. A provider if you want the biggest models
+or have no spare hardware.</p>
<div class="table-scroll">
<table>
<thead>
<tr>
<th></th>
<th>Fully local</th>
-<th>Third-party OSS provider</th>
+<th>Through a provider</th>
</tr>
</thead>
<tbody>
<tr>
<td><strong>Runs on</strong></td>
<td>Your machine (Ollama, LM Studio, vLLM)</td>
-<td>Their GPUs (OpenRouter, Groq, Together, Fireworks, DeepInfra, Novita)</td>
+<td>Their GPUs (OpenRouter, Groq and others)</td>
</tr>
<tr>
<td><strong>Cost</strong></td>
-<td>Electricity. No per-token bill.</td>
-<td>Per-token, billed to your own key.</td>
+<td>Electricity. No per token bill.</td>
+<td>Per token, billed to your own key.</td>
</tr>
<tr>
<td><strong>Privacy</strong></td>
-<td>Total — code never leaves the box.</td>
-<td>Prompts transit a third party.</td>
+<td>Prompts never leave the machine.</td>
+<td>Prompts go to the provider.</td>
</tr>
<tr>
<td><strong>Ceiling</strong></td>
-<td>Bounded by RAM/VRAM (≈7B–70B realistic on a Mac).</td>
-<td>The whole frontier — 235B, 480B, even 1T-parameter models.</td>
+<td>Your memory: roughly 8B to 70B on a Mac, 120B on a very big one.</td>
+<td>Open models with hundreds of billions of parameters.</td>
</tr>
<tr>
<td><strong>Setup</strong></td>
-<td>Pull a model + point the engine at <code>localhost</code>.</td>
-<td>Paste one API key.</td>
+<td>Pull a model and point the engine at localhost.</td>
+<td>Paste one key.</td>
</tr>
<tr>
<td><strong>Best for</strong></td>
-<td>Private work, 24/7 unattended, fixed cost.</td>
-<td>Frontier quality, zero local hardware, bursty use.</td>
+<td>Private work, always on floors, fixed cost.</td>
+<td>Top quality, bursty use, no local hardware.</td>
</tr>
</tbody>
</table>
</div>
-<p>You don’t have to choose globally — Munder Difflin sets the engine and model <em>per agent</em>. A common pattern: a strong provider-hosted model in the orchestrator seat, and cheap local workers for the routine majority. That’s exactly the <a href="/blog/do-more-with-less-model-routing/">capability-routing</a> idea, now with open weights on both ends.</p>
-<p>And route three, for the least wiring of all: <strong>hire the model-maker’s own CLI.</strong> The Qwen and Kimi engines are first-class in the Add-Agent dialog — their vendor CLIs, their auth, no slug to compose — and Qwen3 and Kimi K2.6 are open-weight families. If all you want is “an open-model worker on my floor, now,” that’s one dropdown.</p>
-<h2 id="how-the-three-byok-engines-name-a-model" tabindex="-1">How the three BYOK engines name a model <a class="anchor" href="#how-the-three-byok-engines-name-a-model" aria-hidden="true">#</a></h2>
-<p>One thing to internalize before any wiring: <strong>OpenCode, Crush, and pi all use the same <code>provider/model</code> slug form.</strong> The <em>model</em> part is just the upstream’s id (e.g. <code>openai/gpt-oss-120b</code>, <code>qwen3:30b-a3b</code>). The <em>provider</em> prefix resolves one of two ways:</p>
-<ul>
-<li><strong>A built-in provider</strong> the engine already knows — supply the matching API-key env var and you’re done: <code>openrouter</code>, <code>openai</code>, <code>anthropic</code>, <code>groq</code>, <code>deepseek</code>, <code>mistral</code>, and the local ones.</li>
-<li><strong>A custom OpenAI-compatible provider</strong> you define once (a <code>base_url</code> + key block) for any host that isn’t built in — Together, Fireworks, DeepInfra, Novita, <a href="http://Z.ai">Z.ai</a>, Moonshot. Then the slug is <code>&lt;your-name&gt;/&lt;model-id&gt;</code>.</li>
-</ul>
-<p>The <em>local</em> route differs slightly by engine — same id, different prefix and wiring:</p>
+<p>You do not have to choose once for the whole floor. Engine and model are set per agent, so a strong provider model can sit
+in Michael’s seat while cheap local workers handle the routine majority. That is
+<a href="/blog/do-more-with-less-model-routing/">capability routing</a> with open weights at both ends.</p>
+<h2 id="which-engines-can-run-open-models" tabindex="-1">Which engines can run open models? <a class="anchor" href="#which-engines-can-run-open-models" aria-hidden="true">#</a></h2>
+<p>Four, and each one wires it a little differently.</p>
<div class="table-scroll">
<table>
<thead>
<tr>
<th>Engine</th>
-<th>How the app wires local</th>
-<th>Local slug</th>
+<th>Where the local server goes</th>
+<th>What the app does with it</th>
+<th>Local model slug</th>
</tr>
</thead>
<tbody>
<tr>
<td><strong>OpenCode</strong></td>
-<td>Injects a custom provider named <code>local</code> (OpenAI-compatible) with your base-URL.</td>
-<td><code>local/&lt;id&gt;</code></td>
+<td>Base URL in Settings → AI Engines</td>
+<td>Injects it as a local OpenAI compatible provider named <code>local</code></td>
+<td><code>local/&lt;tag&gt;</code></td>
</tr>
<tr>
<td><strong>Crush</strong></td>
-<td>Writes a provider block (<code>type: ollama / lmstudio / openai-compat</code>) into the agent’s config.</td>
-<td><code>ollama/&lt;id&gt;</code></td>
+<td>Base URL in Settings → AI Engines</td>
+<td>Uses it as the upstream of the local proxy Crush runs through</td>
+<td><code>ollama/&lt;tag&gt;</code></td>
+</tr>
+<tr>
+<td><strong>Qwen</strong></td>
+<td>Base URL and default model in Settings → AI Engines</td>
+<td>The same proxy approach as Crush</td>
+<td>the default model you set</td>
</tr>
<tr>
-<td><strong>pi</strong></td>
-<td>Still <strong>reserved</strong> as of v0.4.4 — the harness doesn’t yet write pi a <code>models.json</code>. Run open models on pi via a provider key (Path B).</td>
-<td>—</td>
+<td><strong>Pi</strong></td>
+<td>Your own <code>~/.pi/agent/models.json</code></td>
+<td>Copies that file into every Pi agent it starts</td>
+<td><code>ollama/&lt;tag&gt;</code></td>
</tr>
</tbody>
</table>
</div>
-<p>Default endpoints are the usual ones: Ollama <code>http://localhost:11434/v1</code>, LM Studio <code>http://127.0.0.1:1234/v1</code>, vLLM whatever you exposed (often <code>:8000/v1</code>). You set these in the app — no shell exports required.</p>
-<figure class="fig fig-inline"><img src="/blog/assets/media/run-munder-difflin-on-open-models/note-1.png" alt="Hand-drawn sketch from “Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider”" loading="lazy" decoding="async" /><figcaption>Same model id everywhere — only the prefix changes: local/ on OpenCode, ollama/ on Crush, provider/ for BYOK.</figcaption></figure>
-<h2 id="path-a-fully-local-ollama-lm-studio-vllm" tabindex="-1">Path A — fully local (Ollama / LM Studio / vLLM) <a class="anchor" href="#path-a-fully-local-ollama-lm-studio-vllm" aria-hidden="true">#</a></h2>
-<p>Three steps: pull a model, tell Munder Difflin where it lives, pick it for an agent.</p>
-<p><strong>1. Pull a model.</strong> With <a href="https://ollama.com">Ollama</a> installed, grab one sized to your RAM:</p>
-<pre class="language-bash"><code class="language-bash">ollama pull gpt-oss:20b <span class="token comment"># 14 GB — runs on a 16 GB Mac, the safe default</span>
-ollama pull qwen3:30b-a3b <span class="token comment"># 19 GB — fast MoE generalist, 32 GB</span>
-ollama pull deepseek-r1:32b <span class="token comment"># 20 GB — strong reasoning, 32 GB</span>
-ollama serve <span class="token comment"># exposes the OpenAI-compatible API on :11434</span></code></pre>
-<p>(LM Studio works the same way — load the model in the app and it serves on <code>:1234</code>. vLLM and llama.cpp expose their own OpenAI-compatible endpoint.)</p>
-<p><strong>2. Point the engine at it.</strong> Open <strong>Settings → AI Engines</strong>, find the engine you’ll use (<strong>OpenCode</strong> or <strong>Crush</strong>), and set its <strong>local base-URL</strong> field to your endpoint — e.g. <code>http://localhost:11434/v1</code> for Ollama. The harness uses it to inject the right provider config when it spawns the agent. No API key needed for local.</p>
-<p><strong>3. Hire an agent on that model.</strong> In the <strong>Add-Agent</strong> modal, choose the engine, then pick the local model. The picker offers the open-model quick-picks; the slug it sends is <code>local/gpt-oss:20b</code> on OpenCode, or <code>ollama/gpt-oss:20b</code> on Crush (keep the colon in the tag). That agent now runs fully on your hardware.</p>
-<p>Which local model? Match it to your machine. These picks are from the project’s open-model catalog, by RAM tier:</p>
+<p>The usual endpoints: Ollama at <code>http://localhost:11434/v1</code>, LM Studio at <code>http://localhost:1234/v1</code>, and vLLM wherever you
+expose it, often <code>:8000/v1</code>.</p>
+<figure class="fig fig-inline"><img src="/blog/assets/media/run-munder-difflin-on-open-models/note-1.png" alt="Hand-drawn sketch from “Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider”" loading="lazy" decoding="async" /><figcaption>Same model id everywhere. Only the prefix changes: local/ on OpenCode, ollama/ on Crush and Pi, and the provider name when you use a key.</figcaption></figure>
+<h2 id="how-do-you-run-a-model-fully-locally" tabindex="-1">How do you run a model fully locally? <a class="anchor" href="#how-do-you-run-a-model-fully-locally" aria-hidden="true">#</a></h2>
+<p>Pull a model, tell the engine where it lives, and pick it for an agent.</p>
+<p><strong>1. Pull a model.</strong> With <a href="https://ollama.com">Ollama</a> installed, grab one sized to your memory:</p>
+<pre class="language-bash"><code class="language-bash">ollama pull gpt-oss:20b <span class="token comment"># about 14 GB, fits a 16 GB Mac</span>
+ollama pull qwen3:30b-a3b <span class="token comment"># about 19 GB, a fast generalist for 32 GB</span>
+ollama pull deepseek-r1:32b <span class="token comment"># about 20 GB, strong reasoning for 32 GB</span>
+ollama serve <span class="token comment"># serves the OpenAI compatible API on port 11434</span></code></pre>
+<p>LM Studio works the same way: load a model in the app and start its local server on port 1234.</p>
+<p><strong>2. Point the engine at it.</strong> For OpenCode, Crush or Qwen, open <strong>Settings → AI Engines</strong> and set that engine’s local base
+URL, for example <code>http://localhost:11434/v1</code>. Local servers need no key. For Pi, add the server to <code>~/.pi/agent/models.json</code>
+instead:</p>
+<pre class="language-json"><code class="language-json"><span class="token punctuation">{</span>
+ <span class="token property">"providers"</span><span class="token operator">:</span> <span class="token punctuation">{</span>
+ <span class="token property">"ollama"</span><span class="token operator">:</span> <span class="token punctuation">{</span>
+ <span class="token property">"baseUrl"</span><span class="token operator">:</span> <span class="token string">"http://localhost:11434/v1"</span><span class="token punctuation">,</span>
+ <span class="token property">"api"</span><span class="token operator">:</span> <span class="token string">"openai-completions"</span><span class="token punctuation">,</span>
+ <span class="token property">"apiKey"</span><span class="token operator">:</span> <span class="token string">"ollama"</span><span class="token punctuation">,</span>
+ <span class="token property">"models"</span><span class="token operator">:</span> <span class="token punctuation">[</span><span class="token punctuation">{</span> <span class="token property">"id"</span><span class="token operator">:</span> <span class="token string">"gpt-oss:20b"</span> <span class="token punctuation">}</span><span class="token punctuation">]</span>
+ <span class="token punctuation">}</span>
+ <span class="token punctuation">}</span>
+<span class="token punctuation">}</span></code></pre>
+<p>Munder Difflin copies that file into each Pi agent when it starts, so start the Pi agent again after you edit it.</p>
+<p><strong>3. Hire an agent on that model.</strong> In <strong>Add agent</strong>, choose the engine and pick one of the open model quick picks, or type
+the slug yourself: <code>local/gpt-oss:20b</code> on OpenCode, <code>ollama/gpt-oss:20b</code> on Crush and Pi. Keep the colon in the tag. That
+agent now runs entirely on your hardware.</p>
+<p>Which local model? These are the app’s local quick picks, by memory:</p>
<div class="table-scroll">
<table>
<thead>
<tr>
<th>Model</th>
<th>Ollama tag</th>
-<th>Min RAM</th>
+<th>Memory</th>
<th>Good for</th>
</tr>
</thead>
@@ -281,59 +313,64 @@ ollama serve <span class="token comment"># exposes the OpenAI-
<td>gpt-oss 20B</td>
<td><code>gpt-oss:20b</code></td>
<td>16 GB</td>
-<td>Smallest capable default</td>
+<td>The smallest capable default</td>
</tr>
<tr>
<td>Mistral Small 24B</td>
<td><code>mistral-small:24b</code></td>
-<td>16–32 GB</td>
-<td>Lightweight generalist</td>
+<td>16 to 32 GB</td>
+<td>A light generalist</td>
</tr>
<tr>
-<td>Qwen3 30B-A3B (MoE)</td>
+<td>Qwen3 30B A3B</td>
<td><code>qwen3:30b-a3b</code></td>
<td>32 GB</td>
-<td>Fast MoE generalist</td>
+<td>A fast generalist</td>
</tr>
<tr>
-<td>Qwen3-Coder 30B</td>
+<td>Qwen3 Coder 30B</td>
<td><code>qwen3-coder:30b</code></td>
<td>32 GB</td>
<td>Coding</td>
</tr>
<tr>
-<td>DeepSeek-R1 32B</td>
+<td>DeepSeek R1 32B</td>
<td><code>deepseek-r1:32b</code></td>
<td>32 GB</td>
<td>Reasoning</td>
</tr>
<tr>
-<td>GLM-4.7-Flash</td>
+<td>GLM 4.7 Flash</td>
<td><code>glm-4.7-flash</code></td>
<td>32 GB</td>
-<td>The only Mac-viable GLM</td>
+<td>The GLM that fits on a Mac</td>
</tr>
<tr>
<td>Llama 3.3 70B</td>
<td><code>llama3.3:70b</code></td>
<td>64 GB</td>
-<td>Bigger generalist</td>
+<td>A bigger generalist</td>
</tr>
<tr>
<td>gpt-oss 120B</td>
<td><code>gpt-oss:120b</code></td>
<td>96 GB</td>
-<td>Top local (Studio-class)</td>
+<td>The top local pick</td>
</tr>
</tbody>
</table>
</div>
-<p>A note on what <em>won’t</em> fit: the headline frontier open models — DeepSeek-V4, Kimi K2.6, GLM-5.2, Qwen3-235B — are server-class. The Ollama tags exist, but no consumer Mac holds them. For those, you want Path B. (Choosing a local model by RAM is the whole subject of the companion <a href="/blog/run-munder-difflin-on-a-mac-mini/">Mac Mini setup guide</a>.)</p>
-<h2 id="path-b-a-third-party-oss-provider-byok" tabindex="-1">Path B — a third-party OSS provider (BYOK) <a class="anchor" href="#path-b-a-third-party-oss-provider-byok" aria-hidden="true">#</a></h2>
-<p>Same open weights, hosted on someone else’s GPUs, billed to your own key. This is how you reach the big models, and it’s a two-field setup.</p>
-<p><strong>1. Get a key.</strong> Sign up with a provider and copy an API key. <a href="https://openrouter.ai">OpenRouter</a> is the easiest start — one key, the widest catalog. <a href="https://groq.com">Groq</a> is the fastest for the models it carries. Together, Fireworks, and DeepInfra host the heavyweights.</p>
-<p><strong>2. Paste it into the app.</strong> In <strong>Settings → AI Engines</strong>, enter the key in the matching field. Keys are stored <strong>write-only through the broker</strong> — the renderer can set a key but never read it back; only the main process injects it as the right env var (<code>OPENROUTER_API_KEY</code>, <code>GROQ_API_KEY</code>, …) when an agent spawns, and only for the provider that agent actually uses. Nothing lands in plaintext config.</p>
-<p><strong>3. Pick a model.</strong> In Add-Agent, select the engine and a provider-hosted model. The slug carries the provider prefix — e.g. <code>openrouter/deepseek/deepseek-v4-flash</code> or <code>groq/openai/gpt-oss-120b</code>. The recommended BYOK quick-picks:</p>
+<p>The headline open flagships, like DeepSeek V4, Kimi K2.6 and Qwen3 235B, are server class. No consumer Mac holds them, so
+use a provider for those. Sizing a model to your memory is the whole subject of the
+<a href="/blog/run-munder-difflin-on-a-mac-mini/">Mac mini guide</a>.</p>
+<h2 id="how-do-you-run-open-models-through-a-provider" tabindex="-1">How do you run open models through a provider? <a class="anchor" href="#how-do-you-run-open-models-through-a-provider" aria-hidden="true">#</a></h2>
+<p>Same open weights, someone else’s GPUs, your own key.</p>
+<p><strong>1. Get a key.</strong> <a href="https://openrouter.ai">OpenRouter</a> is the easiest start, with one key for a wide catalog.
+<a href="https://groq.com">Groq</a> is quick for the models it carries.</p>
+<p><strong>2. Paste it into Settings → AI Engines.</strong> There are key fields for Anthropic, OpenAI, Google Gemini, OpenRouter and Groq.
+Keys are write only: the app stores a key but never shows it back, and it reaches an agent as the right environment variable
+only when that agent starts.</p>
+<p><strong>3. Pick a model.</strong> In Add agent, choose OpenCode, Crush or Pi and a provider hosted quick pick:</p>
<div class="table-scroll">
<table>
<thead>
@@ -341,12 +378,12 @@ ollama serve <span class="token comment"># exposes the OpenAI-
<th>Model</th>
<th>Route</th>
<th>Slug</th>
-<th>Key env</th>
+<th>Key</th>
</tr>
</thead>
<tbody>
<tr>
-<td>gpt-oss 120B (fastest)</td>
+<td>gpt-oss 120B</td>
<td>Groq</td>
<td><code>groq/openai/gpt-oss-120b</code></td>
<td><code>GROQ_API_KEY</code></td>
@@ -358,13 +395,13 @@ ollama serve <span class="token comment"># exposes the OpenAI-
<td><code>GROQ_API_KEY</code></td>
</tr>
<tr>
-<td>DeepSeek-V4-Flash</td>
+<td>DeepSeek V4 Flash</td>
<td>OpenRouter</td>
<td><code>openrouter/deepseek/deepseek-v4-flash</code></td>
<td><code>OPENROUTER_API_KEY</code></td>
</tr>
<tr>
-<td>GLM-4.6</td>
+<td>GLM 4.6</td>
<td>OpenRouter</td>
<td><code>openrouter/z-ai/glm-4.6</code></td>
<td><code>OPENROUTER_API_KEY</code></td>
@@ -376,12 +413,18 @@ ollama serve <span class="token comment"># exposes the OpenAI-
<td><code>OPENROUTER_API_KEY</code></td>
</tr>
<tr>
-<td>Qwen3-Coder 480B</td>
+<td>Qwen3 Coder 480B</td>
<td>OpenRouter</td>
<td><code>openrouter/qwen/qwen3-coder</code></td>
<td><code>OPENROUTER_API_KEY</code></td>
</tr>
<tr>
+<td>Qwen3 235B</td>
+<td>OpenRouter</td>
+<td><code>openrouter/qwen/qwen3-235b-a22b-2507</code></td>
+<td><code>OPENROUTER_API_KEY</code></td>
+</tr>
+<tr>
<td>gpt-oss 120B</td>
<td>OpenRouter</td>
<td><code>openrouter/openai/gpt-oss-120b</code></td>
@@ -390,21 +433,27 @@ ollama serve <span class="token comment"># exposes the OpenAI-
</tbody>
</table>
</div>
-<p>Prefer a model maker’s own API? Those work too: DeepSeek (<code>deepseek/deepseek-v4-flash</code>, <code>DEEPSEEK_API_KEY</code>), Mistral (<code>mistral/...</code>, <code>MISTRAL_API_KEY</code>), <a href="http://Z.ai">Z.ai</a> for GLM, Moonshot for Kimi. On Groq, stick to gpt-oss and <code>llama-3.3-70b-versatile</code>. The full slug-by-slug table, with citations, lives in the project’s open-model catalog (the single source of truth this post and the Mac Mini guide both cite).</p>
-<p>One honest fix from v0.4.4 worth knowing: OpenCode used to preselect a BYOK slug <strong>and silently fall back</strong> to a different model when the key was absent — while every surface kept reporting the model it had asked for. That’s gone. If a key is missing now, you find out; you never unknowingly run a model you didn’t pick.</p>
-<figure class="fig fig-inline"><img src="/blog/assets/media/run-munder-difflin-on-open-models/note-2.png" alt="Hand-drawn sketch from “Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider”" loading="lazy" decoding="async" /><figcaption>One key, the whole frontier: paste it once in Settings, and the write-only broker injects it per spawn — never into plaintext config.</figcaption></figure>
-<h2 id="per-engine-cheat-sheet" tabindex="-1">Per-engine cheat-sheet <a class="anchor" href="#per-engine-cheat-sheet" aria-hidden="true">#</a></h2>
-<p>You rarely touch these directly — the AI Engines panel writes them — but here’s what each engine does under the hood, so the model field makes sense.</p>
-<p><strong>OpenCode</strong> is OpenAI-SDK native and knows most providers out of the box. BYOK is just the env var; local is a custom provider named <code>local</code>. Slugs: <code>openrouter/openai/gpt-oss-120b</code>, <code>local/qwen3:30b-a3b</code>.</p>
-<p><strong>Crush</strong> reads BYOK env vars for its built-in providers and uses a written config block for anything custom or local. For local Ollama it’s literally:</p>
-<pre class="language-json"><code class="language-json"><span class="token punctuation">{</span> <span class="token property">"providers"</span><span class="token operator">:</span> <span class="token punctuation">{</span> <span class="token property">"ollama"</span><span class="token operator">:</span> <span class="token punctuation">{</span> <span class="token property">"type"</span><span class="token operator">:</span> <span class="token string">"ollama"</span><span class="token punctuation">,</span> <span class="token property">"base_url"</span><span class="token operator">:</span> <span class="token string">"http://localhost:11434/v1"</span> <span class="token punctuation">}</span> <span class="token punctuation">}</span> <span class="token punctuation">}</span></code></pre>
-<p>then select <code>ollama/qwen3:30b-a3b</code>. For a host like Together, it’s an <code>openai-compat</code> block with that provider’s <code>base_url</code> and your key.</p>
-<p><strong>pi</strong> ships 15+ built-in providers, so BYOK is just the provider key. Slugs look like <code>groq/llama-3.3-70b-versatile</code> or <code>openrouter/qwen/qwen3-coder</code>. Its local base-URL field remains <strong>reserved</strong> as of v0.4.4 — run open models on pi through a provider key rather than a local endpoint.</p>
-<p><strong>Qwen and Kimi</strong> are the zero-wiring route: vendor CLIs as first-class engines. Sign in the way each CLI wants and hire away — no slugs, no base-URLs.</p>
-<p>All the BYOK engines are orchestrator-eligible, so you can put an open model in Michael’s seat, not just the workers’. Give the seat a strong one — <code>gpt-oss:120b</code> or <code>llama3.3:70b</code> locally (64–96 GB), or a frontier provider model like <code>openrouter/deepseek/deepseek-v4-flash</code>. Sub-8B models are great workers but thin for orchestration.</p>
+<figure class="fig fig-inline"><img src="/blog/assets/media/run-munder-difflin-on-open-models/note-2.png" alt="Hand-drawn sketch from “Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider”" loading="lazy" decoding="async" /><figcaption>One key, the whole catalog: paste it once, and it reaches an agent only when that agent starts.</figcaption></figure>
+<h2 id="what-should-go-in-michaels-seat" tabindex="-1">What should go in Michael’s seat? <a class="anchor" href="#what-should-go-in-michaels-seat" aria-hidden="true">#</a></h2>
+<p>A strong model. Michael does the reasoning, holds the long context and decides who does what. Locally, that is <code>gpt-oss:120b</code>
+or <code>llama3.3:70b</code> on 64 to 96 GB of memory. Through a provider, DeepSeek V4 Flash or Kimi K2.6. Models under 8B make good
+workers and thin orchestrators.</p>
+<h2 id="what-if-it-does-not-work" tabindex="-1">What if it does not work? <a class="anchor" href="#what-if-it-does-not-work" aria-hidden="true">#</a></h2>
+<ul>
+<li><strong>The model name is rejected.</strong> Check the prefix: <code>local/</code> on OpenCode, <code>ollama/</code> on Crush and Pi. Keep the colon in the
+Ollama tag.</li>
+<li><strong>The agent cannot reach the server.</strong> Make sure <code>ollama serve</code> or LM Studio’s server is running, and that the base URL
+ends in <code>/v1</code>.</li>
+<li><strong>A Pi agent cannot see your model.</strong> Check that <code>~/.pi/agent/models.json</code> is valid JSON, then start the Pi agent again so
+the app copies the new file.</li>
+<li><strong>An engine shows as missing.</strong> Settings → Prerequisites lists which engine commands the app can actually find.</li>
+</ul>
<h2 id="the-bottom-line" tabindex="-1">The bottom line <a class="anchor" href="#the-bottom-line" aria-hidden="true">#</a></h2>
-<p>Open weights turn Munder Difflin from “a harness for three vendors’ CLIs” into “a harness for the whole open ecosystem.” Run it <strong>fully local</strong> when privacy and fixed cost matter and your RAM can hold the model; run it on a <strong>third-party provider</strong> when you want frontier quality or no local hardware at all; hire the <strong>vendor CLI</strong> when you want zero wiring — and mix all three across your floor, agent by agent. The setup is two fields in <strong>Settings → AI Engines</strong> and a pick in Add-Agent, with <strong>Settings → Prerequisites</strong> confirming every engine binary the app can see.</p>
-<p>That’s the promise kept: a virtual office of CLI agents on your own computer, running on models whose weights anyone can read. <a href="https://munderdiffl.in/#install">Download Munder Difflin</a> — free, open source, local-first — and point your favorite open model at it. (On a Mac and want the hardware-by-RAM walkthrough? Read the <a href="/blog/run-munder-difflin-on-a-mac-mini/">Mac Mini setup guide</a>.)</p>
+<p>Open weights turn Munder Difflin from a harness for a few vendors’ CLIs into a harness for the whole open ecosystem. Go local
+when privacy and a fixed cost matter, use a provider when you want the biggest models, and mix both across your floor, agent
+by agent.</p>
+<p><a href="https://munderdiffl.in/">Download Munder Difflin</a>, free and open source, and point your favourite open model at it. On a Mac
+and want the sizing walkthrough? Read the <a href="/blog/run-munder-difflin-on-a-mac-mini/">Mac mini guide</a>.</p>
@@ -412,23 +461,28 @@ ollama serve <span class="token comment"># exposes the OpenAI-
<h2 id="faq">FAQ</h2>
<div class="qa">
- <p class="q">Can Munder Difflin run entirely on open-source models?</p>
- <p class="a">Yes. The OpenCode, Crush, and pi engines all support bring-your-own-key (BYOK) and local models, and the Qwen and Kimi CLIs are first-class engines whose flagship models are open-weight. You can run every agent — workers and Michael himself — on open models like gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, or Kimi: fully local on your own hardware, or through a third-party OSS provider with your own API key.</p>
+ <p class="q">Can Munder Difflin run entirely on open source models?</p>
+ <p class="a">Yes. OpenCode, Crush, Qwen and Pi can all run open weight models, either on your own machine or through a provider with your own API key. You can put open models in every seat on the floor, including Michael&#39;s.</p>
+ </div>
+
+ <div class="qa">
+ <p class="q">What is the difference between running local and using a provider?</p>
+ <p class="a">Local means Ollama, LM Studio or vLLM runs the weights on your machine: private, no per token bill, and limited by your memory. A provider such as OpenRouter or Groq hosts the same open weights on its hardware and bills your own key per token, so you can reach models far too big for a laptop.</p>
</div>
<div class="qa">
- <p class="q">What&#39;s the difference between running local and using a third-party provider?</p>
- <p class="a">Local (Ollama, LM Studio, vLLM) runs the weights on your own machine — fully private, no per-token bill, but bounded by your RAM and GPU. A third-party OSS provider (OpenRouter, Groq, Together, Fireworks, DeepInfra) hosts the same open weights on their hardware and you pay per token with your own key — no local hardware limit, so you can reach the 100B–1T-parameter frontier models a laptop can&#39;t hold.</p>
+ <p class="q">Which open model should I give the orchestrator?</p>
+ <p class="a">A strong one, because Michael does the reasoning and the long context coordination. Locally that means gpt-oss 120B or Llama 3.3 70B on a machine with 64 to 96 GB of memory. Through a provider, DeepSeek V4 Flash or Kimi K2.6 on OpenRouter. Models of 8B and under make fine workers and thin orchestrators.</p>
</div>
<div class="qa">
- <p class="q">Which open model should I pick for the orchestrator seat?</p>
- <p class="a">Michael does the reasoning and long-context coordination, so give him a strong model: locally, gpt-oss-120b or Llama 3.3 70B on a 64–96 GB machine; via a provider, DeepSeek-V4-Flash, GLM-4.6, or Kimi-K2.6 on OpenRouter. Small local models (8B and under) are fine for routine workers but underpowered for orchestration.</p>
+ <p class="q">Does each engine need a different model name?</p>
+ <p class="a">The model id stays the same and only the prefix changes. A local model is local/&lt;tag&gt; on OpenCode and ollama/&lt;tag&gt; on Crush and Pi, and a provider model carries the provider first, like openrouter/openai/gpt-oss-120b.</p>
</div>
<div class="qa">
- <p class="q">Do I need a different model id for each engine?</p>
- <p class="a">No — the upstream model id is the same. All three BYOK engines use a provider/model slug; only the provider prefix and how the key or base-URL is wired differ. For local models, OpenCode uses local/&lt;id&gt; while Crush uses ollama/&lt;id&gt;.</p>
+ <p class="q">How do I run Pi on a local model?</p>
+ <p class="a">Add your local server to Pi&#39;s own config file, ~/.pi/agent/models.json. Munder Difflin copies that file into every Pi agent it starts, so the models you define there are available to your Pi agents. The Pi base URL field in the app stays reserved.</p>
</div>
</section>
@@ -447,7 +501,7 @@ ollama serve <span class="token comment"># exposes the OpenAI-
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#two-routes-your-hardware-or-someone-elses">Two routes: your hardware, or someone else’s</a></li><li class="lvl-2"><a href="#how-the-three-byok-engines-name-a-model">How the three BYOK engines name a model</a></li><li class="lvl-2"><a href="#path-a-fully-local-ollama-lm-studio-vllm">Path A — fully local (Ollama / LM Studio / vLLM)</a></li><li class="lvl-2"><a href="#path-b-a-third-party-oss-provider-byok">Path B — a third-party OSS provider (BYOK)</a></li><li class="lvl-2"><a href="#per-engine-cheat-sheet">Per-engine cheat-sheet</a></li><li class="lvl-2"><a href="#the-bottom-line">The bottom line</a></li>
+ <li class="lvl-2"><a href="#should-you-run-open-models-locally-or-through-a-provider">Should you run open models locally or through a provider?</a></li><li class="lvl-2"><a href="#which-engines-can-run-open-models">Which engines can run open models?</a></li><li class="lvl-2"><a href="#how-do-you-run-a-model-fully-locally">How do you run a model fully locally?</a></li><li class="lvl-2"><a href="#how-do-you-run-open-models-through-a-provider">How do you run open models through a provider?</a></li><li class="lvl-2"><a href="#what-should-go-in-michaels-seat">What should go in Michael’s seat?</a></li><li class="lvl-2"><a href="#what-if-it-does-not-work">What if it does not work?</a></li><li class="lvl-2"><a href="#the-bottom-line">The bottom line</a></li>
</ol>
</aside>
@@ -466,13 +520,13 @@ ollama serve <span class="token comment"># exposes the OpenAI-
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -498,10 +552,10 @@ ollama serve <span class="token comment"># exposes the OpenAI-
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/security-for-ai-coding-agents/index.html b/docs/blog/security-for-ai-coding-agents/index.html
index 37150c34..4c17e52d 100644
--- a/docs/blog/security-for-ai-coding-agents/index.html
+++ b/docs/blog/security-for-ai-coding-agents/index.html
@@ -309,13 +309,13 @@ and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -341,10 +341,10 @@ and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/shareable-agent-roles/index.html b/docs/blog/shareable-agent-roles/index.html
index 93a03797..9c8f699e 100644
--- a/docs/blog/shareable-agent-roles/index.html
+++ b/docs/blog/shareable-agent-roles/index.html
@@ -298,7 +298,7 @@
</div>
<footer class="post-foot wrap"><div class="prevnext">
- <a class="pn prev" href="/blog/deploy-a-blog-writer-agent/"><span class="k">← Older</span><span class="t">Deploy a Blog-Writer Agent: The One That Wrote This Post</span></a>
+ <a class="pn prev" href="/blog/deploy-a-blog-writer-agent/"><span class="k">← Older</span><span class="t">Deploy a Blog Writer Agent: The One That Wrote This Post</span></a>
<a class="pn next" href="/blog/launching-munder-difflin-v0-2-8/"><span class="k">Newer →</span><span class="t">Launching Munder Difflin v0.2.8: Shareable Hires</span></a>
</div>
<section class="related">
diff --git a/docs/blog/small-team-ai-agents-playbook/index.html b/docs/blog/small-team-ai-agents-playbook/index.html
index 2f333246..7899a608 100644
--- a/docs/blog/small-team-ai-agents-playbook/index.html
+++ b/docs/blog/small-team-ai-agents-playbook/index.html
@@ -314,13 +314,13 @@ it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -346,10 +346,10 @@ it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/tags/agent-design/index.html b/docs/blog/tags/agent-design/index.html
index afb1fc5c..7b0a0949 100644
--- a/docs/blog/tags/agent-design/index.html
+++ b/docs/blog/tags/agent-design/index.html
@@ -100,10 +100,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/tags/automation/index.html b/docs/blog/tags/automation/index.html
index 4aea62e5..4c740860 100644
--- a/docs/blog/tags/automation/index.html
+++ b/docs/blog/tags/automation/index.html
@@ -116,13 +116,13 @@
<div class="meta">
<time datetime="2026-06-10">Jun 10, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
- <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog-Writer Agent: The One That Wrote This Post</a></h3>
- <p class="dek">How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</p>
+ <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog Writer Agent: The One That Wrote This Post</a></h3>
+ <p class="dek">How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</p>
<div class="foot">
<span class="tag kind">Use Cases</span>
- <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog-Writer Agent: The One That Wrote This Post">Read →</a>
+ <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog Writer Agent: The One That Wrote This Post">Read →</a>
</div>
</div>
</article>
@@ -199,7 +199,7 @@
<span>14 min</span>
</div>
<h3><a href="/blog/how-to-install-and-use-munder-difflin/">How to Install and Use Munder Difflin: A Beginner&#39;s Guide</a></h3>
- <p class="dek">A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
+ <p class="dek">A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-install-and-use-munder-difflin/" aria-label="Read: How to Install and Use Munder Difflin: A Beginner&#39;s Guide">Read →</a>
diff --git a/docs/blog/tags/byok/index.html b/docs/blog/tags/byok/index.html
index 7daee184..fc7daff6 100644
--- a/docs/blog/tags/byok/index.html
+++ b/docs/blog/tags/byok/index.html
@@ -100,10 +100,10 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/run-munder-difflin-on-a-mac-mini/">Run Munder Difflin Locally on a Mac Mini</a></h3>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/run-munder-difflin-on-a-mac-mini/" aria-label="Read: Run Munder Difflin Locally on a Mac Mini">Read →</a>
diff --git a/docs/blog/tags/claude-code/index.html b/docs/blog/tags/claude-code/index.html
index 1864f7b5..f4b90ec1 100644
--- a/docs/blog/tags/claude-code/index.html
+++ b/docs/blog/tags/claude-code/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
@@ -263,7 +263,7 @@
<span>6 min</span>
</div>
<h3><a href="/blog/best-claude-code-multi-agent-tools/">The Best Tools to Run Multiple Claude Code Agents (2026)</a></h3>
- <p class="dek">An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared.</p>
+ <p class="dek">An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
<a class="more" href="/blog/best-claude-code-multi-agent-tools/" aria-label="Read: The Best Tools to Run Multiple Claude Code Agents (2026)">Read →</a>
@@ -788,13 +788,13 @@
<div class="meta">
<time datetime="2026-05-22">May 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>2 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? (Plain-English Guide)</a></h3>
- <p class="dek">A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework.</p>
+ <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? A Plain English Guide</a></h3>
+ <p class="dek">A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one.</p>
<div class="foot">
<span class="tag kind">Concepts</span>
- <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? (Plain-English Guide)">Read →</a>
+ <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? A Plain English Guide">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/cli-agents/index.html b/docs/blog/tags/cli-agents/index.html
index c55550af..6b70186c 100644
--- a/docs/blog/tags/cli-agents/index.html
+++ b/docs/blog/tags/cli-agents/index.html
@@ -132,13 +132,13 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>6 min</span>
</div>
- <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</a></h3>
- <p class="dek">Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</p>
+ <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</a></h3>
+ <p class="dek">Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider">Read →</a>
+ <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open Source Models: Fully Local or Through a Provider">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/command-center/index.html b/docs/blog/tags/command-center/index.html
index 061ae8c3..9c2f7d2b 100644
--- a/docs/blog/tags/command-center/index.html
+++ b/docs/blog/tags/command-center/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/comparisons/index.html b/docs/blog/tags/comparisons/index.html
index 1310a263..d1870b4d 100644
--- a/docs/blog/tags/comparisons/index.html
+++ b/docs/blog/tags/comparisons/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
@@ -215,7 +215,7 @@
<span>6 min</span>
</div>
<h3><a href="/blog/best-claude-code-multi-agent-tools/">The Best Tools to Run Multiple Claude Code Agents (2026)</a></h3>
- <p class="dek">An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared.</p>
+ <p class="dek">An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
<a class="more" href="/blog/best-claude-code-multi-agent-tools/" aria-label="Read: The Best Tools to Run Multiple Claude Code Agents (2026)">Read →</a>
diff --git a/docs/blog/tags/concepts/index.html b/docs/blog/tags/concepts/index.html
index 81f79781..0a358860 100644
--- a/docs/blog/tags/concepts/index.html
+++ b/docs/blog/tags/concepts/index.html
@@ -372,13 +372,13 @@
<div class="meta">
<time datetime="2026-05-22">May 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>2 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? (Plain-English Guide)</a></h3>
- <p class="dek">A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework.</p>
+ <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? A Plain English Guide</a></h3>
+ <p class="dek">A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one.</p>
<div class="foot">
<span class="tag kind">Concepts</span>
- <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? (Plain-English Guide)">Read →</a>
+ <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? A Plain English Guide">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/content/index.html b/docs/blog/tags/content/index.html
index 9ced47f6..400cabce 100644
--- a/docs/blog/tags/content/index.html
+++ b/docs/blog/tags/content/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-06-10">Jun 10, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
- <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog-Writer Agent: The One That Wrote This Post</a></h3>
- <p class="dek">How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</p>
+ <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog Writer Agent: The One That Wrote This Post</a></h3>
+ <p class="dek">How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</p>
<div class="foot">
<span class="tag kind">Use Cases</span>
- <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog-Writer Agent: The One That Wrote This Post">Read →</a>
+ <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog Writer Agent: The One That Wrote This Post">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/cost/index.html b/docs/blog/tags/cost/index.html
index 809741c4..786abc06 100644
--- a/docs/blog/tags/cost/index.html
+++ b/docs/blog/tags/cost/index.html
@@ -116,13 +116,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/getting-started/index.html b/docs/blog/tags/getting-started/index.html
index 0218d8ac..8c43a2ca 100644
--- a/docs/blog/tags/getting-started/index.html
+++ b/docs/blog/tags/getting-started/index.html
@@ -100,10 +100,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
@@ -116,10 +116,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/your-first-hour-with-munder-difflin/">Your First Hour With Munder Difflin</a></h3>
- <p class="dek">A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</p>
+ <p class="dek">A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/your-first-hour-with-munder-difflin/" aria-label="Read: Your First Hour With Munder Difflin">Read →</a>
@@ -135,7 +135,7 @@
<span>14 min</span>
</div>
<h3><a href="/blog/how-to-install-and-use-munder-difflin/">How to Install and Use Munder Difflin: A Beginner&#39;s Guide</a></h3>
- <p class="dek">A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
+ <p class="dek">A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-install-and-use-munder-difflin/" aria-label="Read: How to Install and Use Munder Difflin: A Beginner&#39;s Guide">Read →</a>
diff --git a/docs/blog/tags/guides/index.html b/docs/blog/tags/guides/index.html
index a905aed2..062e0329 100644
--- a/docs/blog/tags/guides/index.html
+++ b/docs/blog/tags/guides/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -132,10 +132,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
@@ -196,10 +196,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/your-first-hour-with-munder-difflin/">Your First Hour With Munder Difflin</a></h3>
- <p class="dek">A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</p>
+ <p class="dek">A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/your-first-hour-with-munder-difflin/" aria-label="Read: Your First Hour With Munder Difflin">Read →</a>
@@ -212,10 +212,10 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/run-munder-difflin-on-a-mac-mini/">Run Munder Difflin Locally on a Mac Mini</a></h3>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/run-munder-difflin-on-a-mac-mini/" aria-label="Read: Run Munder Difflin Locally on a Mac Mini">Read →</a>
@@ -228,13 +228,13 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>6 min</span>
</div>
- <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</a></h3>
- <p class="dek">Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</p>
+ <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</a></h3>
+ <p class="dek">Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider">Read →</a>
+ <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open Source Models: Fully Local or Through a Provider">Read →</a>
</div>
</div>
</article>
@@ -295,7 +295,7 @@
<span>14 min</span>
</div>
<h3><a href="/blog/how-to-install-and-use-munder-difflin/">How to Install and Use Munder Difflin: A Beginner&#39;s Guide</a></h3>
- <p class="dek">A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
+ <p class="dek">A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-install-and-use-munder-difflin/" aria-label="Read: How to Install and Use Munder Difflin: A Beginner&#39;s Guide">Read →</a>
diff --git a/docs/blog/tags/kanban/index.html b/docs/blog/tags/kanban/index.html
index abdafc73..e8c7cf4b 100644
--- a/docs/blog/tags/kanban/index.html
+++ b/docs/blog/tags/kanban/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/lm-studio/index.html b/docs/blog/tags/lm-studio/index.html
index 1825cc40..9c137c46 100644
--- a/docs/blog/tags/lm-studio/index.html
+++ b/docs/blog/tags/lm-studio/index.html
@@ -100,10 +100,10 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/run-munder-difflin-on-a-mac-mini/">Run Munder Difflin Locally on a Mac Mini</a></h3>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/run-munder-difflin-on-a-mac-mini/" aria-label="Read: Run Munder Difflin Locally on a Mac Mini">Read →</a>
diff --git a/docs/blog/tags/local-first/index.html b/docs/blog/tags/local-first/index.html
index 21168e78..ae47b213 100644
--- a/docs/blog/tags/local-first/index.html
+++ b/docs/blog/tags/local-first/index.html
@@ -148,10 +148,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/your-first-hour-with-munder-difflin/">Your First Hour With Munder Difflin</a></h3>
- <p class="dek">A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</p>
+ <p class="dek">A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/your-first-hour-with-munder-difflin/" aria-label="Read: Your First Hour With Munder Difflin">Read →</a>
@@ -164,10 +164,10 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/run-munder-difflin-on-a-mac-mini/">Run Munder Difflin Locally on a Mac Mini</a></h3>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/run-munder-difflin-on-a-mac-mini/" aria-label="Read: Run Munder Difflin Locally on a Mac Mini">Read →</a>
@@ -180,13 +180,13 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>6 min</span>
</div>
- <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</a></h3>
- <p class="dek">Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</p>
+ <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</a></h3>
+ <p class="dek">Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider">Read →</a>
+ <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open Source Models: Fully Local or Through a Provider">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/mac-mini/index.html b/docs/blog/tags/mac-mini/index.html
index d272418a..5e0caf58 100644
--- a/docs/blog/tags/mac-mini/index.html
+++ b/docs/blog/tags/mac-mini/index.html
@@ -100,10 +100,10 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/run-munder-difflin-on-a-mac-mini/">Run Munder Difflin Locally on a Mac Mini</a></h3>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/run-munder-difflin-on-a-mac-mini/" aria-label="Read: Run Munder Difflin Locally on a Mac Mini">Read →</a>
diff --git a/docs/blog/tags/multi-agent/index.html b/docs/blog/tags/multi-agent/index.html
index fca6c236..67e46ccf 100644
--- a/docs/blog/tags/multi-agent/index.html
+++ b/docs/blog/tags/multi-agent/index.html
@@ -180,13 +180,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
@@ -244,10 +244,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
@@ -404,10 +404,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/your-first-hour-with-munder-difflin/">Your First Hour With Munder Difflin</a></h3>
- <p class="dek">A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</p>
+ <p class="dek">A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/your-first-hour-with-munder-difflin/" aria-label="Read: Your First Hour With Munder Difflin">Read →</a>
@@ -484,13 +484,13 @@
<div class="meta">
<time datetime="2026-06-10">Jun 10, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
- <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog-Writer Agent: The One That Wrote This Post</a></h3>
- <p class="dek">How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</p>
+ <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog Writer Agent: The One That Wrote This Post</a></h3>
+ <p class="dek">How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</p>
<div class="foot">
<span class="tag kind">Use Cases</span>
- <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog-Writer Agent: The One That Wrote This Post">Read →</a>
+ <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog Writer Agent: The One That Wrote This Post">Read →</a>
</div>
</div>
</article>
@@ -695,7 +695,7 @@
<span>6 min</span>
</div>
<h3><a href="/blog/best-claude-code-multi-agent-tools/">The Best Tools to Run Multiple Claude Code Agents (2026)</a></h3>
- <p class="dek">An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared.</p>
+ <p class="dek">An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
<a class="more" href="/blog/best-claude-code-multi-agent-tools/" aria-label="Read: The Best Tools to Run Multiple Claude Code Agents (2026)">Read →</a>
@@ -1444,13 +1444,13 @@
<div class="meta">
<time datetime="2026-05-22">May 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>2 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? (Plain-English Guide)</a></h3>
- <p class="dek">A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework.</p>
+ <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? A Plain English Guide</a></h3>
+ <p class="dek">A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one.</p>
<div class="foot">
<span class="tag kind">Concepts</span>
- <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? (Plain-English Guide)">Read →</a>
+ <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? A Plain English Guide">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/non-technical/index.html b/docs/blog/tags/non-technical/index.html
index 65d50d9b..f01bfbde 100644
--- a/docs/blog/tags/non-technical/index.html
+++ b/docs/blog/tags/non-technical/index.html
@@ -103,7 +103,7 @@
<span>14 min</span>
</div>
<h3><a href="/blog/how-to-install-and-use-munder-difflin/">How to Install and Use Munder Difflin: A Beginner&#39;s Guide</a></h3>
- <p class="dek">A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
+ <p class="dek">A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-install-and-use-munder-difflin/" aria-label="Read: How to Install and Use Munder Difflin: A Beginner&#39;s Guide">Read →</a>
diff --git a/docs/blog/tags/observability/index.html b/docs/blog/tags/observability/index.html
index 244cca27..ec9b89d4 100644
--- a/docs/blog/tags/observability/index.html
+++ b/docs/blog/tags/observability/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/ollama/index.html b/docs/blog/tags/ollama/index.html
index 199c0631..15667cf3 100644
--- a/docs/blog/tags/ollama/index.html
+++ b/docs/blog/tags/ollama/index.html
@@ -100,10 +100,10 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/run-munder-difflin-on-a-mac-mini/">Run Munder Difflin Locally on a Mac Mini</a></h3>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/run-munder-difflin-on-a-mac-mini/" aria-label="Read: Run Munder Difflin Locally on a Mac Mini">Read →</a>
diff --git a/docs/blog/tags/onboarding/index.html b/docs/blog/tags/onboarding/index.html
index 4dd173ce..c8e45d88 100644
--- a/docs/blog/tags/onboarding/index.html
+++ b/docs/blog/tags/onboarding/index.html
@@ -100,10 +100,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/your-first-hour-with-munder-difflin/">Your First Hour With Munder Difflin</a></h3>
- <p class="dek">A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</p>
+ <p class="dek">A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/your-first-hour-with-munder-difflin/" aria-label="Read: Your First Hour With Munder Difflin">Read →</a>
diff --git a/docs/blog/tags/open-source/index.html b/docs/blog/tags/open-source/index.html
index 69eb599f..0114fa80 100644
--- a/docs/blog/tags/open-source/index.html
+++ b/docs/blog/tags/open-source/index.html
@@ -244,13 +244,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
@@ -292,13 +292,13 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>6 min</span>
</div>
- <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</a></h3>
- <p class="dek">Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</p>
+ <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</a></h3>
+ <p class="dek">Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider">Read →</a>
+ <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open Source Models: Fully Local or Through a Provider">Read →</a>
</div>
</div>
</article>
@@ -356,13 +356,13 @@
<div class="meta">
<time datetime="2026-06-10">Jun 10, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
- <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog-Writer Agent: The One That Wrote This Post</a></h3>
- <p class="dek">How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</p>
+ <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog Writer Agent: The One That Wrote This Post</a></h3>
+ <p class="dek">How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</p>
<div class="foot">
<span class="tag kind">Use Cases</span>
- <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog-Writer Agent: The One That Wrote This Post">Read →</a>
+ <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog Writer Agent: The One That Wrote This Post">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/tags/security/index.html b/docs/blog/tags/security/index.html
index 5716b714..34074683 100644
--- a/docs/blog/tags/security/index.html
+++ b/docs/blog/tags/security/index.html
@@ -100,10 +100,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/tags/tools/index.html b/docs/blog/tags/tools/index.html
index f9397ec5..ae0f5861 100644
--- a/docs/blog/tags/tools/index.html
+++ b/docs/blog/tags/tools/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
@@ -135,7 +135,7 @@
<span>6 min</span>
</div>
<h3><a href="/blog/best-claude-code-multi-agent-tools/">The Best Tools to Run Multiple Claude Code Agents (2026)</a></h3>
- <p class="dek">An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared.</p>
+ <p class="dek">An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
<a class="more" href="/blog/best-claude-code-multi-agent-tools/" aria-label="Read: The Best Tools to Run Multiple Claude Code Agents (2026)">Read →</a>
diff --git a/docs/blog/tags/tutorial/index.html b/docs/blog/tags/tutorial/index.html
index 416f99d1..58abdffa 100644
--- a/docs/blog/tags/tutorial/index.html
+++ b/docs/blog/tags/tutorial/index.html
@@ -100,13 +100,13 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>6 min</span>
</div>
- <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</a></h3>
- <p class="dek">Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</p>
+ <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</a></h3>
+ <p class="dek">Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider">Read →</a>
+ <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open Source Models: Fully Local or Through a Provider">Read →</a>
</div>
</div>
</article>
@@ -119,7 +119,7 @@
<span>14 min</span>
</div>
<h3><a href="/blog/how-to-install-and-use-munder-difflin/">How to Install and Use Munder Difflin: A Beginner&#39;s Guide</a></h3>
- <p class="dek">A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
+ <p class="dek">A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-install-and-use-munder-difflin/" aria-label="Read: How to Install and Use Munder Difflin: A Beginner&#39;s Guide">Read →</a>
diff --git a/docs/blog/tags/use-cases/index.html b/docs/blog/tags/use-cases/index.html
index 66f88be2..3c50b4eb 100644
--- a/docs/blog/tags/use-cases/index.html
+++ b/docs/blog/tags/use-cases/index.html
@@ -164,13 +164,13 @@
<div class="meta">
<time datetime="2026-06-10">Jun 10, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
- <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog-Writer Agent: The One That Wrote This Post</a></h3>
- <p class="dek">How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</p>
+ <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog Writer Agent: The One That Wrote This Post</a></h3>
+ <p class="dek">How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</p>
<div class="foot">
<span class="tag kind">Use Cases</span>
- <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog-Writer Agent: The One That Wrote This Post">Read →</a>
+ <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog Writer Agent: The One That Wrote This Post">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/the-lethal-trifecta-for-coding-agents/index.html b/docs/blog/the-lethal-trifecta-for-coding-agents/index.html
index 2e37bf3c..7f310910 100644
--- a/docs/blog/the-lethal-trifecta-for-coding-agents/index.html
+++ b/docs/blog/the-lethal-trifecta-for-coding-agents/index.html
@@ -332,13 +332,13 @@ secrets; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -364,10 +364,10 @@ secrets; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/the-multi-agent-cost-playbook/index.html b/docs/blog/the-multi-agent-cost-playbook/index.html
index ea047e70..5e1eea1c 100644
--- a/docs/blog/the-multi-agent-cost-playbook/index.html
+++ b/docs/blog/the-multi-agent-cost-playbook/index.html
@@ -313,13 +313,13 @@ to run a fleet that doesn’t bankrupt you — it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -345,10 +345,10 @@ to run a fleet that doesn’t bankrupt you — it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/tmux-and-scripts-vs-an-agent-harness/index.html b/docs/blog/tmux-and-scripts-vs-an-agent-harness/index.html
index 5abdb7c3..1dfb82e4 100644
--- a/docs/blog/tmux-and-scripts-vs-an-agent-harness/index.html
+++ b/docs/blog/tmux-and-scripts-vs-an-agent-harness/index.html
@@ -269,13 +269,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/topics/comparisons/index.html b/docs/blog/topics/comparisons/index.html
index 70e0253c..ef8d9c01 100644
--- a/docs/blog/topics/comparisons/index.html
+++ b/docs/blog/topics/comparisons/index.html
@@ -126,13 +126,13 @@
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
@@ -241,7 +241,7 @@
<span>6 min</span>
</div>
<h3><a href="/blog/best-claude-code-multi-agent-tools/">The Best Tools to Run Multiple Claude Code Agents (2026)</a></h3>
- <p class="dek">An honest 2026 roundup of tools to run multiple Claude Code agents — Claude Squad, Conductor, Crystal, vibe-kanban, and Munder Difflin — compared.</p>
+ <p class="dek">An honest September 2026 roundup of tools for running several Claude Code agents: Claude Code agent teams, Claude Squad, Conductor, Nimbalyst, Emdash, Vibe Kanban and Munder Difflin, and how to pick one.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
<a class="more" href="/blog/best-claude-code-multi-agent-tools/" aria-label="Read: The Best Tools to Run Multiple Claude Code Agents (2026)">Read →</a>
diff --git a/docs/blog/topics/concepts/index.html b/docs/blog/topics/concepts/index.html
index 9cdae22e..cecb274c 100644
--- a/docs/blog/topics/concepts/index.html
+++ b/docs/blog/topics/concepts/index.html
@@ -446,13 +446,13 @@
<div class="meta">
<time datetime="2026-05-22">May 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>2 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? (Plain-English Guide)</a></h3>
- <p class="dek">A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework.</p>
+ <h3><a href="/blog/what-is-a-multi-agent-harness/">What Is a Multi-Agent Harness? A Plain English Guide</a></h3>
+ <p class="dek">A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one.</p>
<div class="foot">
<span class="tag kind">Concepts</span>
- <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? (Plain-English Guide)">Read →</a>
+ <a class="more" href="/blog/what-is-a-multi-agent-harness/" aria-label="Read: What Is a Multi-Agent Harness? A Plain English Guide">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/topics/guides/index.html b/docs/blog/topics/guides/index.html
index c62982c5..1899c4b8 100644
--- a/docs/blog/topics/guides/index.html
+++ b/docs/blog/topics/guides/index.html
@@ -126,13 +126,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -158,10 +158,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
@@ -222,10 +222,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/your-first-hour-with-munder-difflin/">Your First Hour With Munder Difflin</a></h3>
- <p class="dek">A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</p>
+ <p class="dek">A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/your-first-hour-with-munder-difflin/" aria-label="Read: Your First Hour With Munder Difflin">Read →</a>
@@ -238,10 +238,10 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>5 min</span>
</div>
<h3><a href="/blog/run-munder-difflin-on-a-mac-mini/">Run Munder Difflin Locally on a Mac Mini</a></h3>
- <p class="dek">A step-by-step guide to running a whole Munder Difflin hive offline on an Apple Silicon Mac Mini — how to size models to your unified memory, install Ollama or LM Studio, and wire OpenCode and Crush to a local endpoint.</p>
+ <p class="dek">Run a whole Munder Difflin office offline on an Apple silicon Mac mini: size the model to your unified memory, install Ollama or LM Studio, and wire OpenCode, Crush, Qwen and Pi to it. Current as of Munder Difflin 0.5.2 and the M6 and M5 Pro Mac mini.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/run-munder-difflin-on-a-mac-mini/" aria-label="Read: Run Munder Difflin Locally on a Mac Mini">Read →</a>
@@ -254,13 +254,13 @@
<div class="meta">
<time datetime="2026-06-22">Jun 22, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>8 min</span>
+ <span>6 min</span>
</div>
- <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider</a></h3>
- <p class="dek">Munder Difflin runs your agent floor on open-weight models — gpt-oss, Qwen3, DeepSeek, Llama, Mistral, GLM, Kimi — fully local (Ollama/LM Studio/vLLM), through a third-party OSS provider, or on the model-maker&#39;s own CLI. Here&#39;s the wiring, current as of v0.4.4.</p>
+ <h3><a href="/blog/run-munder-difflin-on-open-models/">Run Munder Difflin on Open Source Models: Fully Local or Through a Provider</a></h3>
+ <p class="dek">Munder Difflin can run your whole agent floor on open weight models like gpt-oss, Qwen3, DeepSeek, Llama, GLM and Kimi: fully local with Ollama, LM Studio or vLLM, or through a provider with your own key. The wiring for each engine, current as of 0.5.2.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open-Source Models — Fully Local or via a Third-Party Provider">Read →</a>
+ <a class="more" href="/blog/run-munder-difflin-on-open-models/" aria-label="Read: Run Munder Difflin on Open Source Models: Fully Local or Through a Provider">Read →</a>
</div>
</div>
</article>
@@ -337,7 +337,7 @@
<span>14 min</span>
</div>
<h3><a href="/blog/how-to-install-and-use-munder-difflin/">How to Install and Use Munder Difflin: A Beginner&#39;s Guide</a></h3>
- <p class="dek">A plain-language guide to Munder Difflin v0.4.6. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
+ <p class="dek">A plain language guide to Munder Difflin 0.5.2. What a coding agent is, which AI engine to pick based on what you already pay for, the one terminal command you&#39;ll ever run, and how to install and set it up on macOS, Windows or Linux.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-install-and-use-munder-difflin/" aria-label="Read: How to Install and Use Munder Difflin: A Beginner&#39;s Guide">Read →</a>
diff --git a/docs/blog/topics/use-cases/index.html b/docs/blog/topics/use-cases/index.html
index fdd3a29f..bbf35c69 100644
--- a/docs/blog/topics/use-cases/index.html
+++ b/docs/blog/topics/use-cases/index.html
@@ -190,13 +190,13 @@
<div class="meta">
<time datetime="2026-06-10">Jun 10, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>6 min</span>
+ <span>5 min</span>
</div>
- <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog-Writer Agent: The One That Wrote This Post</a></h3>
- <p class="dek">How Munder Difflin&#39;s blog is written by an automated writer agent in the hive — drafts in a worktree, single-committer integration, Eleventy build, human-gated deploy, and now hand-drawn illustrations from the same office. Build your own.</p>
+ <h3><a href="/blog/deploy-a-blog-writer-agent/">Deploy a Blog Writer Agent: The One That Wrote This Post</a></h3>
+ <p class="dek">How the Munder Difflin blog is written by agents in our own office: a brief, a house style, a draft in its own worktree, drawings made in code, a pull request, and one human merge. The pipeline, and how to build your own.</p>
<div class="foot">
<span class="tag kind">Use Cases</span>
- <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog-Writer Agent: The One That Wrote This Post">Read →</a>
+ <a class="more" href="/blog/deploy-a-blog-writer-agent/" aria-label="Read: Deploy a Blog Writer Agent: The One That Wrote This Post">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/trigger-ai-agents-from-github-webhook/index.html b/docs/blog/trigger-ai-agents-from-github-webhook/index.html
index 467b35ac..1520e2e7 100644
--- a/docs/blog/trigger-ai-agents-from-github-webhook/index.html
+++ b/docs/blog/trigger-ai-agents-from-github-webhook/index.html
@@ -362,13 +362,13 @@ open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -394,10 +394,10 @@ open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/trigger-ai-agents-from-slack/index.html b/docs/blog/trigger-ai-agents-from-slack/index.html
index 6e3c5073..1ac81290 100644
--- a/docs/blog/trigger-ai-agents-from-slack/index.html
+++ b/docs/blog/trigger-ai-agents-from-slack/index.html
@@ -309,13 +309,13 @@ it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -341,10 +341,10 @@ it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/vibe-kanban-alternative/index.html b/docs/blog/vibe-kanban-alternative/index.html
index 7e45028c..8dcb5b8e 100644
--- a/docs/blog/vibe-kanban-alternative/index.html
+++ b/docs/blog/vibe-kanban-alternative/index.html
@@ -259,13 +259,13 @@ routing happen for you — free and open source.</p>
<div class="meta">
<time datetime="2026-08-06">Aug 6, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>3 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?</a></h3>
- <p class="dek">An honest comparison of YC&#39;s qm and Munder Difflin: both orchestrate fleets of CLI coding agents with security postures, scheduled background work, and Slack — one as a headless multiplayer harness for teams, one as a local-first desktop office with real terminals, voice, and a built-in IDE.</p>
+ <h3><a href="/blog/qm-vs-munder-difflin/">qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?</a></h3>
+ <p class="dek">An honest September 2026 comparison of YC&#39;s qm and Munder Difflin. qm is a multiplayer agent harness a company deploys for Slack and the web. Munder Difflin runs an office of your clones on your own machine, with Teams to connect them.</p>
<div class="foot">
<span class="tag kind">Comparisons</span>
- <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or a Local-First Agent Office?">Read →</a>
+ <a class="more" href="/blog/qm-vs-munder-difflin/" aria-label="Read: qm vs Munder Difflin: YC&#39;s Multiplayer Harness or an Office of Your Clones?">Read →</a>
</div>
</div>
</article>
diff --git a/docs/blog/what-is-a-multi-agent-harness/index.html b/docs/blog/what-is-a-multi-agent-harness/index.html
index 10f1656c..c6f95006 100644
--- a/docs/blog/what-is-a-multi-agent-harness/index.html
+++ b/docs/blog/what-is-a-multi-agent-harness/index.html
@@ -3,8 +3,8 @@
<head>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
-<title>What Is a Multi-Agent Harness? (Plain-English Guide) — Munder Difflin Blog</title>
-<meta name="description" content="A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework." />
+<title>What Is a Multi-Agent Harness? A Plain English Guide — Munder Difflin Blog</title>
+<meta name="description" content="A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one." />
<link rel="canonical" href="https://munderdiffl.in/blog/what-is-a-multi-agent-harness/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -14,8 +14,8 @@
<!-- Open Graph -->
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
-<meta property="og:title" content="What Is a Multi-Agent Harness? (Plain-English Guide) — Munder Difflin Blog" />
-<meta property="og:description" content="A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework." />
+<meta property="og:title" content="What Is a Multi-Agent Harness? A Plain English Guide — Munder Difflin Blog" />
+<meta property="og:description" content="A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one." />
<meta property="og:url" content="https://munderdiffl.in/blog/what-is-a-multi-agent-harness/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -28,8 +28,8 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
-<meta name="twitter:title" content="What Is a Multi-Agent Harness? (Plain-English Guide) — Munder Difflin Blog" />
-<meta name="twitter:description" content="A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework." />
+<meta name="twitter:title" content="What Is a Multi-Agent Harness? A Plain English Guide — Munder Difflin Blog" />
+<meta name="twitter:description" content="A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -42,11 +42,11 @@
{
"@context": "https://schema.org",
"@type": "BlogPosting",
- "headline": "What Is a Multi-Agent Harness? (Plain-English Guide)",
- "description": "A multi-agent harness coordinates several AI coding agents into one team — here's what that means, and how it differs from a single agent or a framework.",
+ "headline": "What Is a Multi-Agent Harness? A Plain English Guide",
+ "description": "A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-05-22T00:00:00.000Z",
- "dateModified": "2026-05-22T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -54,7 +54,7 @@
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/what-is-a-multi-agent-harness/" },
- "keywords": "claude code multi-agent, multi-agent harness, multi-agent ai framework, ai agent harness",
+ "keywords": "claude code multi-agent, multi-agent harness, what is an agent harness, multi-agent ai framework, ai agent harness, claude code agent teams vs harness",
"articleSection": "Concepts",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/what-is-a-multi-agent-harness/"
@@ -67,7 +67,7 @@
"itemListElement": [
{ "@type": "ListItem", "position": 1, "name": "Home", "item": "https://munderdiffl.in/" },
{ "@type": "ListItem", "position": 2, "name": "Blog", "item": "https://munderdiffl.in/blog/" },
- { "@type": "ListItem", "position": 3, "name": "What Is a Multi-Agent Harness? (Plain-English Guide)", "item": "https://munderdiffl.in/blog/what-is-a-multi-agent-harness/" }
+ { "@type": "ListItem", "position": 3, "name": "What Is a Multi-Agent Harness? A Plain English Guide", "item": "https://munderdiffl.in/blog/what-is-a-multi-agent-harness/" }
]
}
</script>
@@ -77,9 +77,11 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "What is a multi-agent harness?", "acceptedAnswer": { "@type": "Answer", "text": "A multi-agent harness is software that runs several AI agents at once and coordinates them — giving each a role, letting them message each other, sharing memory, and routing work — so they act as one team instead of isolated sessions." } },
- { "@type": "Question", "name": "How is a harness different from a framework like LangGraph or CrewAI?", "acceptedAnswer": { "@type": "Answer", "text": "A framework is a library you write an agent app with. A harness wraps agents you already run (like Claude Code terminals) and adds coordination — messaging, memory, orchestration, and visibility — without you rebuilding your agent from scratch." } },
- { "@type": "Question", "name": "Do I need a multi-agent harness to use Claude Code?", "acceptedAnswer": { "@type": "Answer", "text": "No. One Claude Code session is plenty for many tasks. A harness helps once you're running several at once and the coordination overhead (who does what, who knows what) starts to cost you time." } }
+ { "@type": "Question", "name": "What is a multi-agent harness?", "acceptedAnswer": { "@type": "Answer", "text": "Software that runs several AI agents at the same time and makes them work as one team. Each agent gets a role, agents pass messages to each other, they share long term memory, and a coordinator hands out the work." } },
+ { "@type": "Question", "name": "How is a harness different from a framework like LangGraph or CrewAI?", "acceptedAnswer": { "@type": "Answer", "text": "A framework is a library you write a new agent application with. A harness wraps agents you already run, like Claude Code or Codex in a terminal, and adds the coordination around them. You do not rebuild anything." } },
+ { "@type": "Question", "name": "Is Claude Code's agent teams feature a multi-agent harness?", "acceptedAnswer": { "@type": "Answer", "text": "It covers part of the idea inside Claude Code. Agent teams are experimental and off by default. A lead session spawns teammates that share a task list and message each other, and a team belongs to that one session. A harness keeps agents working across sessions, remembers between them, and can mix CLIs from different vendors." } },
+ { "@type": "Question", "name": "Do I need a multi-agent harness to use Claude Code?", "acceptedAnswer": { "@type": "Answer", "text": "No. One session is plenty for most tasks. A harness starts paying for itself when you run three or more agents at once and keeping them coordinated costs you more time than the work does." } },
+ { "@type": "Question", "name": "Is Munder Difflin a multi-agent harness?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. It is a free, open source harness that runs twelve terminal coding CLIs, including Claude Code, Codex, Gemini CLI and Copilot, on your own machine, with your clone Michael coordinating them." } }
]
}
@@ -148,15 +150,15 @@
<span>/</span>
<a href="/blog/topics/concepts/">Concepts</a>
</nav>
- <h1>What Is a Multi-Agent Harness? (Plain-English Guide)</h1>
- <p class="dek">A multi-agent harness coordinates several AI coding agents into one team — here&#39;s what that means, and how it differs from a single agent or a framework.</p>
+ <h1>What Is a Multi-Agent Harness? A Plain English Guide</h1>
+ <p class="dek">A multi-agent harness runs several AI coding agents as one team, with roles, messages, shared memory and an orchestrator. What it is, how it differs from a framework, subagents or agent teams, and when you need one.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-05-22">May 22, 2026</time>
<span class="dot" aria-hidden="true">·</span>
- <span>2 min read</span>
+ <span>4 min read</span>
<span class="tag kind">Concepts</span>
</div>
</header>
@@ -165,60 +167,78 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p><strong>A multi-agent harness</strong> is
-software that runs several AI agents at once and makes them act like a team: each gets a role, they
-message each other, share long-term memory, and a coordinator routes work between them. It's the
-layer that turns "five chat windows" into "one office."</p></div>
-<p>If you’ve run more than one AI coding agent at the same time, you’ve felt the problem: two agents
-edit the same file, neither remembers what the other did, and you become the human message bus
-alt-tabbing between windows. A <strong>multi-agent harness</strong> is the software that fixes that.</p>
-<h2 id="a-multi-agent-harness-in-one-sentence" tabindex="-1">A multi-agent harness, in one sentence <a class="anchor" href="#a-multi-agent-harness-in-one-sentence" aria-hidden="true">#</a></h2>
-<blockquote>
-<p>A multi-agent harness runs several AI agents concurrently and coordinates them — roles, messaging,
-shared memory, and work routing — so they behave as a single team.</p>
-</blockquote>
-<p>That’s the whole idea. Everything else is detail about <em>how</em> the coordination happens.</p>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p>A <strong>multi-agent harness</strong> runs
+several AI coding agents at once and makes them behave like a team: each has a role, they message each
+other, they share memory, and a coordinator routes the work. It is the layer that turns five terminal
+windows into one office.</p></div>
+<p>If you have ever run two AI coding agents at the same time, you already know the problem. They edit the
+same file, neither knows what the other did, and you become the message bus, copying context from one
+window to the next. A <strong>multi-agent harness</strong> is the software that takes that job off your hands.</p>
+<h2 id="what-is-a-multi-agent-harness" tabindex="-1">What is a multi-agent harness? <a class="anchor" href="#what-is-a-multi-agent-harness" aria-hidden="true">#</a></h2>
+<p>It is software that runs several AI agents at the same time and coordinates them. That coordination comes
+down to four things: roles, messaging, shared memory and routing. Everything else is detail about how a
+particular harness does those four things.</p>
+<p>The easiest picture is an office. The agents are the staff. The harness is the building: desks, mailboxes,
+a filing cabinet everyone can read, and a manager who decides who does what.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/what-is-a-multi-agent-harness/note-1.png" alt="Hand-drawn sketch from “What Is a Multi-Agent Harness? (Plain-English Guide)”" loading="lazy" decoding="async" /></figure>
-<h2 id="what-it-adds-on-top-of-a-single-agent" tabindex="-1">What it adds on top of a single agent <a class="anchor" href="#what-it-adds-on-top-of-a-single-agent" aria-hidden="true">#</a></h2>
-<p>A single agent is a loop: read context, take an action, repeat. A harness wraps <strong>many</strong> of those
-loops and adds the parts a lone agent doesn’t have:</p>
+<h2 id="what-does-a-harness-add-on-top-of-a-single-agent" tabindex="-1">What does a harness add on top of a single agent? <a class="anchor" href="#what-does-a-harness-add-on-top-of-a-single-agent" aria-hidden="true">#</a></h2>
+<p>Five things a lone agent does not have. A single coding agent is a loop: read the context, act, check the
+result, repeat. A harness wraps many of those loops and adds:</p>
<ul>
-<li><strong>Roles.</strong> Each agent gets a job (researcher, builder, reviewer) instead of all of them doing
-everything.</li>
-<li><strong>Messaging.</strong> Agents pass information to each other directly, rather than through you.</li>
-<li><strong>Shared memory.</strong> Durable, cross-session knowledge so agent B can use what agent A learned.</li>
-<li><strong>Orchestration.</strong> A coordinator that decomposes your intent and assigns work — in Munder Difflin
-that’s the <a href="https://munderdiffl.in/#how">GOD orchestrator</a>, an agent you talk to in plain language.</li>
-<li><strong>Visibility.</strong> A way to <em>see</em> what the team is doing, so it’s not a black box.</li>
+<li><strong>Roles.</strong> A researcher, a builder and a reviewer, instead of five generalists stepping on each other.</li>
+<li><strong>Messaging.</strong> Agents hand findings to each other directly, not through you.</li>
+<li><strong>Shared memory.</strong> What one agent learned on Monday is still there for another agent on Thursday.</li>
+<li><strong>Orchestration.</strong> One coordinator turns your request into tasks and assigns them. In Munder Difflin that
+coordinator is Michael, a clone of you that you talk to in plain language.</li>
+<li><strong>Visibility.</strong> A way to see what the whole team is doing without tailing five terminals.</li>
</ul>
-<figure class="fig fig-inline"><img src="/blog/assets/media/what-is-a-multi-agent-harness/note-2.png" alt="Hand-drawn sketch from “What Is a Multi-Agent Harness? (Plain-English Guide)”" loading="lazy" decoding="async" /></figure>
-<h2 id="harness-vs-framework-vs-subagents" tabindex="-1">Harness vs. framework vs. subagents <a class="anchor" href="#harness-vs-framework-vs-subagents" aria-hidden="true">#</a></h2>
-<p>These three get conflated. The difference is what you bring to the table:</p>
+<h2 id="what-is-the-difference-between-a-harness-a-framework-subagents-and-agent-teams" tabindex="-1">What is the difference between a harness, a framework, subagents and agent teams? <a class="anchor" href="#what-is-the-difference-between-a-harness-a-framework-subagents-and-agent-teams" aria-hidden="true">#</a></h2>
+<p>What you bring, and how long the team lasts. These four get mixed up constantly, so here they are side by side.</p>
<h3 id="framework" tabindex="-1">Framework <a class="anchor" href="#framework" aria-hidden="true">#</a></h3>
-<p>A library (LangGraph, CrewAI, AutoGen) you <strong>build an agent application with</strong>. You write the graph,
-the tools, the prompts. Powerful, but it’s a from-scratch build.</p>
+<p>A library such as LangGraph, CrewAI or AutoGen that you <strong>build an agent application with</strong>. You write the
+graph, the tools and the prompts. Powerful, and a build from scratch.</p>
<h3 id="subagents" tabindex="-1">Subagents <a class="anchor" href="#subagents" aria-hidden="true">#</a></h3>
-<p>A single agent <strong>spawning helpers</strong> inside its own run. Useful for fan-out, but the helpers are
-short-lived and scoped to that one parent — no shared memory across your whole workflow.</p>
+<p>One agent <strong>spawning short lived helpers</strong> inside its own run. Great for fanning out a search or a review.
+The helpers report back to their parent and are gone when the job ends.</p>
+<h3 id="agent-teams-inside-claude-code" tabindex="-1">Agent teams inside Claude Code <a class="anchor" href="#agent-teams-inside-claude-code" aria-hidden="true">#</a></h3>
+<p>Claude Code’s own <a href="https://code.claude.com/docs/en/agent-teams">agent teams</a> sit in between, and they are
+still experimental and off by default. A lead session spawns teammates that share a task list and message
+each other directly. A team belongs to a single session, and every teammate is a Claude Code instance.</p>
<h3 id="harness" tabindex="-1">Harness <a class="anchor" href="#harness" aria-hidden="true">#</a></h3>
-<p>Software that <strong>wraps agents you already run</strong> and coordinates them. You don’t rebuild your agent;
-you keep using Claude Code, and the harness adds messaging, memory, orchestration, and a view of the
-floor. That’s the <a href="https://munderdiffl.in/#what">multi-agent harness</a> approach.</p>
-<h2 id="when-you-actually-need-one" tabindex="-1">When you actually need one <a class="anchor" href="#when-you-actually-need-one" aria-hidden="true">#</a></h2>
-<p>You don’t need a harness to use Claude Code. You start wanting one when:</p>
+<p>Software that <strong>wraps the agents you already run</strong> and keeps them working together over time. You keep
+using Claude Code, Codex or Gemini CLI. The harness adds messaging, memory that survives a restart, an
+orchestrator, and a view of the floor. Because it sits outside any single CLI, it can put agents from
+different vendors on the same team.</p>
+<figure class="fig fig-inline"><img src="/blog/assets/media/what-is-a-multi-agent-harness/note-2.png" alt="Hand-drawn sketch from “What Is a Multi-Agent Harness? (Plain-English Guide)”" loading="lazy" decoding="async" /></figure>
+<h2 id="when-do-you-actually-need-one" tabindex="-1">When do you actually need one? <a class="anchor" href="#when-do-you-actually-need-one" aria-hidden="true">#</a></h2>
+<p>When coordinating your agents costs more time than the agents save you. The usual signs:</p>
<ul>
-<li>you’re running <strong>three or more</strong> sessions and losing track of which is doing what,</li>
-<li>you keep <strong>re-explaining context</strong> because each session forgets, or</li>
-<li>two agents <strong>collide</strong> on the same files.</li>
+<li>you run <strong>three or more</strong> sessions and lose track of which one is doing what,</li>
+<li>you keep <strong>explaining the same context</strong> because every session starts from zero,</li>
+<li>two agents <strong>collide</strong> on the same files,</li>
+<li>you want work to <strong>keep moving</strong> while you are in a meeting, with the machine left on.</li>
+</ul>
+<p>If none of that sounds familiar, one session is fine, and you can stop reading here with a clear conscience.</p>
+<h2 id="what-does-a-harness-look-like-in-practice" tabindex="-1">What does a harness look like in practice? <a class="anchor" href="#what-does-a-harness-look-like-in-practice" aria-hidden="true">#</a></h2>
+<p>Here is how <a href="https://munderdiffl.in/">Munder Difflin</a> does it. Every agent is a real terminal CLI running
+on your machine, and twelve are supported: Claude Code, Codex, Gemini CLI, Antigravity, Grok, Kimi Code,
+Qwen, OpenCode, Crush, Pi, Copilot and Cursor. Each agent gets a desk on a 2D office floor, a mailbox and
+long term memory. Michael, your clone, takes your request, breaks it into tasks, hires workers and routes
+messages between them. When something needs a human decision, it lands on the ASK ME board instead of
+vanishing into a scrollback.</p>
+<p>The floor is a simulation and uses no tokens. The real work happens in the terminals underneath, on the
+subscriptions you already pay for and within their normal usage limits.</p>
+<h2 id="where-to-go-next" tabindex="-1">Where to go next <a class="anchor" href="#where-to-go-next" aria-hidden="true">#</a></h2>
+<ul>
+<li><a href="/blog/manage-multiple-claude-code-sessions/">How to manage multiple Claude Code sessions</a> without losing
+track of them.</li>
+<li><a href="/blog/give-claude-code-long-term-memory/">How to give Claude Code long term memory</a> so the team stops
+forgetting.</li>
+<li><a href="/blog/best-claude-code-multi-agent-tools/">The best tools to run multiple Claude Code agents</a>, compared.</li>
</ul>
-<p>If that’s you, the next step is a practical one:
-<a href="/blog/how-to-run-multiple-claude-code-agents/">how to run multiple Claude Code agents</a> without
-losing track, and <a href="/blog/give-claude-code-long-term-memory/">how to give Claude Code long-term memory</a>
-so the team stops forgetting.</p>
<hr>
-<p>Munder Difflin is exactly this: a local, open-source multi-agent harness for Claude Code. If you
-want to watch a coordinated team of agents work an office floor,
-<a href="https://munderdiffl.in/#install">download Munder Difflin</a> — it’s free and MIT-licensed.</p>
+<p>Munder Difflin is free and open source under the MIT license. <a href="https://munderdiffl.in/">Download it</a> for
+macOS, Windows or Linux and watch a coordinated team work an office floor.</p>
@@ -227,17 +247,27 @@ want to watch a coordinated team of agents work an office floor,
<div class="qa">
<p class="q">What is a multi-agent harness?</p>
- <p class="a">A multi-agent harness is software that runs several AI agents at once and coordinates them — giving each a role, letting them message each other, sharing memory, and routing work — so they act as one team instead of isolated sessions.</p>
+ <p class="a">Software that runs several AI agents at the same time and makes them work as one team. Each agent gets a role, agents pass messages to each other, they share long term memory, and a coordinator hands out the work.</p>
</div>
<div class="qa">
<p class="q">How is a harness different from a framework like LangGraph or CrewAI?</p>
- <p class="a">A framework is a library you write an agent app with. A harness wraps agents you already run (like Claude Code terminals) and adds coordination — messaging, memory, orchestration, and visibility — without you rebuilding your agent from scratch.</p>
+ <p class="a">A framework is a library you write a new agent application with. A harness wraps agents you already run, like Claude Code or Codex in a terminal, and adds the coordination around them. You do not rebuild anything.</p>
+ </div>
+
+ <div class="qa">
+ <p class="q">Is Claude Code&#39;s agent teams feature a multi-agent harness?</p>
+ <p class="a">It covers part of the idea inside Claude Code. Agent teams are experimental and off by default. A lead session spawns teammates that share a task list and message each other, and a team belongs to that one session. A harness keeps agents working across sessions, remembers between them, and can mix CLIs from different vendors.</p>
</div>
<div class="qa">
<p class="q">Do I need a multi-agent harness to use Claude Code?</p>
- <p class="a">No. One Claude Code session is plenty for many tasks. A harness helps once you&#39;re running several at once and the coordination overhead (who does what, who knows what) starts to cost you time.</p>
+ <p class="a">No. One session is plenty for most tasks. A harness starts paying for itself when you run three or more agents at once and keeping them coordinated costs you more time than the work does.</p>
+ </div>
+
+ <div class="qa">
+ <p class="q">Is Munder Difflin a multi-agent harness?</p>
+ <p class="a">Yes. It is a free, open source harness that runs twelve terminal coding CLIs, including Claude Code, Codex, Gemini CLI and Copilot, on your own machine, with your clone Michael coordinating them.</p>
</div>
</section>
@@ -256,7 +286,7 @@ want to watch a coordinated team of agents work an office floor,
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#a-multi-agent-harness-in-one-sentence">A multi-agent harness, in one sentence</a></li><li class="lvl-2"><a href="#what-it-adds-on-top-of-a-single-agent">What it adds on top of a single agent</a></li><li class="lvl-2"><a href="#harness-vs-framework-vs-subagents">Harness vs. framework vs. subagents</a></li><li class="lvl-3"><a href="#framework">Framework</a></li><li class="lvl-3"><a href="#subagents">Subagents</a></li><li class="lvl-3"><a href="#harness">Harness</a></li><li class="lvl-2"><a href="#when-you-actually-need-one">When you actually need one</a></li>
+ <li class="lvl-2"><a href="#what-is-a-multi-agent-harness">What is a multi-agent harness?</a></li><li class="lvl-2"><a href="#what-does-a-harness-add-on-top-of-a-single-agent">What does a harness add on top of a single agent?</a></li><li class="lvl-2"><a href="#what-is-the-difference-between-a-harness-a-framework-subagents-and-agent-teams">What is the difference between a harness, a framework, subagents and agent teams?</a></li><li class="lvl-3"><a href="#framework">Framework</a></li><li class="lvl-3"><a href="#subagents">Subagents</a></li><li class="lvl-3"><a href="#agent-teams-inside-claude-code">Agent teams inside Claude Code</a></li><li class="lvl-3"><a href="#harness">Harness</a></li><li class="lvl-2"><a href="#when-do-you-actually-need-one">When do you actually need one?</a></li><li class="lvl-2"><a href="#what-does-a-harness-look-like-in-practice">What does a harness look like in practice?</a></li><li class="lvl-2"><a href="#where-to-go-next">Where to go next</a></li>
</ol>
</aside>
diff --git a/docs/blog/whats-new-in-agentic-ai-june-2026/index.html b/docs/blog/whats-new-in-agentic-ai-june-2026/index.html
index dcc5aa75..747c9792 100644
--- a/docs/blog/whats-new-in-agentic-ai-june-2026/index.html
+++ b/docs/blog/whats-new-in-agentic-ai-june-2026/index.html
@@ -369,13 +369,13 @@ hive run; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -401,10 +401,10 @@ hive run; it’s free and open source.</p>
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/blog/your-first-hour-with-munder-difflin/index.html b/docs/blog/your-first-hour-with-munder-difflin/index.html
index abf74b25..44d9396b 100644
--- a/docs/blog/your-first-hour-with-munder-difflin/index.html
+++ b/docs/blog/your-first-hour-with-munder-difflin/index.html
@@ -4,7 +4,7 @@
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Your First Hour With Munder Difflin — Munder Difflin Blog</title>
-<meta name="description" content="A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running." />
+<meta name="description" content="A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running." />
<link rel="canonical" href="https://munderdiffl.in/blog/your-first-hour-with-munder-difflin/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
@@ -15,7 +15,7 @@
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
<meta property="og:title" content="Your First Hour With Munder Difflin — Munder Difflin Blog" />
-<meta property="og:description" content="A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running." />
+<meta property="og:description" content="A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running." />
<meta property="og:url" content="https://munderdiffl.in/blog/your-first-hour-with-munder-difflin/" />
<meta property="og:image" content="https://munderdiffl.in/media/og.png" />
<meta property="og:locale" content="en_US" />
@@ -31,7 +31,7 @@
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="Your First Hour With Munder Difflin — Munder Difflin Blog" />
-<meta name="twitter:description" content="A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running." />
+<meta name="twitter:description" content="A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running." />
<meta name="twitter:image" content="https://munderdiffl.in/media/og.png" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
@@ -45,10 +45,10 @@
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "Your First Hour With Munder Difflin",
- "description": "A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.",
+ "description": "A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.",
"image": ["https://munderdiffl.in/media/og.png"],
"datePublished": "2026-07-03T00:00:00.000Z",
- "dateModified": "2026-08-20T00:00:00.000Z",
+ "dateModified": "2026-09-10T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
@@ -56,7 +56,7 @@
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/your-first-hour-with-munder-difflin/" },
- "keywords": "munder difflin onboarding, getting started with munder difflin, multi-agent harness tutorial, first hour with a multi-agent harness, ai agent approval queue, scheduled agent missions, munder difflin skills",
+ "keywords": "munder difflin onboarding, getting started with munder difflin, munder difflin tutorial, first hour with a multi-agent harness, munder difflin setup, scheduled agent missions, munder difflin skills",
"articleSection": "Guides",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/your-first-hour-with-munder-difflin/"
@@ -79,12 +79,13 @@
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
- { "@type": "Question", "name": "How long does it take to get Munder Difflin running?", "acceptedAnswer": { "@type": "Answer", "text": "About five minutes. Grab the signed build for macOS, Windows, or Linux from the releases page, or run from source with Node.js 18+. Onboarding validates your setup at the very first step — as of v0.4.4 it checks your home folder immediately instead of failing four steps later, and the Prerequisites page in Settings shows live status for every tool an engine needs, with a button that asks Michael to install what's missing." } },
- { "@type": "Question", "name": "What do I need before installing Munder Difflin?", "acceptedAnswer": { "@type": "Answer", "text": "At least one supported agent CLI — Claude Code, Antigravity, Codex, Grok, Kimi, Qwen, OpenCode, Crush, pi, or GitHub Copilot CLI. If you're missing one (or git, Node, uv, MemPalace), Settings → Prerequisites shows exactly what's absent with the platform-correct install command, and can hand the whole job to Michael." } },
- { "@type": "Question", "name": "Who is Michael and how do I give him work?", "acceptedAnswer": { "@type": "Answer", "text": "Michael is a clone of you — the boss of the agents, while you stay the boss of him. You type a brief into his terminal (or talk to him by voice) and he adjudicates it: creating tasks, assigning them to workers, routing messages between inboxes, and escalating only critical items back to you. He seats himself in his office automatically on first launch." } },
- { "@type": "Question", "name": "Do I have to approve everything the agents do?", "acceptedAnswer": { "@type": "Answer", "text": "No. Michael resolves routine requests himself so the system stays autonomous. Only critical items — spend, destructive operations, scope changes — land in the human-in-the-loop approvals queue for you to act on. A circuit breaker and per-agent token budgets guard the rest." } },
- { "@type": "Question", "name": "How do I review what an agent actually changed?", "acceptedAnswer": { "@type": "Answer", "text": "Open the built-in Monaco IDE from the title-bar IDE button. Its git CHANGES rail lists every modified file, and clicking one opens a read-only side-by-side diff against HEAD. There's a per-agent git tab with status, log, and commit graph — and since v0.4.4 the IDE previews images too, so a design change is as reviewable as a code change." } },
- { "@type": "Question", "name": "Can Munder Difflin keep working after I walk away?", "acceptedAnswer": { "@type": "Answer", "text": "Yes — that's the point. The Triggers tab in the Command Center holds recurring missions with a label, interval, target agent, and body, plus a heartbeat that re-engages the floor when it goes quiet. It's local-first, so the office runs on hardware you already own — and the app keeps itself current, telling you exactly what's in each update before you restart into it." } }
+ { "@type": "Question", "name": "How long does it take to get Munder Difflin running?", "acceptedAnswer": { "@type": "Answer", "text": "About ten minutes if one coding CLI, like Claude Code, Codex or Antigravity, is already installed and signed in. Download the build for macOS, Windows or Linux, answer a short setup, and you are on the floor. Settings, then Prerequisites, shows anything that is still missing." } },
+ { "@type": "Question", "name": "What do I need before installing Munder Difflin?", "acceptedAnswer": { "@type": "Answer", "text": "One supported terminal coding CLI, installed and signed in. Twelve are supported: Claude Code, Codex, Gemini CLI, Antigravity, Grok, Kimi Code, Qwen, OpenCode, Crush, Pi, Copilot and Cursor. If you do not pay for any AI subscription, Antigravity is free." } },
+ { "@type": "Question", "name": "Who is Michael and how do I give him work?", "acceptedAnswer": { "@type": "Answer", "text": "Michael is your clone, the boss of the floor, and you stay the boss of him. You type what you want into his terminal or talk to him by voice. He breaks it into tasks, hires workers, routes messages between them, and brings you only the decisions that need you." } },
+ { "@type": "Question", "name": "Do I have to approve everything the agents do?", "acceptedAnswer": { "@type": "Answer", "text": "Only if you want to. In ask first mode agents pause for tool approval, which is a good way to start. With auto mode on they carry on alone, and anything that genuinely needs you lands on the ASK ME board. Per agent token budgets and a circuit breaker catch runaway agents either way." } },
+ { "@type": "Question", "name": "How do I review what an agent changed?", "acceptedAnswer": { "@type": "Answer", "text": "Open the IDE on the agent you are looking at. You can browse and edit files in its workspace and see its uncommitted changes as a diff before you accept anything." } },
+ { "@type": "Question", "name": "Does Munder Difflin keep working after I walk away?", "acceptedAnswer": { "@type": "Answer", "text": "Yes, as long as the machine stays on. The Triggers tab holds schedules that start work on an interval or on chosen weekdays, and agents keep working through their queue while you are away. Hosted sandboxes that keep going with the lid closed are not available yet." } },
+ { "@type": "Question", "name": "Is Munder Difflin free?", "acceptedAnswer": { "@type": "Answer", "text": "Yes. The classic office is free and MIT licensed, with unlimited local agents. Pro and Teams are optional paid plans, each with a 14 day trial." } }
]
}
@@ -154,14 +155,14 @@
<a href="/blog/topics/guides/">Guides</a>
</nav>
<h1>Your First Hour With Munder Difflin</h1>
- <p class="dek">A minute-by-minute walkthrough of your first hour with Munder Difflin v0.4.4: install, the clone-of-you onboarding, your first brief to Michael, watching the floor, approving your first escalation, and leaving a schedule running.</p>
+ <p class="dek">A minute by minute walkthrough of your first hour with Munder Difflin 0.5.2: install, setup, your first job for Michael, watching the floor, answering an ASK ME card, reading a diff, and leaving a schedule running.</p>
<div class="byline">
<span class="av" aria-hidden="true">CG</span>
<strong>Chaitanya Giri</strong>
<span class="dot" aria-hidden="true">·</span>
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true">·</span>
- <span>6 min read</span>
+ <span>5 min read</span>
<span class="tag kind">Guides</span>
</div>
</header>
@@ -170,47 +171,90 @@
<div class="post-wrap wrap">
<div class="prose">
- <div class="callout tldr"><span class="ic">TL;DR</span><p><strong>You can go from zero to a running AI office in one hour.</strong> Minute 0: install. Minute 5: onboarding — it opens on the honest pitch, <strong>a clone of you, working 24/7</strong>, and validates your setup at step one. Minute 10: your <strong>first brief to Michael</strong>. Minute 20: watch avatars work the floor and open a real desk terminal. Minute 40: your <strong>first approval</strong> and a side-by-side diff in the built-in <strong>Monaco IDE</strong>. Minute 60: set a <strong>schedule</strong> and walk away — the office keeps working.</p></div>
+ <div class="callout tldr"><span class="ic">TL;DR</span><p><strong>Zero to a working AI office in an
+hour.</strong> Minute 0: download. Minute 5: a short setup, including your clone's engine. Minute 15: your
+<strong>first job for Michael</strong>. Minute 25: watch the floor and open a real terminal. Minute 40:
+answer an <strong>ASK ME</strong> card and read a <strong>diff</strong>. Minute 50: give an agent a
+<strong>skill</strong>. Minute 60: leave a <strong>schedule</strong> running and walk away.</p></div>
<video controls preload="none" playsinline poster="/media/demo/intro-poster.jpg" style="width:100%; border-radius:12px; margin:12px 0 24px;">
<source src="/media/demo/intro.mp4" type="video/mp4" />
</video>
-<p>Most tools ask you to learn them before they do anything. Munder Difflin is the other kind: within an hour you’ve briefed an orchestrator, watched agents work, approved a real change, and left the office running without you. Here’s that hour, minute by minute — current as of v0.4.4.</p>
-<div class="yt yt-placeholder" role="img" aria-label="Video coming soon: Your first hour with Munder Difflin — full walkthrough">
- <div class="ph-art"><span class="yt-badge" aria-hidden="true"><svg viewBox="0 0 24 24" width="26" height="26" fill="currentColor" aria-hidden="true"><path d="M8 5.5v13l11-6.5z"/></svg></span><span class="ph-note">video on its way</span><span class="yt-ph-title">Your first hour with Munder Difflin — full walkthrough</span></div>
-</div>
-<h2 id="minute-0-install" tabindex="-1">Minute 0 — Install <a class="anchor" href="#minute-0-install" aria-hidden="true">#</a></h2>
-<p>Two paths: grab a signed build (macOS, Windows, Linux) from the <a href="https://github.com/chaitanyagiri/munder-difflin/releases/latest">releases page</a>, or clone and run from source with <code>npm install &amp;&amp; npm run dev</code>. You’ll want at least one supported agent CLI on your <code>PATH</code> — the engine card now honestly names all ten: Claude Code, Antigravity, Codex, Grok, Kimi, Qwen, OpenCode, Crush, pi, and GitHub Copilot CLI.</p>
-<p>Missing something? You no longer have to find out the hard way. <strong>Settings → Prerequisites</strong> shows live status for git, Node, uv, MemPalace and every engine — real paths, platform-correct install commands, and a button that simply asks Michael to fill the gaps for you. The <a href="/blog/how-to-install-and-use-munder-difflin/">install and usage guide</a> covers the deeper details. Budget five minutes.</p>
-<figure class="fig fig-inline"><img src="/blog/assets/media/your-first-hour-with-munder-difflin/note-1.png" alt="Hand-drawn sketch from “Your First Hour With Munder Difflin”" loading="lazy" decoding="async" /><figcaption>Prerequisites, before they become surprises: live status for every tool, and a button that hands the gaps to Michael.</figcaption></figure>
-<h2 id="minute-5-onboarding-meet-your-clone" tabindex="-1">Minute 5 — Onboarding: meet your clone <a class="anchor" href="#minute-5-onboarding-meet-your-clone" aria-hidden="true">#</a></h2>
-<p>First launch opens on the pitch the whole product keeps: <strong>a clone of you, working 24/7.</strong> The wizard’s few questions are the actual shape of the product:</p>
+<p>Most tools want you to study them before they do anything useful. This one should be doing real work inside an
+hour: you brief an orchestrator, watch agents work, check a real change, and leave the office running without you.
+Here is that hour, minute by minute, current as of Munder Difflin 0.5.2.</p>
+<h2 id="minute-0-download-and-install" tabindex="-1">Minute 0: Download and install <a class="anchor" href="#minute-0-download-and-install" aria-hidden="true">#</a></h2>
+<p>Grab the build for your system from <a href="https://munderdiffl.in/">munderdiffl.in</a> or the
+<a href="https://github.com/chaitanyagiri/munder-difflin/releases/latest">latest release</a>:</p>
<ul>
-<li><strong>Your clone’s engine</strong> — Michael runs on a pluggable engine; pick from the ten, change it later.</li>
-<li><strong>Harness home</strong> — the folder where the hive lives: per-agent memory, mailboxes, the blackboard, the event log. Plain files in a local git repo. The wizard validates this at step one — an empty or impossible folder fails immediately, not after step four.</li>
-<li><strong>Registered repos</strong> — the codebases your agents will work on.</li>
-<li><strong>Auto-mode</strong> — whether agents proceed without per-tool prompts (the approval gate still catches critical items).</li>
+<li><strong>macOS:</strong> one universal <code>.dmg</code> for Apple Silicon and Intel, signed and notarized by Apple.</li>
+<li><strong>Windows 10 and 11:</strong> a setup installer, or a portable <code>.exe</code> if you would rather not install anything.</li>
+<li><strong>Linux:</strong> an <code>.AppImage</code>.</li>
</ul>
-<p>Finish the wizard and you land on the floor: a pixel-art office, empty except for Michael, who seats himself in his office automatically — with a <strong>BOSS</strong> tag on his card, because he’s the boss of the agents while you stay the boss of him. Everything the hive needs starts with him: as of v0.4.4 a brand-new install boots its message router, hook server and mission scheduler on the very first run.</p>
-<h2 id="minute-10-your-first-brief-to-michael" tabindex="-1">Minute 10 — Your first brief to Michael <a class="anchor" href="#minute-10-your-first-brief-to-michael" aria-hidden="true">#</a></h2>
-<p>You don’t manage the workers. You talk to your clone, and he runs the floor. Click into Michael’s terminal and type a brief the way you’d brief a colleague: what you want, which repo, what “done” looks like. Prefer talking? His voice mode opens with a live snapshot of the floor and can run nearly the whole app.</p>
-<p>Michael adjudicates. He creates tasks on the kanban, assigns them, and routes messages between agent inboxes. Workers you hire via <strong>Add agent</strong> each get a real CLI process in its own pseudo-terminal and, with the git-isolation toggle, their own worktree — so nobody collides on branches. How he decides what to route, resolve, or escalate is its own post: <a href="/blog/how-the-god-orchestrator-works/">how the orchestrator works</a>.</p>
-<p>A good first brief is small and self-contained: “read this repo and write a <a href="http://REPORT.md">REPORT.md</a> summarizing the architecture” beats “refactor everything” for hour one.</p>
-<h2 id="minute-20-watch-the-floor-then-open-a-desk" tabindex="-1">Minute 20 — Watch the floor, then open a desk <a class="anchor" href="#minute-20-watch-the-floor-then-open-a-desk" aria-hidden="true">#</a></h2>
-<p>Now the part that makes the product legible: the floor is not a decoration, it’s the state of the system. Avatars walk to stations as they work. When the hive routes a message, an envelope flies from sender to recipient; escalations fly to the door. The cast is an affectionate parody of The Office, and every movement maps to a real event.</p>
-<figure class="fig fig-inline"><img src="/blog/assets/media/your-first-hour-with-munder-difflin/floor-view.png" alt="The office floor: suited agents at desks, one carrying a sticky note to the kanban wall" loading="lazy" decoding="async" /><figcaption>The floor mid-task: every movement maps to a real event.</figcaption></figure>
-<p>Click any agent and you get their desk: the live terminal (you can type back into it), a sandboxed file browser, and a git tab with status, log, and commit graph. Go fullscreen and the roster cards show each agent’s model, project, and a live context gauge — the fuel dial for every worker at a glance. This is the moment the abstraction clicks: that avatar is a real CLI process, and you’re reading its actual stdout.</p>
-<figure class="fig fig-inline"><img src="/blog/assets/media/your-first-hour-with-munder-difflin/note-2.png" alt="Hand-drawn sketch from “Your First Hour With Munder Difflin”" loading="lazy" decoding="async" /><figcaption>Every desk is real: a live terminal, a file browser, a git tab — and a context gauge telling you how much runway the agent has left.</figcaption></figure>
-<h2 id="minute-40-your-first-approval-and-the-diff" tabindex="-1">Minute 40 — Your first approval, and the diff <a class="anchor" href="#minute-40-your-first-approval-and-the-diff" aria-hidden="true">#</a></h2>
-<p>Sometime in the first hour, something lands in the approvals queue — a spend threshold, a destructive operation, a scope change. This is by design: Michael resolves routine requests himself and escalates only critical items, so the queue stays short and every item in it deserves your attention. (The philosophy behind that gate: <a href="/blog/human-in-the-loop-approving-ai-agents/">approving AI agents without babysitting them</a>.)</p>
-<p>Before you click approve, look at the work. Hit the title-bar <strong>IDE</strong> button and the built-in Monaco editor — the VS Code editor engine, fully self-hosted — opens over the floor. The git CHANGES rail lists what changed; click a file for a read-only side-by-side diff against HEAD. Images preview right in the IDE now, so a screenshot or SVG an agent produced is one click to inspect. Read the diff, approve the item, watch the envelope fly.</p>
-<h2 id="minute-50-give-an-agent-a-skill" tabindex="-1">Minute 50 — Give an agent a skill <a class="anchor" href="#minute-50-give-an-agent-a-skill" aria-hidden="true">#</a></h2>
-<p>New in v0.4.4 and worth five minutes of your first hour: <strong>Skills.</strong> A browsable catalog of hundreds of skills — search, category and publisher filters — installable across Claude Code, OpenCode and Codex with scope precedence handled for you. Give your reviewer a review checklist, your writer a style guide, your researcher a source-citing discipline. It’s the difference between hiring generalists and hiring people who’ve read the manual. (Background: <a href="/blog/mcp-and-skills-in-a-hive/">MCP and skills in a hive</a>.)</p>
-<h2 id="minute-60-leave-it-running" tabindex="-1">Minute 60 — Leave it running <a class="anchor" href="#minute-60-leave-it-running" aria-hidden="true">#</a></h2>
-<p>The last move of the hour is the one that changes your relationship with the tool: schedule something. The Command Center’s <strong>Triggers</strong> tab takes a label, an interval, a target agent, and a mission body — “every morning, triage new GitHub issues,” say — and a heartbeat re-engages the floor if it goes quiet. Last-fired and next-fired times are right there in the tab.</p>
-<p>Close the laptop. It’s local-first, so tomorrow the office is exactly where you left it — probably with mail in your queue. And when the next release ships, the app tells you itself: the update toast says what’s actually inside before you restart into it. For patterns on what’s worth automating, see <a href="/blog/scheduling-autonomous-agent-missions/">scheduling autonomous agent missions</a>.</p>
+<p>Each release also carries a <code>SHA256SUMS.txt</code>, so you can check the file you downloaded is the real one.</p>
+<p>You need one terminal coding CLI installed and signed in. Twelve are supported: Claude Code, Codex, Gemini CLI,
+Antigravity, Grok, Kimi Code, Qwen, OpenCode, Crush, Pi, Copilot and Cursor. If you do not pay for any AI
+subscription, Antigravity is free. The <a href="/blog/how-to-install-and-use-munder-difflin/">install guide</a> has the exact
+commands for every one of them.</p>
+<h2 id="minute-5-setup" tabindex="-1">Minute 5: Setup <a class="anchor" href="#minute-5-setup" aria-hidden="true">#</a></h2>
+<p>First launch walks you through a short setup. It starts by asking whether you are technical, which only changes how
+much jargon the app shows you. Then four steps:</p>
+<ul>
+<li><strong>A home folder</strong> for the app’s own files: settings, agent memory and mailboxes. Use a new, empty folder.</li>
+<li><strong>Your clone.</strong> Name it and pick the engine that powers it. Give it the most capable model you have, because it
+does the thinking and the delegating.</li>
+<li><strong>Your projects.</strong> The folders agents may work in. A project is just a folder.</li>
+<li><strong>Permissions.</strong> Whether agents act on their own or ask first. Start with ask first. You can change it any time.</li>
+</ul>
+<p>Every install starts in the classic office, the free version, with the floor you are about to meet.</p>
+<figure class="fig fig-inline"><img src="/blog/assets/media/your-first-hour-with-munder-difflin/note-1.png" alt="Hand-drawn sketch from “Your First Hour With Munder Difflin”" loading="lazy" decoding="async" /><figcaption>Prerequisites in Settings shows the live status of every tool an engine needs, before it becomes a surprise.</figcaption></figure>
+<h2 id="minute-15-your-first-job-for-michael" tabindex="-1">Minute 15: Your first job for Michael <a class="anchor" href="#minute-15-your-first-job-for-michael" aria-hidden="true">#</a></h2>
+<p>You do not manage the workers. You talk to your clone, and he runs the floor. Click into Michael’s terminal and
+describe the job the way you would brief a capable new hire: the outcome, the folder, and what done looks like.
+Prefer talking? He has a voice mode.</p>
+<p>A good first job is small and self contained. “Read this repo and write <a href="http://REPORT.md">REPORT.md</a> explaining how it fits together”
+beats “refactor everything” for hour one.</p>
+<p>Michael turns the job into tasks on the kanban, hires a worker if he needs one, and routes messages between their
+mailboxes. Each worker is a real CLI process in its own terminal and, with git isolation on, its own worktree. How he
+decides what to handle and what to send your way is its own post: <a href="/blog/how-the-god-orchestrator-works/">how Michael routes work</a>.</p>
+<h2 id="minute-25-watch-the-floor-then-open-a-desk" tabindex="-1">Minute 25: Watch the floor, then open a desk <a class="anchor" href="#minute-25-watch-the-floor-then-open-a-desk" aria-hidden="true">#</a></h2>
+<p>The floor is not decoration. It is the state of the system. Characters walk to their desks as they work, and envelopes
+fly between them when they message each other. The cast is an affectionate parody of The Office, and the whole floor is
+a simulation that uses no tokens.</p>
+<figure class="fig fig-inline"><img src="/blog/assets/media/your-first-hour-with-munder-difflin/floor-view.png" alt="The office floor: suited agents at desks, one carrying a sticky note to the kanban wall" loading="lazy" decoding="async" /><figcaption>The floor mid task: every movement maps to a real event.</figcaption></figure>
+<p>Click any agent to open its terminal. You can read the live output and type straight back into it. Want to work with
+one agent without the office around it? Focus mode gives you the terminal full width, and Esc brings the floor back.
+This is the moment it clicks: that character is a real CLI process, and you are reading its actual output.</p>
+<figure class="fig fig-inline"><img src="/blog/assets/media/your-first-hour-with-munder-difflin/note-2.png" alt="Hand-drawn sketch from “Your First Hour With Munder Difflin”" loading="lazy" decoding="async" /><figcaption>Every desk is real: a live terminal, plus a context gauge showing how much runway the agent has left.</figcaption></figure>
+<h2 id="minute-40-answer-an-ask-me-card-then-read-the-diff" tabindex="-1">Minute 40: Answer an ASK ME card, then read the diff <a class="anchor" href="#minute-40-answer-an-ask-me-card-then-read-the-diff" aria-hidden="true">#</a></h2>
+<p>At some point in the first hour an agent will need you. It might be a question only you can answer, or a step only you
+can take. It shows up on the <strong>ASK ME</strong> board instead of hiding in a scrollback, formatted so you can read it at a
+glance. Michael settles routine questions himself, so whatever reaches you deserves your attention.</p>
+<p>Before you say yes to a change, look at it. Open the <strong>IDE</strong> on the agent’s workspace to browse its files and see its
+uncommitted changes as a diff. Read it, answer the card, and watch the work carry on. The thinking behind that gate is in
+<a href="/blog/human-in-the-loop-approving-ai-agents/">approving AI agents without babysitting them</a>.</p>
+<h2 id="minute-50-give-an-agent-a-skill" tabindex="-1">Minute 50: Give an agent a skill <a class="anchor" href="#minute-50-give-an-agent-a-skill" aria-hidden="true">#</a></h2>
+<p>The <strong>skills</strong> tab is a catalog of skills you can install for your agents. Give your reviewer a review checklist, your
+writer a style guide, and your researcher the habit of citing sources. It is the difference between hiring generalists
+and hiring people who actually read the manual. More in <a href="/blog/mcp-and-skills-in-a-hive/">MCP and skills in a hive</a>.</p>
+<h2 id="minute-60-leave-it-running" tabindex="-1">Minute 60: Leave it running <a class="anchor" href="#minute-60-leave-it-running" aria-hidden="true">#</a></h2>
+<p>The last move of the hour is the one that changes how you use the tool: schedule something. Open the <strong>Triggers</strong> tab and
+create a schedule with a label, who it goes to, and the prompt, which is sent word for word on every run. It can repeat on
+an interval or on chosen weekdays at a set time. “Every weekday at 9, triage new GitHub issues and summarise them for me” is
+a good first one.</p>
+<p>Leave the machine on and walk away. The agents keep working, and when a new version ships the app updates itself. Hosted
+sandboxes that keep working with the lid closed are not available yet, so for now the office runs wherever you run it. For
+ideas on what is worth automating, read <a href="/blog/scheduling-autonomous-agent-missions/">scheduling autonomous agent missions</a>.</p>
+<h2 id="want-the-office-in-one-window" tabindex="-1">Want the office in one window? <a class="anchor" href="#want-the-office-in-one-window" aria-hidden="true">#</a></h2>
+<p>Everything above is the free classic office. If you would rather work in a single window, <strong>Pro</strong> puts the orchestrator,
+agents, tasks, inbox, automations and memory in one workspace and adds <strong>Stapler</strong>, a small floating window that sends a
+screenshot, a recorded message or a meeting transcript to your orchestrator with one line from you. <strong>Teams</strong> lets your
+clone work with your teammates’ clones. Both come with a 14 day trial, and the <a href="https://munderdiffl.in/#pricing">pricing page</a>
+has the details.</p>
<h2 id="the-hour-in-one-line" tabindex="-1">The hour, in one line <a class="anchor" href="#the-hour-in-one-line" aria-hidden="true">#</a></h2>
-<p>Install, meet your clone, brief him once, watch the floor, read one diff, install one skill, set one schedule — and you’ve gone from “a CLI in a terminal” to “an office that works while you don’t.”</p>
-<p><a href="https://github.com/chaitanyagiri/munder-difflin/releases/latest">Download Munder Difflin</a> — free, MIT-licensed, local-first — and if the first hour earns it, <a href="https://github.com/chaitanyagiri/munder-difflin">a GitHub star</a> helps more people find it.</p>
+<p>Install, set up your clone, give him one job, watch the floor, answer one card, read one diff, install one skill, set one
+schedule. That is the whole distance from “a CLI in a terminal” to “an office that works while you do not”.</p>
+<p><a href="https://munderdiffl.in/">Download Munder Difflin</a>, free and MIT licensed. If the first hour earns it,
+<a href="https://github.com/chaitanyagiri/munder-difflin">a GitHub star</a> helps other people find it.</p>
@@ -219,32 +263,37 @@
<div class="qa">
<p class="q">How long does it take to get Munder Difflin running?</p>
- <p class="a">About five minutes. Grab the signed build for macOS, Windows, or Linux from the releases page, or run from source with Node.js 18+. Onboarding validates your setup at the very first step — as of v0.4.4 it checks your home folder immediately instead of failing four steps later, and the Prerequisites page in Settings shows live status for every tool an engine needs, with a button that asks Michael to install what&#39;s missing.</p>
+ <p class="a">About ten minutes if one coding CLI, like Claude Code, Codex or Antigravity, is already installed and signed in. Download the build for macOS, Windows or Linux, answer a short setup, and you are on the floor. Settings, then Prerequisites, shows anything that is still missing.</p>
</div>
<div class="qa">
<p class="q">What do I need before installing Munder Difflin?</p>
- <p class="a">At least one supported agent CLI — Claude Code, Antigravity, Codex, Grok, Kimi, Qwen, OpenCode, Crush, pi, or GitHub Copilot CLI. If you&#39;re missing one (or git, Node, uv, MemPalace), Settings → Prerequisites shows exactly what&#39;s absent with the platform-correct install command, and can hand the whole job to Michael.</p>
+ <p class="a">One supported terminal coding CLI, installed and signed in. Twelve are supported: Claude Code, Codex, Gemini CLI, Antigravity, Grok, Kimi Code, Qwen, OpenCode, Crush, Pi, Copilot and Cursor. If you do not pay for any AI subscription, Antigravity is free.</p>
</div>
<div class="qa">
<p class="q">Who is Michael and how do I give him work?</p>
- <p class="a">Michael is a clone of you — the boss of the agents, while you stay the boss of him. You type a brief into his terminal (or talk to him by voice) and he adjudicates it: creating tasks, assigning them to workers, routing messages between inboxes, and escalating only critical items back to you. He seats himself in his office automatically on first launch.</p>
+ <p class="a">Michael is your clone, the boss of the floor, and you stay the boss of him. You type what you want into his terminal or talk to him by voice. He breaks it into tasks, hires workers, routes messages between them, and brings you only the decisions that need you.</p>
</div>
<div class="qa">
<p class="q">Do I have to approve everything the agents do?</p>
- <p class="a">No. Michael resolves routine requests himself so the system stays autonomous. Only critical items — spend, destructive operations, scope changes — land in the human-in-the-loop approvals queue for you to act on. A circuit breaker and per-agent token budgets guard the rest.</p>
+ <p class="a">Only if you want to. In ask first mode agents pause for tool approval, which is a good way to start. With auto mode on they carry on alone, and anything that genuinely needs you lands on the ASK ME board. Per agent token budgets and a circuit breaker catch runaway agents either way.</p>
</div>
<div class="qa">
- <p class="q">How do I review what an agent actually changed?</p>
- <p class="a">Open the built-in Monaco IDE from the title-bar IDE button. Its git CHANGES rail lists every modified file, and clicking one opens a read-only side-by-side diff against HEAD. There&#39;s a per-agent git tab with status, log, and commit graph — and since v0.4.4 the IDE previews images too, so a design change is as reviewable as a code change.</p>
+ <p class="q">How do I review what an agent changed?</p>
+ <p class="a">Open the IDE on the agent you are looking at. You can browse and edit files in its workspace and see its uncommitted changes as a diff before you accept anything.</p>
</div>
<div class="qa">
- <p class="q">Can Munder Difflin keep working after I walk away?</p>
- <p class="a">Yes — that&#39;s the point. The Triggers tab in the Command Center holds recurring missions with a label, interval, target agent, and body, plus a heartbeat that re-engages the floor when it goes quiet. It&#39;s local-first, so the office runs on hardware you already own — and the app keeps itself current, telling you exactly what&#39;s in each update before you restart into it.</p>
+ <p class="q">Does Munder Difflin keep working after I walk away?</p>
+ <p class="a">Yes, as long as the machine stays on. The Triggers tab holds schedules that start work on an interval or on chosen weekdays, and agents keep working through their queue while you are away. Hosted sandboxes that keep going with the lid closed are not available yet.</p>
+ </div>
+
+ <div class="qa">
+ <p class="q">Is Munder Difflin free?</p>
+ <p class="a">Yes. The classic office is free and MIT licensed, with unlimited local agents. Pro and Teams are optional paid plans, each with a 14 day trial.</p>
</div>
</section>
@@ -263,7 +312,7 @@
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
- <li class="lvl-2"><a href="#minute-0-install">Minute 0 — Install</a></li><li class="lvl-2"><a href="#minute-5-onboarding-meet-your-clone">Minute 5 — Onboarding: meet your clone</a></li><li class="lvl-2"><a href="#minute-10-your-first-brief-to-michael">Minute 10 — Your first brief to Michael</a></li><li class="lvl-2"><a href="#minute-20-watch-the-floor-then-open-a-desk">Minute 20 — Watch the floor, then open a desk</a></li><li class="lvl-2"><a href="#minute-40-your-first-approval-and-the-diff">Minute 40 — Your first approval, and the diff</a></li><li class="lvl-2"><a href="#minute-50-give-an-agent-a-skill">Minute 50 — Give an agent a skill</a></li><li class="lvl-2"><a href="#minute-60-leave-it-running">Minute 60 — Leave it running</a></li><li class="lvl-2"><a href="#the-hour-in-one-line">The hour, in one line</a></li>
+ <li class="lvl-2"><a href="#minute-0-download-and-install">Minute 0: Download and install</a></li><li class="lvl-2"><a href="#minute-5-setup">Minute 5: Setup</a></li><li class="lvl-2"><a href="#minute-15-your-first-job-for-michael">Minute 15: Your first job for Michael</a></li><li class="lvl-2"><a href="#minute-25-watch-the-floor-then-open-a-desk">Minute 25: Watch the floor, then open a desk</a></li><li class="lvl-2"><a href="#minute-40-answer-an-ask-me-card-then-read-the-diff">Minute 40: Answer an ASK ME card, then read the diff</a></li><li class="lvl-2"><a href="#minute-50-give-an-agent-a-skill">Minute 50: Give an agent a skill</a></li><li class="lvl-2"><a href="#minute-60-leave-it-running">Minute 60: Leave it running</a></li><li class="lvl-2"><a href="#want-the-office-in-one-window">Want the office in one window?</a></li><li class="lvl-2"><a href="#the-hour-in-one-line">The hour, in one line</a></li>
</ol>
</aside>
@@ -282,13 +331,13 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
- <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet, and Budgets in One Place</a></h3>
- <p class="dek">A guide to Munder Difflin&#39;s Command Center — the task kanban, live fleet monitoring, per-agent budgets and cost telemetry — and when to watch the board instead of the floor.</p>
+ <h3><a href="/blog/command-center-guide/">The Command Center: Kanban, Fleet and Budgets in One Place</a></h3>
+ <p class="dek">A guide to Munder Difflin&#39;s Command Center in 0.5.2: the kanban with task dependencies, the ask me tab, triggers, memory, live token use against each agent&#39;s budget, and when to watch the board instead of the floor.</p>
<div class="foot">
<span class="tag kind">Guides</span>
- <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet, and Budgets in One Place">Read →</a>
+ <a class="more" href="/blog/command-center-guide/" aria-label="Read: The Command Center: Kanban, Fleet and Budgets in One Place">Read →</a>
</div>
</div>
</article>
@@ -314,10 +363,10 @@
<div class="meta">
<time datetime="2026-07-03">Jul 3, 2026</time>
<span class="dot" aria-hidden="true"></span>
- <span>5 min</span>
+ <span>4 min</span>
</div>
<h3><a href="/blog/how-to-hire-from-the-agent-gallery/">How to Hire From the Agent Gallery</a></h3>
- <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: browse six off-the-shelf hires, import one from a link or file, review the pre-filled manifest, customize identity, workspace, engine, and briefing — then spawn it yourself.</p>
+ <p class="dek">A practical guide to Munder Difflin&#39;s Agent Gallery: pick one of 80 ready made roles, download its manifest, import it with Add agent, review every field, customise identity, workspace, engine and briefing, then spawn it yourself.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/how-to-hire-from-the-agent-gallery/" aria-label="Read: How to Hire From the Agent Gallery">Read →</a>
diff --git a/docs/sitemap.xml b/docs/sitemap.xml
index 58d17762..7e163487 100644
--- a/docs/sitemap.xml
+++ b/docs/sitemap.xml
@@ -10,7 +10,7 @@
<urlset xmlns="http://www.sitemaps.org/schemas/sitemap/0.9">
<url>
<loc>https://munderdiffl.in/</loc>
- <lastmod>2026-08-27</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>weekly</changefreq>
<priority>1.0</priority>
</url>
@@ -112,13 +112,13 @@
</url>
<url>
<loc>https://munderdiffl.in/blog/qm-vs-munder-difflin/</loc>
- <lastmod>2026-08-06</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
<url>
<loc>https://munderdiffl.in/blog/command-center-guide/</loc>
- <lastmod>2026-07-03</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
@@ -142,7 +142,7 @@
</url>
<url>
<loc>https://munderdiffl.in/blog/how-to-hire-from-the-agent-gallery/</loc>
- <lastmod>2026-07-03</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
@@ -214,7 +214,7 @@
</url>
<url>
<loc>https://munderdiffl.in/blog/your-first-hour-with-munder-difflin/</loc>
- <lastmod>2026-08-20</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
@@ -226,13 +226,13 @@
</url>
<url>
<loc>https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/</loc>
- <lastmod>2026-06-22</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
<url>
<loc>https://munderdiffl.in/blog/run-munder-difflin-on-open-models/</loc>
- <lastmod>2026-08-20</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
@@ -262,7 +262,7 @@
</url>
<url>
<loc>https://munderdiffl.in/blog/deploy-a-blog-writer-agent/</loc>
- <lastmod>2026-08-20</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
@@ -352,7 +352,7 @@
</url>
<url>
<loc>https://munderdiffl.in/blog/how-to-install-and-use-munder-difflin/</loc>
- <lastmod>2026-08-27</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
@@ -394,7 +394,7 @@
</url>
<url>
<loc>https://munderdiffl.in/blog/best-claude-code-multi-agent-tools/</loc>
- <lastmod>2026-06-04</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>
@@ -802,7 +802,7 @@
</url>
<url>
<loc>https://munderdiffl.in/blog/what-is-a-multi-agent-harness/</loc>
- <lastmod>2026-05-22</lastmod>
+ <lastmod>2026-09-10</lastmod>
<changefreq>monthly</changefreq>
<priority>0.8</priority>
</url>