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
HTML agent-harnessagent-orchestrationagentsai-agentsautonomous-agentsclaude-codecodexcoding-agentsdesktop-appdeveloper-toolselectrongemini-cliharnessharness-engineeringlocal-firstmemorymulti-agentopencodeorchestrationtypescript
blog: stop wide tables from dragging the whole page sideways on phones
chaitanyagirie5e1ba03f398
diff --git a/blog/eleventy.config.js b/blog/eleventy.config.jsindex 1f0eed2c..90ca7297 100644--- a/blog/eleventy.config.js+++ b/blog/eleventy.config.js@@ -31,6 +31,12 @@ export default function (eleventyConfig) {slugify,});+ // A markdown table of npm commands is wider than a phone. Left bare it widens+ // the whole document, which drags the sticky masthead off-screen with it.+ // Wrapping every table in its own scroll box keeps the overflow local.+ md.renderer.rules.table_open = () => '<div class="table-scroll">\n<table>\n';+ md.renderer.rules.table_close = () => '</table>\n</div>\n';+eleventyConfig.setLibrary("md", md);// ---- plugins ----diff --git a/blog/src/assets/blog-press.css b/blog/src/assets/blog-press.cssindex 0514b3f4..3d73cb97 100644--- a/blog/src/assets/blog-press.css+++ b/blog/src/assets/blog-press.css@@ -213,6 +213,8 @@ a.tag:hover { background: var(--ink); color: var(--paper); }font-family: var(--disp); font-weight: 600; font-size: 1.25em; line-height: 1.45; font-style: italic; color: var(--ink);}.prose hr { border: none; border-top: 4px solid var(--rule); width: 88px; margin: 3em 0; }+.prose .table-scroll { margin: 1.8em 0; max-width: 100%; overflow-x: auto; -webkit-overflow-scrolling: touch; }+.prose .table-scroll > table { margin: 0; min-width: 100%; }.prose table { width: 100%; border-collapse: collapse; margin: 1.8em 0; font-size: 16.5px; border: 2px solid var(--rule); }.prose th { font-family: var(--grot); font-size: 13px; letter-spacing: 0.06em; text-transform: uppercase; text-align: left; background: var(--ink); color: var(--paper); padding: 10px 14px; }.prose td { padding: 10px 14px; border-top: 1px solid var(--line); }@@ -316,6 +318,11 @@ footer.site-foot .wrap { padding: 40px 24px; }.prevnext { grid-template-columns: 1fr; }.prose p, .prose li { font-size: 18px; }.prose > p:first-of-type::first-letter { font-size: 3.6em; }+ /* Let the table size to its content and scroll, rather than cramming a+ 40-character npm command into a 90px column. */+ .prose .table-scroll > table { width: max-content; font-size: 15px; }+ .prose .table-scroll th, .prose .table-scroll td { padding: 8px 12px; white-space: nowrap; }+ .prose .table-scroll td code { white-space: nowrap; }}/* youtube placeholder (no id yet) */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.mdindex 80bfd614..511e69ee 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@@ -243,14 +243,14 @@ Same idea: install, then run the command once on its own to sign in.| Engine | Install it with | Then run | Official docs ||---|---|---|---|-| Gemini CLI | `npm install -g @google/gemini-cli` | `gemini` | [github.com/google-gemini/gemini-cli](https://github.com/google-gemini/gemini-cli) |-| Grok | `npm install -g @xai-official/grok` | `grok` | [docs.x.ai/build/overview](https://docs.x.ai/build/overview) |-| Kimi Code | `npm install -g @moonshot-ai/kimi-code` | `kimi` | [kimi.com/code/docs](https://www.kimi.com/code/docs/en/kimi-code-cli/guides/getting-started.html) |-| Qwen | `npm install -g @qwen-code/qwen-code@latest` | `qwen` | [github.com/QwenLM/qwen-code](https://github.com/QwenLM/qwen-code) |-| Crush | `npm install -g @charmland/crush` | `crush` | [github.com/charmbracelet/crush](https://github.com/charmbracelet/crush) |+| Gemini CLI | `npm install -g @google/gemini-cli` | `gemini` | [google-gemini/gemini-cli](https://github.com/google-gemini/gemini-cli) |+| Grok | `npm install -g @xai-official/grok` | `grok` | [docs.x.ai](https://docs.x.ai/build/overview) |+| Kimi Code | `npm install -g @moonshot-ai/kimi-code` | `kimi` | [kimi.com/code](https://www.kimi.com/code/docs/en/kimi-code-cli/guides/getting-started.html) |+| Qwen | `npm install -g @qwen-code/qwen-code@latest` | `qwen` | [QwenLM/qwen-code](https://github.com/QwenLM/qwen-code) |+| Crush | `npm install -g @charmland/crush` | `crush` | [charmbracelet/crush](https://github.com/charmbracelet/crush) || Copilot | `npm install -g @github/copilot` | `copilot` | [docs.github.com](https://docs.github.com/copilot/concepts/agents/about-copilot-cli) |-| Cursor | `curl https://cursor.com/install -fsS \| bash` | `cursor-agent` | [cursor.com/docs/cli](https://cursor.com/docs/cli/overview) |-| pi | `npm install -g --ignore-scripts @earendil-works/pi-coding-agent` | `pi` | [pi.dev/docs/latest](https://pi.dev/docs/latest) |+| Cursor | `curl https://cursor.com/install -fsS \| bash` | `cursor-agent` | [cursor.com/docs](https://cursor.com/docs/cli/overview) |+| pi | `npm install -g --ignore-scripts @earendil-works/pi-coding-agent` | `pi` | [pi.dev/docs](https://pi.dev/docs/latest) |On Windows, the Cursor line is `irm 'https://cursor.com/install?win32=true' | iex` in PowerShell.diff --git a/docs/blog/assets/blog-press.css b/docs/blog/assets/blog-press.cssindex 0514b3f4..3d73cb97 100644--- a/docs/blog/assets/blog-press.css+++ b/docs/blog/assets/blog-press.css@@ -213,6 +213,8 @@ a.tag:hover { background: var(--ink); color: var(--paper); }font-family: var(--disp); font-weight: 600; font-size: 1.25em; line-height: 1.45; font-style: italic; color: var(--ink);}.prose hr { border: none; border-top: 4px solid var(--rule); width: 88px; margin: 3em 0; }+.prose .table-scroll { margin: 1.8em 0; max-width: 100%; overflow-x: auto; -webkit-overflow-scrolling: touch; }+.prose .table-scroll > table { margin: 0; min-width: 100%; }.prose table { width: 100%; border-collapse: collapse; margin: 1.8em 0; font-size: 16.5px; border: 2px solid var(--rule); }.prose th { font-family: var(--grot); font-size: 13px; letter-spacing: 0.06em; text-transform: uppercase; text-align: left; background: var(--ink); color: var(--paper); padding: 10px 14px; }.prose td { padding: 10px 14px; border-top: 1px solid var(--line); }@@ -316,6 +318,11 @@ footer.site-foot .wrap { padding: 40px 24px; }.prevnext { grid-template-columns: 1fr; }.prose p, .prose li { font-size: 18px; }.prose > p:first-of-type::first-letter { font-size: 3.6em; }+ /* Let the table size to its content and scroll, rather than cramming a+ 40-character npm command into a 90px column. */+ .prose .table-scroll > table { width: max-content; font-size: 15px; }+ .prose .table-scroll th, .prose .table-scroll td { padding: 8px 12px; white-space: nowrap; }+ .prose .table-scroll td code { white-space: nowrap; }}/* youtube placeholder (no id yet) */diff --git a/docs/blog/best-ai-coding-agents/index.html b/docs/blog/best-ai-coding-agents/index.htmlindex 36814f6d..a3fa46d6 100644--- a/docs/blog/best-ai-coding-agents/index.html+++ b/docs/blog/best-ai-coding-agents/index.html@@ -193,6 +193,7 @@ shipping bigger work. The “one tool wins everything” framing is over.</p><blockquote><p>Pricing/features change monthly — treat as 2026 reference points; verify on each tool’s site.</p></blockquote>+<div class="table-scroll"><table><thead><tr>@@ -248,6 +249,7 @@ shipping bigger work. The “one tool wins everything” framing is over.</p></tr></tbody></table>+</div><figure class="fig fig-inline"><img src="/blog/assets/media/best-ai-coding-agents/note-1.png" alt="Hand-drawn sketch from “The Best AI Coding Agents in 2026: A Field Guide”" loading="lazy" decoding="async" /></figure><h2 id="the-tools-fairly" tabindex="-1">The tools, fairly <a class="anchor" href="#the-tools-fairly" aria-hidden="true">#</a></h2><p><strong>Cursor</strong> — a dedicated AI-first IDE that leads on in-editor agent ergonomics; reference pricing runsdiff --git a/docs/blog/best-claude-code-multi-agent-tools/index.html b/docs/blog/best-claude-code-multi-agent-tools/index.htmlindex 8407f70e..2ba1ec31 100644--- a/docs/blog/best-claude-code-multi-agent-tools/index.html+++ b/docs/blog/best-claude-code-multi-agent-tools/index.html@@ -198,6 +198,7 @@ harness</a> idea.</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>+<div class="table-scroll"><table><thead><tr>@@ -252,6 +253,7 @@ harness</a> idea.</li></tr></tbody></table>+</div><p>The single row that most separates the field is <strong>shared memory + orchestrator</strong>: that’s the jumpfrom “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>diff --git a/docs/blog/claude-code-orchestration-tools-compared/index.html b/docs/blog/claude-code-orchestration-tools-compared/index.htmlindex 41958368..6393b5fb 100644--- a/docs/blog/claude-code-orchestration-tools-compared/index.html+++ b/docs/blog/claude-code-orchestration-tools-compared/index.html@@ -179,6 +179,7 @@ We’ll fold it into the verdict.</p><h2 id="the-comparison" tabindex="-1">The comparison <a class="anchor" href="#the-comparison" aria-hidden="true">#</a></h2><p>Scored as ●●● strong · ●●○ partial · ●○○ minimal, to the best of our knowledge at the time of writing.Tools evolve — verify against each project’s current docs.</p>+<div class="table-scroll"><table><thead><tr>@@ -241,6 +242,7 @@ Tools evolve — verify against each project’s current docs.</p></tr></tbody></table>+</div><p>Read the table as a fit, not a scoreboard. The tools clustered on the left optimize for <strong>control andsimplicity</strong>; Munder Difflin optimizes for <strong>coordination</strong> (memory + messaging + a routingorchestrator). Both are legitimate.</p>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.htmlindex c1bf19a9..03b1b855 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@@ -221,6 +221,7 @@ and escalates only the critical calls to you. The parent-as-bottleneck problem g</ul><p>Put simply: subagents give one agent <em>helpers</em>; a harness gives you a <em>team</em>.</p><h2 id="a-side-by-side" tabindex="-1">A side-by-side <a class="anchor" href="#a-side-by-side" aria-hidden="true">#</a></h2>+<div class="table-scroll"><table><thead><tr>@@ -262,6 +263,7 @@ and escalates only the critical calls to you. The parent-as-bottleneck problem g</tr></tbody></table>+</div><p>The table isn’t a scoreboard — it’s a map. Each column wins its own territory.</p><figure class="fig fig-inline"><img src="/blog/assets/media/claude-code-subagents-vs-multi-agent-harness/note-2.png" alt="Hand-drawn sketch from “Claude Code Subagents vs a Multi-Agent Harness”" loading="lazy" decoding="async" /></figure><h2 id="theyre-better-together" tabindex="-1">They’re better together <a class="anchor" href="#theyre-better-together" aria-hidden="true">#</a></h2>diff --git a/docs/blog/claude-squad-vs-munder-difflin/index.html b/docs/blog/claude-squad-vs-munder-difflin/index.htmlindex 262205e2..0d403a08 100644--- a/docs/blog/claude-squad-vs-munder-difflin/index.html+++ b/docs/blog/claude-squad-vs-munder-difflin/index.html@@ -181,6 +181,7 @@ head-to-head: where each is genuinely better, and how to decide.</p><p>Everything below is detail on that distinction.</p><figure class="fig fig-inline"><img src="/blog/assets/media/claude-squad-vs-munder-difflin/note-1.png" alt="Hand-drawn sketch from “Claude Squad vs Munder Difflin: Which Should You Use?”" loading="lazy" decoding="async" /></figure><h2 id="feature-comparison" tabindex="-1">Feature comparison <a class="anchor" href="#feature-comparison" aria-hidden="true">#</a></h2>+<div class="table-scroll"><table><thead><tr>@@ -242,6 +243,7 @@ head-to-head: where each is genuinely better, and how to decide.</p></tr></tbody></table>+</div><p>No row makes one “win” outright — they reflect two philosophies. Claude Squad optimizes for <em>less</em>;Munder Difflin optimizes for <em>coordination</em>.</p><h2 id="where-claude-squad-is-the-better-pick" tabindex="-1">Where Claude Squad is the better pick <a class="anchor" href="#where-claude-squad-is-the-better-pick" aria-hidden="true">#</a></h2>diff --git a/docs/blog/cline-vs-munder-difflin/index.html b/docs/blog/cline-vs-munder-difflin/index.htmlindex 807b47a9..01242945 100644--- a/docs/blog/cline-vs-munder-difflin/index.html+++ b/docs/blog/cline-vs-munder-difflin/index.html@@ -229,6 +229,7 @@ several agents run at once.</li>not an editor extension’s.</p><figure class="fig fig-inline"><img src="/blog/assets/media/cline-vs-munder-difflin/note-2.png" alt="Hand-drawn sketch from “Cline vs Munder Difflin: In-Editor Agent or Local Hive?”" loading="lazy" decoding="async" /></figure><h2 id="how-to-choose" tabindex="-1">How to choose <a class="anchor" href="#how-to-choose" aria-hidden="true">#</a></h2>+<div class="table-scroll"><table><thead><tr>@@ -259,6 +260,7 @@ not an editor extension’s.</p></tr></tbody></table>+</div><p>And honestly: they can coexist. Use Cline as your in-editor agent and Munder Difflin to orchestrate ahive for the bigger, parallel work. The layers don’t conflict. For a structured way to decide acrossthe whole field, see <a href="/blog/how-to-choose-a-multi-agent-tool/">how to choose a multi-agent tool</a> anddiff --git a/docs/blog/from-one-terminal-to-a-team/index.html b/docs/blog/from-one-terminal-to-a-team/index.htmlindex cd044f35..5df04cd2 100644--- a/docs/blog/from-one-terminal-to-a-team/index.html+++ b/docs/blog/from-one-terminal-to-a-team/index.html@@ -197,6 +197,7 @@ instead of routing through you.</li><figure class="fig fig-inline"><img src="/blog/assets/media/from-one-terminal-to-a-team/note-2.png" alt="Hand-drawn sketch from “From One Terminal to a Team: Scaling Your Claude Code Workflow”" loading="lazy" decoding="async" /></figure><h2 id="the-before-and-after" tabindex="-1">The before-and-after <a class="anchor" href="#the-before-and-after" aria-hidden="true">#</a></h2><p>The same morning, two ways:</p>+<div class="table-scroll"><table><thead><tr>@@ -238,6 +239,7 @@ instead of routing through you.</li></tr></tbody></table>+</div><p>Nothing about the <em>agents</em> changed — they’re the same Claude Code sessions. What changed is the layeraround them, and that’s what moves you from operator to manager.</p><h2 id="knowing-when-to-make-the-jump" tabindex="-1">Knowing when to make the jump <a class="anchor" href="#knowing-when-to-make-the-jump" aria-hidden="true">#</a></h2>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.htmlindex a20ff42c..5b1584aa 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@@ -205,6 +205,7 @@<figure class="fig fig-inline"><img src="/blog/assets/media/github-copilot-cli-vs-claude-code-for-hive-work/note-2.png" alt="Hand-drawn sketch from “GitHub Copilot CLI vs Claude Code as Hive Workers”" loading="lazy" decoding="async" /></figure><h2 id="verdict-dont-pick-mix" tabindex="-1">Verdict: don’t pick — mix <a class="anchor" href="#verdict-dont-pick-mix" aria-hidden="true">#</a></h2><p>The comparison table writes itself once you frame it as lifecycle:</p>+<div class="table-scroll"><table><thead><tr>@@ -246,6 +247,7 @@</tr></tbody></table>+</div><p>Because Munder Difflin makes the engine a per-hire choice, this isn’t an either/or. The pattern that works: <strong>Claude Code as the orchestrator and your long-lived workers</strong> — the agents that need to receive messages, hold context across turns, and run under hook-driven approval gates — and <strong>Copilot CLI workers for burst tasks</strong> the orchestrator dispatches: a scoped refactor, a test-writing pass, a one-shot investigation. Different engines for different roles is the general lesson of mixed fleets, and it also spreads cost across two subscriptions you likely already pay for.</p><p>For everything else in the release — including the built-in Monaco IDE — see the <a href="/blog/launching-munder-difflin-v0-3-3/">v0.3.3 launch post</a>.</p><h2 id="try-both-on-one-floor" tabindex="-1">Try both on one floor <a class="anchor" href="#try-both-on-one-floor" aria-hidden="true">#</a></h2>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.htmlindex b638a07e..35da0f70 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@@ -209,6 +209,7 @@ cloud? For many developers, local-first and open source are non-negotiable.</p></ul><p>Then weight by importance to <em>you</em> (×1 for “nice to have”, ×2 for “matters”, ×3 for “dealbreaker”).Multiply, sum, compare. A worked example for a team whose bottleneck is coordination:</p>+<div class="table-scroll"><table><thead><tr>@@ -263,6 +264,7 @@ Multiply, sum, compare. A worked example for a team whose bottleneck is coordina</tr></tbody></table>+</div><p>Flip the weights for a team whose bottleneck is “just run three parallel tasks,” and the sessionmanager wins handily — that’s the point. The rubric encodes <em>your</em> priorities, so it gives <em>your</em>answer. For a ready-made criteria breakdown across the real tools, see <a href="/blog/claude-code-orchestration-tools-compared/">Claude Code orchestrationdiff --git a/docs/blog/how-to-install-and-use-munder-difflin/index.html b/docs/blog/how-to-install-and-use-munder-difflin/index.htmlindex 8e0ba7e0..91902e13 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@@ -239,6 +239,7 @@ everyone anyway. Check Jio's page for current terms, offers change.</p></div>built against first, and it is the most thoroughly tested of the twelve.</p><h3 id="the-other-engines" tabindex="-1">The other engines <a class="anchor" href="#the-other-engines" aria-hidden="true">#</a></h3><p>Any of these work. Pick one only if you already use it.</p>+<div class="table-scroll"><table><thead><tr>@@ -290,6 +291,7 @@ built against first, and it is the most thoroughly tested of the twelve.</p></tr></tbody></table>+</div><p>You can add more engines later and run several at once, with different agents on different engines.</p><h2 id="step-2-install-your-engine-and-log-in" tabindex="-1">Step 2: Install your engine, and log in <a class="anchor" href="#step-2-install-your-engine-and-log-in" aria-hidden="true">#</a></h2><p><strong>This is the only step where you open a terminal.</strong> Once it is done you will not need it again.</p>@@ -341,6 +343,7 @@ claude</code></pre><p>Then run <code>claude</code> and sign in when the browser opens.</p><h3 id="the-other-engines-1" tabindex="-1">The other engines <a class="anchor" href="#the-other-engines-1" aria-hidden="true">#</a></h3><p>Same idea: install, then run the command once on its own to sign in.</p>+<div class="table-scroll"><table><thead><tr>@@ -355,31 +358,31 @@ claude</code></pre><td>Gemini CLI</td><td><code>npm install -g @google/gemini-cli</code></td><td><code>gemini</code></td>-<td><a href="https://github.com/google-gemini/gemini-cli">github.com/google-gemini/gemini-cli</a></td>+<td><a href="https://github.com/google-gemini/gemini-cli">google-gemini/gemini-cli</a></td></tr><tr><td>Grok</td><td><code>npm install -g @xai-official/grok</code></td><td><code>grok</code></td>-<td><a href="https://docs.x.ai/build/overview">docs.x.ai/build/overview</a></td>+<td><a href="https://docs.x.ai/build/overview">docs.x.ai</a></td></tr><tr><td>Kimi Code</td><td><code>npm install -g @moonshot-ai/kimi-code</code></td><td><code>kimi</code></td>-<td><a href="https://www.kimi.com/code/docs/en/kimi-code-cli/guides/getting-started.html">kimi.com/code/docs</a></td>+<td><a href="https://www.kimi.com/code/docs/en/kimi-code-cli/guides/getting-started.html">kimi.com/code</a></td></tr><tr><td>Qwen</td><td><code>npm install -g @qwen-code/qwen-code@latest</code></td><td><code>qwen</code></td>-<td><a href="https://github.com/QwenLM/qwen-code">github.com/QwenLM/qwen-code</a></td>+<td><a href="https://github.com/QwenLM/qwen-code">QwenLM/qwen-code</a></td></tr><tr><td>Crush</td><td><code>npm install -g @charmland/crush</code></td><td><code>crush</code></td>-<td><a href="https://github.com/charmbracelet/crush">github.com/charmbracelet/crush</a></td>+<td><a href="https://github.com/charmbracelet/crush">charmbracelet/crush</a></td></tr><tr><td>Copilot</td>@@ -391,16 +394,17 @@ claude</code></pre><td>Cursor</td><td><code>curl https://cursor.com/install -fsS | bash</code></td><td><code>cursor-agent</code></td>-<td><a href="https://cursor.com/docs/cli/overview">cursor.com/docs/cli</a></td>+<td><a href="https://cursor.com/docs/cli/overview">cursor.com/docs</a></td></tr><tr><td>pi</td><td><code>npm install -g --ignore-scripts @earendil-works/pi-coding-agent</code></td><td><code>pi</code></td>-<td><a href="https://pi.dev/docs/latest">pi.dev/docs/latest</a></td>+<td><a href="https://pi.dev/docs/latest">pi.dev/docs</a></td></tr></tbody></table>+</div><p>On Windows, the Cursor line is <code>irm 'https://cursor.com/install?win32=true' | iex</code> in PowerShell.</p><h3 id="check-it-actually-installed" tabindex="-1">Check it actually installed <a class="anchor" href="#check-it-actually-installed" aria-hidden="true">#</a></h3><p>Run this, swapping <code>claude</code> for whichever command you installed:</p>diff --git a/docs/blog/local-first-vs-cloud-agent-sdks/index.html b/docs/blog/local-first-vs-cloud-agent-sdks/index.htmlindex 98cad8be..7bf39a69 100644--- a/docs/blog/local-first-vs-cloud-agent-sdks/index.html+++ b/docs/blog/local-first-vs-cloud-agent-sdks/index.html@@ -245,6 +245,7 @@ hive adds <strong>no platform tax</strong> — you pay for the inference you’dfree because it’s just code running on your machine. As you scale from two agents to a dozen, that gapcompounds. (Related: <a href="/blog/do-more-with-less-model-routing/">doing more with less via model routing</a>.)</p><h2 id="the-decision-table" tabindex="-1">The decision table <a class="anchor" href="#the-decision-table" aria-hidden="true">#</a></h2>+<div class="table-scroll"><table><thead><tr>@@ -279,6 +280,7 @@ compounds. (Related: <a href="/blog/do-more-with-less-model-routing/">doing more</tr></tbody></table>+</div><h2 id="the-bottom-line" tabindex="-1">The bottom line <a class="anchor" href="#the-bottom-line" aria-hidden="true">#</a></h2><p>The 2026 cloud agent SDKs are a real, well-built answer to <em>distributed, governed, elastic</em> agent work.A local-first hive is the right answer to <em>private, predictable, own-your-code</em> agent work. Thediff --git a/docs/blog/orca-vs-munder-difflin/index.html b/docs/blog/orca-vs-munder-difflin/index.htmlindex 9deaf81d..a12842f7 100644--- a/docs/blog/orca-vs-munder-difflin/index.html+++ b/docs/blog/orca-vs-munder-difflin/index.html@@ -198,6 +198,7 @@<p>v0.3.3 did add a built-in Monaco IDE — file tree, tabs, side-by-side diffs vs HEAD — but let’s be honest about what it’s for: it’s a <strong>review surface for what your agents changed</strong>, not a place you live. Orca’s editing experience is deeper. Munder Difflin’s bet is that you’ll spend less time in an editor at all, because the office handles the routing, memory, and follow-through that would otherwise pull you back to the keyboard.</p><figure class="fig fig-inline"><img src="/blog/assets/media/orca-vs-munder-difflin/note-2.png" alt="Hand-drawn sketch from “Orca vs Munder Difflin: Agent IDE or Agent Office?”" loading="lazy" decoding="async" /></figure><h2 id="the-boundary-drawn-plainly" tabindex="-1">The boundary, drawn plainly <a class="anchor" href="#the-boundary-drawn-plainly" aria-hidden="true">#</a></h2>+<div class="table-scroll"><table><thead><tr>@@ -254,6 +255,7 @@</tr></tbody></table>+</div><h2 id="when-to-pick-which" tabindex="-1">When to pick which <a class="anchor" href="#when-to-pick-which" aria-hidden="true">#</a></h2><p><strong>Pick Orca</strong> if your bottleneck is interactive throughput: you’re at the keyboard, you want several agents on one problem, and you want first-class tools for comparing, annotating, and shipping their diffs — plus a phone app to check in from the couch.</p><p><strong>Pick Munder Difflin</strong> if your bottleneck is <em>you being required at all</em>: you want work triggered from Slack or a schedule, routed by an orchestrator, informed by shared memory, and bounded by budgets and approval gates while it runs overnight.</p>diff --git a/docs/blog/qm-vs-munder-difflin/index.html b/docs/blog/qm-vs-munder-difflin/index.htmlindex 3db42395..52e3c3ac 100644--- a/docs/blog/qm-vs-munder-difflin/index.html+++ b/docs/blog/qm-vs-munder-difflin/index.html@@ -196,6 +196,7 @@ Grok, Kimi, and more, riding the subscriptions you already pay for. The floor is(yes, it looks like The Office), a GOD orchestrator routes work, agents share long-termmemory, 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="table-scroll"><table><thead><tr>@@ -237,6 +238,7 @@ memory, and background work runs as <strong>scheduled missions</strong> with Sla</tr></tbody></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,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.htmlindex 24b7e248..a8b85c2a 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@@ -205,6 +205,7 @@ ceiling on a dedicated box with <code>sudo sysctl iogpu.wired_limit_mb=<MB>model. A practical rule: <strong>plan for the model to use about two-thirds of your unified memory</strong> and leavethe rest for context and the OS.</p><p>Current Mac Mini lineup and what each tier comfortably runs (4-bit quant):</p>+<div class="table-scroll"><table><thead><tr>@@ -247,11 +248,13 @@ the rest for context and the OS.</p></tr></tbody></table>+</div><p><em>(Newer Mac Mini chips raise bandwidth and the memory ceiling, but the same “model ≈ two-thirds ofunified 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 asGGUF/MLX): <code>ollama pull <tag></code>, then reference the model as <code>local/<tag></code> in OpenCode or <code>ollama/<tag></code> inCrush (more on that in Step 3).</p>+<div class="table-scroll"><table><thead><tr>@@ -300,6 +303,7 @@ Crush (more on that in Step 3).</p></tr></tbody></table>+</div><p>These picks are drawn from the model families our open-source model catalog tracks — the single source oftruth we keep version-current.The companion guide, <a href="/blog/run-munder-difflin-on-open-models/">running Munder Difflin on opendiff --git a/docs/blog/run-munder-difflin-on-open-models/index.html b/docs/blog/run-munder-difflin-on-open-models/index.htmlindex e0965a7d..35f25405 100644--- a/docs/blog/run-munder-difflin-on-open-models/index.html+++ b/docs/blog/run-munder-difflin-on-open-models/index.html@@ -173,6 +173,7 @@<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="table-scroll"><table><thead><tr>@@ -214,6 +215,7 @@</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>@@ -223,6 +225,7 @@<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><your-name>/<model-id></code>.</li></ul><p>The <em>local</em> route differs slightly by engine — same id, different prefix and wiring:</p>+<div class="table-scroll"><table><thead><tr>@@ -249,6 +252,7 @@</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>@@ -262,6 +266,7 @@ ollama serve <span class="token comment"># exposes the OpenAI-<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>+<div class="table-scroll"><table><thead><tr>@@ -322,12 +327,14 @@ ollama serve <span class="token comment"># exposes the OpenAI-</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>+<div class="table-scroll"><table><thead><tr>@@ -382,6 +389,7 @@ ollama serve <span class="token comment"># exposes the OpenAI-</tr></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>diff --git a/docs/blog/what-is-aeo-for-dev-tools/index.html b/docs/blog/what-is-aeo-for-dev-tools/index.htmlindex 348196bf..a0290fda 100644--- a/docs/blog/what-is-aeo-for-dev-tools/index.html+++ b/docs/blog/what-is-aeo-for-dev-tools/index.html@@ -185,6 +185,7 @@ a <em>quoted sentence</em>. This post is the playbook we actually run, with the<p><strong>SEO optimizes a page to rank so a human clicks it. AEO optimizes content to be the source an AIquotes in its answer — often with no click at all.</strong> They share foundations (crawlable, fast,well-structured pages) but diverge on intent:</p>+<div class="table-scroll"><table><thead><tr>@@ -221,6 +222,7 @@ well-structured pages) but diverge on intent:</p></tr></tbody></table>+</div><p>The good news: you don’t choose between them. AEO is mostly SEO done with a tighter discipline —clearer answers, stronger structure, honest facts. A page that an LLM can confidently quote is also apage Google’s AI Overviews and traditional results reward.</p>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.htmlindex e4a8ebbb..dcc5aa75 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@@ -184,6 +184,7 @@ weekly and vendors revise details. Where a figure comes from a secondary aggregaprimary vendor, we say so. Munder Difflin is our own project; we’ve tried to report the rest straight.</p></blockquote><h2 id="at-a-glance" tabindex="-1">At a glance <a class="anchor" href="#at-a-glance" aria-hidden="true">#</a></h2>+<div class="table-scroll"><table><thead><tr>@@ -220,6 +221,7 @@ primary vendor, we say so. Munder Difflin is our own project; we’ve tried to r</tr></tbody></table>+</div><h2 id="models-learned-to-act-not-just-answer" tabindex="-1">Models learned to act, not just answer <a class="anchor" href="#models-learned-to-act-not-just-answer" aria-hidden="true">#</a></h2><p>The headline release is <a href="https://www.anthropic.com/claude/opus">Anthropic’s Claude Opus 4.8</a>, out<a href="https://techcrunch.com/2026/05/28/anthropic-releases-opus-4-8-with-new-dynamic-workflow-tool/">May 28, 2026 — 41 days after Opus 4.7</a>.