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
<!doctype html>
<html lang="en">
<head>
<!-- Google tag (gtag.js), Google Ads AW-18388216863. Loaded only when Do Not Track is
off, so it follows the same rule as the PostHog block on this site. -->
<script>
window.dataLayer = window.dataLayer || [];
function gtag(){ dataLayer.push(arguments); }
if (navigator.doNotTrack !== '1') {
var gtagScript = document.createElement('script');
gtagScript.async = true;
gtagScript.src = 'https://www.googletagmanager.com/gtag/js?id=AW-18388216863';
document.head.appendChild(gtagScript);
gtag('js', new Date());
gtag('config', 'AW-18388216863');
}
</script>
<meta charset="UTF-8" />
<meta name="viewport" content="width=device-width, initial-scale=1.0" />
<title>Running AI Agents Safely: Permission Modes and Sandboxing — Munder Difflin Blog</title>
<meta name="description" content="How to run autonomous AI agents safely: permission modes and the bypassPermissions foot-gun, workspace isolation, and what to gate vs. allow." />
<link rel="canonical" href="https://munderdiffl.in/blog/agent-security-and-sandboxing/" />
<link rel="icon" type="image/png" href="https://munderdiffl.in/logo.png" />
<link rel="apple-touch-icon" href="https://munderdiffl.in/logo.png" />
<meta name="theme-color" content="#F5F2E8" />
<!-- Open Graph -->
<meta property="og:site_name" content="Munder Difflin Blog" />
<meta property="og:type" content="article" />
<meta property="og:title" content="Running AI Agents Safely: Permission Modes and Sandboxing — Munder Difflin Blog" />
<meta property="og:description" content="How to run autonomous AI agents safely: permission modes and the bypassPermissions foot-gun, workspace isolation, and what to gate vs. allow." />
<meta property="og:url" content="https://munderdiffl.in/blog/agent-security-and-sandboxing/" />
<meta property="og:image" content="https://munderdiffl.in/blog/assets/media/agent-security-and-sandboxing/hero.png" />
<meta property="og:image:width" content="1600" />
<meta property="og:image:height" content="900" />
<meta property="og:image:alt" content="Illustration: Running AI Agents Safely: Permission Modes and Sandboxing" />
<meta property="og:locale" content="en_US" />
<meta property="article:published_time" content="2026-06-04T00:00:00.000Z" />
<meta property="article:author" content="Chaitanya Giri" />
<meta property="article:tag" content="Security" />
<meta property="article:tag" content="Guardrails" />
<meta property="article:tag" content="Autonomous" />
<meta property="article:tag" content="Internals" />
<!-- Twitter -->
<meta name="twitter:card" content="summary_large_image" />
<meta name="twitter:title" content="Running AI Agents Safely: Permission Modes and Sandboxing — Munder Difflin Blog" />
<meta name="twitter:description" content="How to run autonomous AI agents safely: permission modes and the bypassPermissions foot-gun, workspace isolation, and what to gate vs. allow." />
<meta name="twitter:image" content="https://munderdiffl.in/blog/assets/media/agent-security-and-sandboxing/hero.png" />
<meta name="twitter:image:alt" content="Illustration: Running AI Agents Safely: Permission Modes and Sandboxing" />
<link rel="alternate" type="application/atom+xml" title="Munder Difflin Blog" href="https://munderdiffl.in/blog/feed.xml" />
<link rel="preconnect" href="https://fonts.googleapis.com" />
<link rel="preconnect" href="https://fonts.gstatic.com" crossorigin />
<link href="https://fonts.googleapis.com/css2?family=Fraunces:opsz,[email protected],500;9..144,600;9..144,700;9..144,900&family=Source+Serif+4:opsz,[email protected],400;8..60,600;8..60,700&family=Space+Grotesk:wght@500;600;700&family=JetBrains+Mono:wght@400;600&display=swap" rel="stylesheet" />
<link rel="stylesheet" href="/blog/assets/blog-press.css" />
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "BlogPosting",
"headline": "Running AI Agents Safely: Permission Modes and Sandboxing",
"description": "How to run autonomous AI agents safely: permission modes and the bypassPermissions foot-gun, workspace isolation, and what to gate vs. allow.",
"image": ["https://munderdiffl.in/blog/assets/media/agent-security-and-sandboxing/hero.png"],
"datePublished": "2026-06-04T00:00:00.000Z",
"dateModified": "2026-06-04T00:00:00.000Z",
"author": { "@type": "Person", "name": "Chaitanya Giri" },
"publisher": {
"@type": "Organization",
"name": "Munder Difflin",
"logo": { "@type": "ImageObject", "url": "https://munderdiffl.in/logo.png" }
},
"mainEntityOfPage": { "@type": "WebPage", "@id": "https://munderdiffl.in/blog/agent-security-and-sandboxing/" },
"keywords": "ai agent security, ai agent sandboxing, claude code permission modes, bypasspermissions safety, autonomous agent guardrails",
"articleSection": "Guides",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/agent-security-and-sandboxing/"
}
</script>
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "BreadcrumbList",
"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": "Running AI Agents Safely: Permission Modes and Sandboxing", "item": "https://munderdiffl.in/blog/agent-security-and-sandboxing/" }
]
}
</script>
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{ "@type": "Question", "name": "What is the safest way to run autonomous AI coding agents?", "acceptedAnswer": { "@type": "Answer", "text": "Defense in depth, not a single switch. The per-action permission prompt is one gate; once you turn it off for unattended speed (bypassPermissions / 'auto mode'), safety has to come from structure: run each agent in an isolated workspace, scope its filesystem and credentials, keep destructive operations out of reach, observe every action, and audit. Gate the few genuinely dangerous things; let the rest flow." } },
{ "@type": "Question", "name": "Is it safe to run Claude Code with --permission-mode bypassPermissions?", "acceptedAnswer": { "@type": "Answer", "text": "It's the right default for unattended, control-room workflows — but it's a foot-gun on production repos, because the agent won't pause before file edits or shell commands. Make it safe by containing it: a throwaway git worktree, no production credentials in the environment, and an audit log. Reserve un-sandboxed bypass mode for repos where the worst case is acceptable." } },
{ "@type": "Question", "name": "How do you sandbox an AI agent?", "acceptedAnswer": { "@type": "Answer", "text": "Layer constraints: isolate its workspace (a per-agent git worktree so changes are a throwaway branch), limit what it can reach (scoped filesystem, no prod secrets), keep a single committer so agents can't rewrite history, use a PreToolUse hook as a programmable policy gate, and validate any untrusted input at the boundary before the agent acts on it." } }
]
}
</script>
<script>
// Anonymous blog analytics (PostHog) — same posture as the main site and the
// app's TELEMETRY.md: public write-only token ('' = inert), anonymous-only
// (person_profiles 'never'), no autocapture/recording, DNT respected.
(function () {
var PH_KEY = 'phc_qJzsokwcbtTGHemnvywR3nSqivNNgMjHeNVEWkZNydFB';
if (!PH_KEY || navigator.doNotTrack === '1') return;
var s = document.createElement('script');
s.async = true;
s.src = 'https://us-assets.i.posthog.com/static/array.js';
s.onload = function () {
if (!window.posthog) return;
window.posthog.init(PH_KEY, {
api_host: 'https://us.i.posthog.com',
person_profiles: 'never',
autocapture: false,
before_send: function (event) {
// PostHog fills $ip from the connection unless the event carries one.
// Null it here and disable the GeoIP lookup so no IP, city, postal
// code or lat/long is ever derived or stored for this page view.
if (event && event.properties) {
event.properties.$ip = null;
event.properties.$geoip_disable = true;
}
return event;
},
disable_session_recording: true,
respect_dnt: true
});
};
document.head.appendChild(s);
})();
</script>
</head>
<body class="theme-press">
<a class="skip-link" href="#main">Skip to content</a>
<nav id="nav">
<a class="brand" href="https://munderdiffl.in" aria-label="Munder Difflin home">
<img class="brand-logo" src="https://munderdiffl.in/logo.png" width="1318" height="840" alt="Munder Difflin, Inc. — Multi-Agent Harness" />
</a>
<span class="spacer"></span>
<div class="links">
<a href="/blog/">Latest</a>
<a href="/blog/topics/">Topics</a>
<a href="/blog/about/">About</a>
<a href="https://munderdiffl.in">↗ Product</a>
</div>
<div class="nav-actions">
<a class="btn sm" href="https://munderdiffl.in/blog/feed.xml" aria-label="RSS feed">⤵ RSS</a>
<a class="btn sm primary" href="https://munderdiffl.in/#install">⤓ Download</a>
</div>
</nav>
<main id="main">
<article class="t-guides">
<header class="post-head wrap">
<nav class="crumbs" aria-label="Breadcrumb">
<a href="/blog/">Blog</a>
<span>/</span>
<a href="/blog/topics/guides/">Guides</a>
</nav>
<h1>Running AI Agents Safely: Permission Modes and Sandboxing</h1>
<p class="dek">How to run autonomous AI agents safely: permission modes and the bypassPermissions foot-gun, workspace isolation, and what to gate vs. allow.</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="2026-06-04">Jun 4, 2026</time>
<span class="dot" aria-hidden="true">·</span>
<span>5 min read</span>
<span class="tag kind">Guides</span>
</div>
</header>
<div class="hero-wrap wrap"><figure class="fig fig-hero"><img src="/blog/assets/media/agent-security-and-sandboxing/hero.png" alt="Illustration: Running AI Agents Safely: Permission Modes and Sandboxing" loading="lazy" decoding="async" /></figure></div>
<div class="post-wrap wrap">
<div class="prose">
<div class="callout tldr"><span class="ic">TL;DR</span><p>Security for autonomous agents isn't one
switch — it's <strong>defense in depth</strong>. The per-action <strong>permission prompt</strong> is one
gate, but the moment you turn it off for unattended speed (<code>bypassPermissions</code>, a.k.a. "auto
mode"), you've removed the last interactive check — so safety has to become <strong>structural</strong>:
isolate the workspace, scope the filesystem and credentials, keep destructive ops out of reach, observe
every action, and audit. <strong>Gate the few genuinely dangerous things; let the rest flow.</strong></p></div>
<p>An AI agent that can edit files and run shell commands is, by design, holding a loaded tool. The whole
point of an agent is that it acts without asking you about every step — which is also exactly what makes
security a real design question, not an afterthought. This post is a practical map of the layers that
keep autonomous agents safe, grounded in how a <a href="/#what">multi-agent harness</a> actually wires them.</p>
<h2 id="the-core-tension-autonomy-vs-a-gate" tabindex="-1">The core tension: autonomy vs. a gate <a class="anchor" href="#the-core-tension-autonomy-vs-a-gate" aria-hidden="true">#</a></h2>
<p>Coding agents ship with a built-in gate: the <strong>permission prompt</strong>. By default Claude Code pauses before
a sensitive action and asks. That’s great when a human is watching — and useless when the entire premise
is “run a hive of agents unattended overnight.” You can’t approve a prompt you’re asleep for. If the prompts are wearing you down in the daytime too, see <a href="/blog/why-does-claude-code-keep-asking-for-permission/">why Claude Code keeps asking for permission</a>.</p>
<p>So autonomous setups reach for the other end of the dial. In Munder Difflin that’s <strong>auto mode</strong>, which
spawns each agent with <code>--permission-mode bypassPermissions</code>. The onboarding flow says it plainly:</p>
<blockquote>
<p>Agents in the harness run <strong>unattended</strong>. By default, every agent is spawned with
<code>--permission-mode bypassPermissions</code> — meaning claude won’t pause to ask you before file edits or
shell commands. This is the right default for the “control room” experience; it’s also a loaded
foot-gun on production repos.</p>
</blockquote>
<p>That honesty is the right framing. Bypass mode isn’t reckless — it’s the <em>correct</em> default for an
unattended control room. But it removes the last interactive gate, so the safety has to live somewhere
else. The rest of this post is “somewhere else.” One side effect worth knowing: a bypass session also skips plan mode’s edit block, as <a href="/blog/how-to-use-claude-code-plan-mode/">how to use Claude Code plan mode</a> explains.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/agent-security-and-sandboxing/note-1.png" alt="Hand-drawn sketch from “Running AI Agents Safely: Permission Modes and Sandboxing”" loading="lazy" decoding="async" /></figure>
<h2 id="when-the-prompt-is-off-structure-is-your-safety" tabindex="-1">When the prompt is off, structure is your safety <a class="anchor" href="#when-the-prompt-is-off-structure-is-your-safety" aria-hidden="true">#</a></h2>
<p>If you can’t rely on a human clicking “approve,” you make the dangerous things structurally hard to
reach. Five layers do most of the work.</p>
<p><strong>1. Isolate the workspace.</strong> Run each agent in its own <a href="/blog/claude-code-git-worktrees-vs-hive/">git worktree</a>,
so its changes are a throwaway branch in a separate directory — never your main checkout. An agent that
goes off the rails overnight has made a mess you can delete, not a mess you have to untangle. Isolation
turns “the agent broke something” into “the agent’s branch is wrong,” which is a much smaller problem.</p>
<p><strong>2. Shrink the blast radius.</strong> In a hive, <strong>agents never call git</strong> — only the coordinating main
process commits, the <a href="/blog/single-committer-git-pattern/">single-committer pattern</a>. That alone removes
a whole class of damage: an agent can’t force-push, rewrite history, or stomp another agent’s branch,
because it doesn’t hold the git hammer at all. The same instinct applies to everything sharp: keep
production credentials out of the agent’s environment, and scope its filesystem access. (Munder Difflin’s
own file IPC is sandboxed to a root path — operations take a root plus a relative path, so a request
can’t wander outside it.)</p>
<p><strong>3. Keep a programmable gate.</strong> Turning off the <em>prompt</em> doesn’t mean turning off <em>policy</em>. Claude
Code’s <strong>PreToolUse hook</strong> fires before every tool call — Munder Difflin already wires one on every tool
(it uses it to animate the live office floor), and that exact hook point is where you can inspect a tool
call and allow, ask, or <strong>deny</strong> it. A few lines of policy at PreToolUse — “never let <code>rm -rf</code> outside
the worktree,” “block writes to <code>.env</code>” — is a gate that works even when no human is watching. (More on
the lifecycle in <a href="/blog/claude-code-hooks-explained/">Claude Code hooks, explained</a>.)</p>
<p><strong>4. Don’t trust input from outside.</strong> The moment an agent acts on input from beyond its own session —
a webhook, a Slack message, an issue body — that input is untrusted. The harness’s Slack integration is
the model to copy: every inbound request is HMAC-verified against a signing secret, replay-guarded to a
five-minute window, and size-capped <em>before a single byte is parsed</em>. Anything that fails gets a <code>403</code>.
Validate at the boundary, then let the agent act.</p>
<p><strong>5. Observe and audit.</strong> You can’t secure what you can’t see. An <a href="/blog/append-only-event-log-agents/">append-only event
log</a> records every action in order, so “what did the agent do at
2 a.m.?” has a precise, timestamped answer — and the genuinely critical calls can still pause for a human
through <a href="/blog/human-in-the-loop-approving-ai-agents/">native approvals</a>. Bypass mode and a human gate
aren’t mutually exclusive: most actions flow, the few that matter still wait.</p>
<figure class="fig fig-inline"><img src="/blog/assets/media/agent-security-and-sandboxing/note-2.png" alt="Hand-drawn sketch from “Running AI Agents Safely: Permission Modes and Sandboxing”" loading="lazy" decoding="async" /></figure>
<h2 id="what-to-gate-vs-what-to-allow" tabindex="-1">What to gate vs. what to allow <a class="anchor" href="#what-to-gate-vs-what-to-allow" aria-hidden="true">#</a></h2>
<p>Security theater is gating everything until the human rubber-stamps a hundred prompts and stops reading.
The goal is a <em>narrow</em> gate. A reasonable default policy:</p>
<p><strong>Gate (require a human, or simply disallow):</strong></p>
<ul>
<li>Destructive or irreversible actions — dropping data, deleting branches, <code>rm -rf</code>, force-push.</li>
<li>Spending real money — provisioning paid infra, large API spend, anything with a bill.</li>
<li>Touching secrets or production — reading credentials, writing to a prod database, deploying.</li>
<li>Network calls to systems of record outside the sandbox.</li>
</ul>
<p><strong>Allow (let it flow):</strong></p>
<ul>
<li>Reads — inspecting code, running searches, reading docs.</li>
<li>Edits <strong>inside an isolated worktree</strong>, where the worst case is a bad branch.</li>
<li>Running the test suite and build — in fact, <a href="/blog/claude-code-automation-while-you-sleep/">make tests the agent’s bar</a> for “done.”</li>
</ul>
<p>The dividing line is <em>reversibility</em>. If a mistake is a <code>git checkout</code> away from undone, let the agent
move; if it isn’t, gate it.</p>
<h2 id="a-safe-by-default-setup-checklist" tabindex="-1">A safe-by-default setup checklist <a class="anchor" href="#a-safe-by-default-setup-checklist" aria-hidden="true">#</a></h2>
<p>Before you let agents run unattended:</p>
<ol>
<li>Each agent in its <strong>own worktree</strong> (or at least its own branch), never shared main.</li>
<li><strong>No production credentials</strong> in the agent’s environment — dev/sandbox only.</li>
<li>A <strong>PreToolUse policy</strong> for the handful of actions you never want to happen.</li>
<li>An <strong>audit log</strong> you can read in the morning.</li>
<li><strong>Bypass mode only where the worst case is acceptable</strong> — keep prompts on for any repo where it isn’t.</li>
<li>The genuinely critical still routed to a <strong>human gate</strong>.</li>
</ol>
<p>Get those in place and unattended agents are a force multiplier, not a liability. The pattern is always
the same: autonomy where mistakes are cheap, gates where they’re not, and isolation so “cheap” stays
cheap. A good <a href="/#how">harness orchestrator</a> bakes most of this in — isolated worktrees, single-committer
git, lifecycle hooks, and an audit trail — so safe-by-default is the path of least resistance.</p>
<p>Want to see unattended agents that are isolated, observable, and auditable by default? You can
<a href="/#install">download Munder Difflin</a> free — it’s open source.</p>
<section class="faq" aria-label="Frequently asked questions">
<h2 id="faq">FAQ</h2>
<div class="qa">
<p class="q">What is the safest way to run autonomous AI coding agents?</p>
<p class="a">Defense in depth, not a single switch. The per-action permission prompt is one gate; once you turn it off for unattended speed (bypassPermissions / 'auto mode'), safety has to come from structure: run each agent in an isolated workspace, scope its filesystem and credentials, keep destructive operations out of reach, observe every action, and audit. Gate the few genuinely dangerous things; let the rest flow.</p>
</div>
<div class="qa">
<p class="q">Is it safe to run Claude Code with --permission-mode bypassPermissions?</p>
<p class="a">It's the right default for unattended, control-room workflows — but it's a foot-gun on production repos, because the agent won't pause before file edits or shell commands. Make it safe by containing it: a throwaway git worktree, no production credentials in the environment, and an audit log. Reserve un-sandboxed bypass mode for repos where the worst case is acceptable.</p>
</div>
<div class="qa">
<p class="q">How do you sandbox an AI agent?</p>
<p class="a">Layer constraints: isolate its workspace (a per-agent git worktree so changes are a throwaway branch), limit what it can reach (scoped filesystem, no prod secrets), keep a single committer so agents can't rewrite history, use a PreToolUse hook as a programmable policy gate, and validate any untrusted input at the boundary before the agent acts on it.</p>
</div>
</section>
<hr />
<div class="share">
<span class="lbl">Tags</span>
<a class="tag" href="/blog/tags/security/">Security</a><a class="tag" href="/blog/tags/guardrails/">Guardrails</a><a class="tag" href="/blog/tags/autonomous/">Autonomous</a><a class="tag" href="/blog/tags/internals/">Internals</a>
</div>
</div>
<aside class="toc-rail" aria-label="Table of contents">
<div class="tt">On this page</div>
<ol>
<li class="lvl-2"><a href="#the-core-tension-autonomy-vs-a-gate">The core tension: autonomy vs. a gate</a></li><li class="lvl-2"><a href="#when-the-prompt-is-off-structure-is-your-safety">When the prompt is off, structure is your safety</a></li><li class="lvl-2"><a href="#what-to-gate-vs-what-to-allow">What to gate vs. what to allow</a></li><li class="lvl-2"><a href="#a-safe-by-default-setup-checklist">A safe-by-default setup checklist</a></li>
</ol>
</aside>
</div>
<footer class="post-foot wrap"><div class="prevnext">
<a class="pn prev" href="/blog/agents-rule-of-two/"><span class="k">← Older</span><span class="t">The Agents Rule of Two: A Simple Safety Rule for Coding Agents</span></a>
<a class="pn next" href="/blog/the-multi-agent-cost-playbook/"><span class="k">Newer →</span><span class="t">The Multi-Agent Cost Playbook: Five Levers to Cut AI Agent Spend</span></a>
</div>
<section class="related">
<h2>Related in Guides</h2>
<div class="post-grid">
<article class="card t-guides">
<a class="thumb t-guides" href="/blog/claude-md-best-practices/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/claude-md-best-practices/hero.png" alt="" loading="lazy" decoding="async" /></a>
<div class="body">
<div class="meta">
<time datetime="2026-10-26">Oct 26, 2026</time>
<span class="dot" aria-hidden="true"></span>
<span>4 min</span>
</div>
<h3><a href="/blog/claude-md-best-practices/">CLAUDE.md Best Practices: What to Write, Where It Lives, How Long (Oct 2026)</a></h3>
<p class="dek">CLAUDE.md best practices from Anthropic's docs, checked 26 Oct 2026: what to put in it, where each file lives, the 200 line target and how AGENTS.md loads.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/claude-md-best-practices/" aria-label="Read: CLAUDE.md Best Practices: What to Write, Where It Lives, How Long (Oct 2026)">Read →</a>
</div>
</div>
</article>
<article class="card t-guides">
<a class="thumb t-guides" href="/blog/codex-computer-use/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/codex-computer-use/hero.png" alt="" loading="lazy" decoding="async" /></a>
<div class="body">
<div class="meta">
<time datetime="2026-10-26">Oct 26, 2026</time>
<span class="dot" aria-hidden="true"></span>
<span>4 min</span>
</div>
<h3><a href="/blog/codex-computer-use/">Codex Computer Use: How to Turn It On, Plans and Limits (Oct 2026)</a></h3>
<p class="dek">Codex computer use lets the ChatGPT desktop app see and click apps on macOS and Windows. Setup, which plans get it, locked use and how Claude's compares.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/codex-computer-use/" aria-label="Read: Codex Computer Use: How to Turn It On, Plans and Limits (Oct 2026)">Read →</a>
</div>
</div>
</article>
<article class="card t-guides">
<a class="thumb t-guides" href="/blog/cursor-pricing/" aria-hidden="true" tabindex="-1"><img src="/blog/assets/media/cursor-pricing/hero.png" alt="" loading="lazy" decoding="async" /></a>
<div class="body">
<div class="meta">
<time datetime="2026-10-26">Oct 26, 2026</time>
<span class="dot" aria-hidden="true"></span>
<span>5 min</span>
</div>
<h3><a href="/blog/cursor-pricing/">Cursor Pricing (Oct 2026): Every Plan After the 12 Oct Change</a></h3>
<p class="dek">Cursor pricing as of 26 Oct 2026: Free, Plus $8, Premium $30, Super $100, Ultra $200, Teams seats, what changed on 12 Oct and what old plans keep.</p>
<div class="foot">
<span class="tag kind">Guides</span>
<a class="more" href="/blog/cursor-pricing/" aria-label="Read: Cursor Pricing (Oct 2026): Every Plan After the 12 Oct Change">Read →</a>
</div>
</div>
</article>
</div>
</section>
</footer>
</article>
</main>
<footer class="site-foot">
<div class="wrap">
<div class="foot-top">
<a class="foot-brand brand" href="https://munderdiffl.in" aria-label="Munder Difflin home">
<img class="brand-logo" src="https://munderdiffl.in/logo.png" width="1318" height="840" alt="Munder Difflin, Inc. — Multi-Agent Harness" />
</a>
<div class="foot-links">
<a href="/blog/">Latest</a>
<a href="/blog/topics/">Topics</a>
<a href="/blog/about/">About</a>
<a href="https://munderdiffl.in/blog/feed.xml">RSS</a>
<a href="https://github.com/chaitanyagiri/munder-difflin">GitHub</a>
<a href="https://munderdiffl.in">munderdiffl.in ↗</a>
</div>
</div>
<div class="foot-meta">
© 2026 Munder Difflin · Local-first · MIT-licensed ·
Built on <a href="https://munderdiffl.in">Electron, React, Pixi.js, xterm.js & node-pty</a>.
Run an office of agents while you sleep.
</div>
</div>
</footer>
<script>
// sticky-nav border on scroll
(function () {
var nav = document.getElementById('nav');
if (!nav) return;
var onScroll = function () { nav.classList.toggle('scrolled', window.scrollY > 8); };
onScroll(); window.addEventListener('scroll', onScroll, { passive: true });
})();
// YouTube click-to-load: thumbnail button → youtube-nocookie iframe
(function () {
document.addEventListener('click', function (e) {
var btn = e.target.closest('.yt-load');
if (!btn) return;
var box = btn.closest('.yt');
var id = box && box.getAttribute('data-yt');
// Only an 11 character YouTube video id may reach the iframe URL.
if (!id || !/^[A-Za-z0-9_-]{11}$/.test(id)) return;
var frame = document.createElement('iframe');
frame.src = 'https://www.youtube-nocookie.com/embed/' + encodeURIComponent(id) + '?autoplay=1';
frame.title = box.getAttribute('data-title') || 'YouTube video';
frame.allow = 'accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture; web-share';
frame.setAttribute('allowfullscreen', '');
box.replaceChildren(frame);
});
})();
</script>
</body>
</html>