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>Run Munder Difflin Locally on a Mac Mini — Munder Difflin Blog</title>
<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" />
<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="Run Munder Difflin Locally on a Mac Mini — Munder Difflin Blog" />
<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/blog/assets/media/run-munder-difflin-on-a-mac-mini/hero.png" />
<meta property="og:image:width" content="1600" />
<meta property="og:image:height" content="900" />
<meta property="og:image:alt" content="Illustration: Run Munder Difflin Locally on a Mac Mini" />
<meta property="og:locale" content="en_US" />
<meta property="article:published_time" content="2026-06-22T00:00:00.000Z" />
<meta property="article:author" content="Chaitanya Giri" />
<meta property="article:tag" content="Guides" />
<meta property="article:tag" content="Local-First" />
<meta property="article:tag" content="Mac Mini" />
<meta property="article:tag" content="Ollama" />
<meta property="article:tag" content="LM Studio" />
<meta property="article:tag" content="BYOK" />
<!-- 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="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/blog/assets/media/run-munder-difflin-on-a-mac-mini/hero.png" />
<meta name="twitter:image:alt" content="Illustration: Run Munder Difflin Locally on a Mac Mini" />
<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": "Run Munder Difflin Locally on a Mac Mini",
"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/blog/assets/media/run-munder-difflin-on-a-mac-mini/hero.png"],
"datePublished": "2026-06-22T00:00:00.000Z",
"dateModified": "2026-09-10T00:00:00.000Z",
"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/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, m6 mac mini local llm",
"articleSection": "Guides",
"inLanguage": "en-US",
"url": "https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/"
}
</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": "Run Munder Difflin Locally on a Mac Mini", "item": "https://munderdiffl.in/blog/run-munder-difflin-on-a-mac-mini/" }
]
}
</script>
<script type="application/ld+json">
{
"@context": "https://schema.org",
"@type": "FAQPage",
"mainEntity": [
{ "@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." } }
]
}
</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>Run Munder Difflin Locally on a Mac Mini</h1>
<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>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/run-munder-difflin-on-a-mac-mini/hero.png" alt="Illustration: Run Munder Difflin Locally on a Mac Mini" 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>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=<MB></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>Memory</th>
<th>Comfortable model size</th>
</tr>
</thead>
<tbody>
<tr>
<td>M6</td>
<td>16 GB</td>
<td>About 8B</td>
</tr>
<tr>
<td>M6</td>
<td>24 GB</td>
<td>Up to about 14B</td>
</tr>
<tr>
<td>M6</td>
<td>32 GB</td>
<td>Up to about 32B, with shorter context</td>
</tr>
<tr>
<td>M5 Pro</td>
<td>24 GB</td>
<td>Up to about 14B, generated faster</td>
</tr>
<tr>
<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>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 <tag></code>:</p>
<div class="table-scroll">
<table>
<thead>
<tr>
<th>Memory</th>
<th>Pick (Ollama tag)</th>
<th>Good for</th>
</tr>
</thead>
<tbody>
<tr>
<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>24 GB</td>
<td><code>qwen3:14b</code> or <code>mistral-small:24b</code></td>
<td>A roomy generalist</td>
</tr>
<tr>
<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>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>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>
<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-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/<tag></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>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/<tag></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">"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>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-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 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><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>
<section class="faq" aria-label="Frequently asked questions">
<h2 id="faq">FAQ</h2>
<div class="qa">
<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'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.</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">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 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">How does Pi use a local model?</p>
<p class="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.</p>
</div>
</section>
<hr />
<div class="share">
<span class="lbl">Tags</span>
<a class="tag" href="/blog/tags/guides/">Guides</a><a class="tag" href="/blog/tags/local-first/">Local-First</a><a class="tag" href="/blog/tags/mac-mini/">Mac Mini</a><a class="tag" href="/blog/tags/ollama/">Ollama</a><a class="tag" href="/blog/tags/lm-studio/">LM Studio</a><a class="tag" href="/blog/tags/byok/">BYOK</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="#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 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">
<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>