Local AI Agent Wiki — A knowledge wiki for AI agents

    
  
      
      
      


    
  
  
  

    
      
    
    
  

    
      
    
      
    
    __md_scope=new URL(".",location),__md_hash=e=>[...e].reduce(((e,_)=>(e<<5)-e+_.charCodeAt(0)),0),__md_get=(e,_=localStorage,t=__md_scope)=>JSON.parse(_.getItem(t.pathname+"."+e)),__md_set=(e,_,t=localStorage,a=__md_scope)=>{try{t.setItem(a.pathname+"."+e,JSON.stringify(_))}catch(e){}}
    
      

    
    
  
  
  
    
  
    
    
    
    
    <div>
      
    </div>
    <div>
      
    </div>
    
    
    <div class="md-container">
      
      
        
          
        
      
      <main class="md-main">
        <div class="md-main__inner md-grid">
          
          
  
  



<nav class="landing-nav">
  <div class="landing-nav__container">
    <a href="/" class="landing-nav__wordmark">
      <span class="doto-w700">local ai</span>
    </a>
    <div class="landing-nav__links">
      <a href="#features" class="landing-nav__link">features</a>
      <a href="#docs" class="landing-nav__link">wiki</a>
      <a href="#agent-layer" class="landing-nav__link">for agents</a>
      <a href="/content/" class="landing-nav__link">browse</a>
    </div>
  </div>
</nav>


<section class="landing-hero dot-grid--light">
  <div class="landing-hero__container">
    <h1 class="landing-hero__wordmark doto-w700">Local AI Agent Wiki</h1>
    <p class="landing-hero__subhead">
      A knowledge wiki that AI agents can read, analyze, and reproduce local AI workflows on their own machines.
      Structured markdown, cited sources, and machine-readable endpoints — no JS required.
    </p>
    <div class="landing-hero__ctas">
      <a href="/content/" class="landing-hero__cta-primary">browse the wiki</a>
      <a href="#agent-layer" class="landing-hero__cta-secondary">point your agent at it</a>
    </div>
  </div>
</section>


<section class="stat-strip">
  <div class="stat-strip__item">
    <span class="stat-strip__label">pages</span>
    <span class="stat-strip__value">52</span>
  </div>
  <div class="stat-strip__item">
    <span class="stat-strip__label">sources cited</span>
    <span class="stat-strip__value">41</span>
  </div>
  <div class="stat-strip__item">
    <span class="stat-strip__label">last updated</span>
    <span class="stat-strip__value">2026-10-09</span>
  </div>
  <div class="stat-strip__item">
    <span class="stat-strip__label">build</span>
    <span class="stat-strip__value">0.65s</span>
  </div>
</section>


<section class="wiki-section wiki-section--dark dot-grid--dark">
  <div class="wiki-container">
    <p class="wiki-section__eyebrow wiki-section__eyebrow--dark">What this is</p>
    <h2 class="wiki-section__head">A research library for running open-weight LLMs locally</h2>
    <p class="about-section__lead">
      This wiki exists because local AI has too much scattered, conflicting, and unverified
      information. Every page is built from live sources — vendor docs, GitHub repos, academic
      papers, and benchmarks — with uncertainty flags preserved inline (verified, unverified,
      vendor-reported). It is readable by humans and structured for AI agents.
    </p>
    <p class="about-section__lead">
      The content model is deliberately sparse: a question, findings, sources. Six categories
      cover inference engines, open-weight models, hardware, the application stack, practice/ops,
      and an overview. Each page carries front-matter (title, category, tags, summary, source
      count, status) so agents can ingest it programmatically.
    </p>
  </div>
</section>


<section class="wiki-section wiki-section--light">
  <div class="wiki-container">
    <p class="wiki-section__eyebrow">Six categories</p>
    <h2 class="wiki-section__head">Browse by topic</h2>
    <div class="category-grid">
      <a href="/content/inference-engines/overview/" class="category-card">
        <div class="category-card__name">Inference Engines</div>
        <div class="category-card__count">18 pages</div>
        <div class="category-card__desc">llama.cpp, Ollama, vLLM, SGLang, TensorRT-LLM, and 14 more engine profiles — pick by workload.</div>
      </a>
      <a href="/content/models/landscape/" class="category-card">
        <div class="category-card__name">Open-Weight Models</div>
        <div class="category-card__count">3 pages</div>
        <div class="category-card__desc">Landscape, VRAM sizing, licensing — what fits and what you can use.</div>
      </a>
      <a href="/content/hardware/buying-guide/" class="category-card">
        <div class="category-card__name">Hardware</div>
        <div class="category-card__count">9 pages</div>
        <div class="category-card__desc">VRAM math, quantization, multi-GPU, Apple Silicon, AMD ROCm — sizing before buying.</div>
      </a>
      <a href="/content/application-stack/overview/" class="category-card">
        <div class="category-card__name">Application Stack</div>
        <div class="category-card__count">9 pages</div>
        <div class="category-card__desc">RAG, agents, fine-tuning, evaluation, speech, image gen — composability layer.</div>
      </a>
      <a href="/content/practice-ops/learning-resources/" class="category-card">
        <div class="category-card__name">Practice &amp; Ops</div>
        <div class="category-card__count">10 pages</div>
        <div class="category-card__desc">Drivers, monitoring, security, cost analysis, EU AI Act compliance.</div>
      </a>
      <a href="/content/_overview/reading-guide/" class="category-card">
        <div class="category-card__name">Overview</div>
        <div class="category-card__count">2 pages</div>
        <div class="category-card__desc">Reading guide + master report — start here if you don't know where to begin.</div>
      </a>
    </div>
  </div>
</section>


<section class="agent-layer" id="agent-layer">
  <div class="wiki-container">
    <p class="wiki-section__eyebrow wiki-section__eyebrow--dark">For agents</p>
    <h2 class="wiki-section__head wiki-section__head--dark">Machine-readable endpoints</h2>
    <p class="agent-layer__desc">
      Agents <strong>do not</strong> browse HTML. Point your agent at these endpoints to discover the full wiki:
    </p>

    <div class="agent-endpoint">
      <code class="agent-endpoint__path">GET /llms.txt</code>
      <span class="agent-endpoint__desc">Project index with all 52 pages, one-line descriptions (llms.txt v2 spec).</span>
    </div>
    <div class="agent-endpoint">
      <code class="agent-endpoint__path">GET /llms-full.txt</code>
      <span class="agent-endpoint__desc">Concatenated markdown of every page — ingest the entire wiki at once.</span>
    </div>
    <div class="agent-endpoint">
      <code class="agent-endpoint__path">GET /content/{category}/{slug}/</code>
      <span class="agent-endpoint__desc">Any page rendered as HTML. Append <code>.md</code> for raw source.</span>
    </div>
    <div class="agent-endpoint">
      <code class="agent-endpoint__path">GET /index.json</code>
      <span class="agent-endpoint__desc">JSON index: every page with url, raw-markdown URL, title, summary, tags, status.</span>
    </div>
  </div>
</section>


<section class="wiki-section wiki-section--light">
  <div class="wiki-container">
    <p class="wiki-section__eyebrow">Features</p>
    <h2 class="wiki-section__head">Built for AI agents, readable by humans</h2>

    <div class="feature-list">
      <div class="feature-item">
        <div class="feature-item__head">
          <span class="feature-item__icon">llms.txt</span>
          <span class="feature-item__label">Agent discovery</span>
        </div>
        <p class="feature-item__desc">
          Every page indexed in <code>llms.txt</code> and <code>llms-full.txt</code>, following the aug-2026 spec with
          <code>rel=alternate</code> link headers. Agents discover the entire wiki from one entry point.
        </p>
      </div>

      <div class="feature-item">
        <div class="feature-item__head">
          <span class="feature-item__icon">raw</span>
          <span class="feature-item__label">Raw markdown</span>
        </div>
        <p class="feature-item__desc">
          Every page is served byte-identical to its source <code>.md</code> at a predictable URL
          (<code>/content/{category}/{slug}.md</code>), so agents ingest verified source text — no HTML parsing.
        </p>
      </div>

      <div class="feature-item">
        <div class="feature-item__head">
          <span class="feature-item__icon">json</span>
          <span class="feature-item__label">Structured index</span>
        </div>
        <p class="feature-item__desc">
          A JSON index (<code>/index.json</code>) exposes every page's url, raw-markdown url, title, summary,
          category, tags, last-updated, source count, and verification status — the programmatic entry point.
        </p>
      </div>

      <div class="feature-item">
        <div class="feature-item__head">
          <span class="feature-item__icon">sitemap</span>
          <span class="feature-item__label">Sitemap + no-JS</span>
        </div>
        <p class="feature-item__desc">
          <code>sitemap.xml</code> + <code>robots.txt</code> (allow all crawlers). The site builds as static HTML
          with zero runtime JS for content — readable with JavaScript disabled, fast to crawl.
        </p>
      </div>
    </div>
  </div>
</section>


<section class="wiki-section wiki-section--dark dot-grid--dark" id="docs">
  <div class="wiki-container">
    <p class="wiki-section__eyebrow wiki-section__eyebrow--dark">Docs preview</p>
    <h2 class="wiki-section__head">Top pages — verified research, cited sources</h2>
    <p class="about-section__lead">
      Start with any of these curated entries. Each carries inline uncertainty flags
      (verified, vendor-reported, unverified) so agents can weight claims appropriately.
    </p>

    <div class="preview-grid">
      <a href="/content/inference-engines/overview/" class="preview-card">
        <p class="preview-card__eyebrow">inference engines</p>
        <div class="preview-card__title">The 2026 Local Inference Stack</div>
        <p class="preview-card__desc">GGUF/C++ single-user engines, batched GPU serving, and agent-oriented engines. Engine choice is now gated by model architecture.</p>
      </a>
      <a href="/content/hardware/buying-guide/" class="preview-card">
        <p class="preview-card__eyebrow">hardware</p>
        <div class="preview-card__title">GPU Buying Guide for Local LLMs</div>
        <p class="preview-card__desc">VRAM is the binding constraint. RTX 3090 24GB for 32T, RTX 5090 for speed, M5 Ultra for Apple Silicon.</p>
      </a>
      <a href="/content/models/landscape/" class="preview-card">
        <p class="preview-card__eyebrow">models</p>
        <div class="preview-card__title">Open-Weight Model Landscape</div>
        <p class="preview-card__desc">Qwen3.8-27B is the standout dense line. MoE for big, dense for local. License flags preserved per model.</p>
      </a>
      <a href="/content/application-stack/overview/" class="preview-card">
        <p class="preview-card__eyebrow">application stack</p>
        <div class="preview-card__title">The Local AI Application Stack</div>
        <p class="preview-card__desc">Everything speaks OpenAI-compatible APIs. vLLM → LlamaIndex/CrewAI → MCP → Qdrant → Qwen3-Embedding.</p>
      </a>
      <a href="/content/practice-ops/security-local-vs-cloud/" class="preview-card">
        <p class="preview-card__eyebrow">security</p>
        <div class="preview-card__title">Security — Local vs Cloud, Supply Chain</div>
        <p class="preview-card__desc">Local is private by default, not automatically secure. SafeTensors eliminates the pickle RCE vector. Verify SHA-256 before loading.</p>
      </a>
    </div>
  </div>
</section>


<section class="wiki-section wiki-section--light">
  <div class="wiki-container">
    <p class="wiki-section__eyebrow">Latest entries</p>
    <h2 class="wiki-section__head">Recent pages</h2>

    <ul class="entry-list">
      <li class="entry-list__item">
        <a href="/content/application-stack/multi-agent/" class="entry-list__link">
          <span class="entry-list__title">Fully-Local Multi-Agent Orchestration</span>
          <span class="entry-list__meta">2026-10-09 · application-stack</span>
        </a>
      </li>
      <li class="entry-list__item">
        <a href="/content/hardware/vram-math/" class="entry-list__link">
          <span class="entry-list__title">VRAM Math: Sizing Your GPU Before You Buy</span>
          <span class="entry-list__meta">2026-10-09 · hardware</span>
        </a>
      </li>
      <li class="entry-list__item">
        <a href="/content/practice-ops/monitoring/" class="entry-list__link">
          <span class="entry-list__title">Monitoring &amp; Observability — OTel, Prometheus</span>
          <span class="entry-list__meta">2026-10-09 · practice-ops</span>
        </a>
      </li>
      <li class="entry-list__item">
        <a href="/content/inference-engines/emergent-engines/" class="entry-list__link">
          <span class="entry-list__title">Emergent Agent-Oriented Engines (2026)</span>
          <span class="entry-list__meta">2026-10-09 · inference-engines</span>
        </a>
      </li>
    </ul>
  </div>
</section>


<section class="landing-cta">
  <div class="wiki-container">
    <h2 class="landing-cta__head">Point your agent at this wiki</h2>
    <p class="landing-cta__sub">
      Give your agent the <code>llms.txt</code> URL and it can discover, read, and reproduce
      every local AI workflow documented here — no human browsing required.
    </p>
    <a href="/llms.txt" class="landing-cta__button">GET /llms.txt</a>
  </div>
</section>


<footer class="wiki-footer">
  <div class="wiki-container">
    <div class="wiki-footer__grid">
      <div class="wiki-footer__col">
        <div class="wiki-footer__col-title">Local AI Agent Wiki</div>
        <p class="wiki-footer__col-desc">A knowledge wiki for running open-weight LLMs locally, built from verified research with cited sources.</p>
      </div>
      <div class="wiki-footer__col">
        <div class="wiki-footer__col-title">Links</div>
        <ul class="wiki-footer__list">
          <li><a href="/content/_overview/reading-guide/">Wiki</a></li>
          <li><a href="https://github.com/paulo/local-ai-agent-wiki">Source</a></li>
          <li><a href="/content/practice-ops/licensing-open-vs-open-source/">License</a></li>
          <li><a href="/llms.txt">For Agents</a></li>
        </ul>
      </div>
      <div class="wiki-footer__col">
        <div class="wiki-footer__col-title">Categories</div>
        <ul class="wiki-footer__list">
          <li><a href="/content/inference-engines/overview/">Inference Engines</a></li>
          <li><a href="/content/models/landscape/">Models</a></li>
          <li><a href="/content/hardware/buying-guide/">Hardware</a></li>
          <li><a href="/content/application-stack/overview/">Application Stack</a></li>
          <li><a href="/content/practice-ops/learning-resources/">Practice &amp; Ops</a></li>
        </ul>
      </div>
    </div>
    <div class="wiki-footer__legal">
      <p>Content licensed CC-BY-SA 4.0. Fonts: Doto (OFL-1.1), Inter (OFL-1.1).
      Built with MkDocs + Material. This is a local project — not deployed.</p>
    </div>
  </div>
</footer>

          
var target=document.getElementById(location.hash.slice(1));target&amp;&amp;target.name&amp;&amp;(target.checked=target.name.startsWith("__tabbed_"))
        </div>
        
      </main>
      <footer class="md-footer">
  
  <div class="md-footer-meta md-typeset">
    <div class="md-footer-meta__inner md-grid">
      <div class="md-copyright">
  
  
    Made with
    <a href="https://squidfunk.github.io/mkdocs-material/">
      Material for MkDocs
    </a>
  
</div>
      
    </div>
  </div>
</footer>
    </div>
    <div class="md-dialog">
      <div class="md-dialog__inner md-typeset"></div>
    </div>
    
    
    
  
      
      
      {"annotate": null, "base": ".", "features": [], "search": "assets/javascripts/workers/search.2c215733.min.js", "tags": null, "translations": {"clipboard.copied": "Copied to clipboard", "clipboard.copy": "Copy to clipboard", "search.result.more.one": "1 more on this page", "search.result.more.other": "# more on this page", "search.result.none": "No matching documents", "search.result.one": "1 matching document", "search.result.other": "# matching documents", "search.result.placeholder": "Type to start searching", "search.result.term.missing": "Missing", "select.version": "Select version"}, "version": null}
    
  
    :root {
      --md-text-font: 'Inter', ui-sans, system-ui, sans-serif;
      --md-code-font: 'SFMono-Regular', 'JetBrains Mono', 'Fira Code', 'Consolas', 'Menlo', 'ui-monospace', monospace;
    }