Skip to content

armgit / xerj

Public Imported from GitHub
0

AI search engine and MCP server for bots and agents. One Rust binary indexes code, docs, logs, JSONL/CSV and PDFs for BM25 full-text, vector and hybrid search, code search without grep, agent memory and RAG. Elasticsearch-compatible API; Corpus Hub serves hundreds of reference memory corpora.

45.9 KB
Raw Blame
<!DOCTYPE html>
<html lang="en" data-theme="day">
<head>
<!-- ga4:begin — generated by scripts/seo/seo_head.py · do not hand-edit -->
<!-- Google tag (gtag.js) -->
<script async src="https://www.googletagmanager.com/gtag/js?id=G-Z77ETCDJYL"></script>
<script>
  window.dataLayer = window.dataLayer || [];
  function gtag(){dataLayer.push(arguments);}
  gtag('js', new Date());

  gtag('config', 'G-Z77ETCDJYL');
</script>
<!-- ga4:end -->
<meta charset="UTF-8">
<meta name="viewport" content="width=device-width, initial-scale=1.0">
<title>XERJ · Docs · Recipe · Agent memory</title>
<meta name="description" content="Give an AI agent long-term memory with XERJ&#x27;s namespaced offline memory API — store, recall by vector or text, filter by metadata, forget, stay isolated.">
<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=Big+Shoulders+Display:wght@400;700;800;900&family=Inter:wght@400;500;600;700&family=IBM+Plex+Sans:wght@400;500;600;700&family=JetBrains+Mono:wght@400;500;700&display=swap" rel="stylesheet">
<link rel="icon" href="data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg' viewBox='0 0 100 100'><rect width='100' height='100' fill='%230b0b0d'/><text x='50' y='68' font-size='60' font-family='Big Shoulders Display,sans-serif' font-weight='900' fill='%23ffc400' text-anchor='middle'>X</text></svg>">
<script src="/theme.js"></script>
<link rel="stylesheet" href="/style.css?v=90f643e">
<!-- seo:begin — generated by scripts/seo/fix_heads.py · do not hand-edit -->
<link rel="canonical" href="https://xerj.org/docs/recipes/agentic-memory">
<meta property="og:site_name" content="XERJ.ai">
<meta property="og:type" content="article">
<meta property="og:title" content="XERJ · Docs · Recipe · Agent memory">
<meta property="og:description" content="Give an AI agent long-term memory with XERJ&#x27;s namespaced offline memory API — store, recall by vector or text, filter by metadata, forget, stay isolated.">
<meta property="og:url" content="https://xerj.org/docs/recipes/agentic-memory">
<meta property="og:image" content="https://xerj.org/og/xerj-card.png">
<meta property="og:image:width" content="1200">
<meta property="og:image:height" content="630">
<meta property="og:image:alt" content="XERJ.ai — the search engine built for AI agents">
<meta property="og:locale" content="en_US">
<meta name="twitter:card" content="summary_large_image">
<meta name="twitter:title" content="XERJ · Docs · Recipe · Agent memory">
<meta name="twitter:description" content="Give an AI agent long-term memory with XERJ&#x27;s namespaced offline memory API — store, recall by vector or text, filter by metadata, forget, stay isolated.">
<meta name="twitter:image" content="https://xerj.org/og/xerj-card.png">
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "Organization",
  "@id": "https://xerj.org/#organization",
  "name": "XERJ",
  "alternateName": "XERJ.ai",
  "url": "https://xerj.org/",
  "logo": "https://xerj.org/brandbook/xerj-mark.svg",
  "description": "A new way for AI to search for data: one local Rust binary that indexes your code, docs, specs and logs, plus Corpus Hub packs, and returns cited answers.",
  "sameAs": [
    "https://github.com/xerj-org/xerj",
    "https://x.com/xerj_org"
  ]
}
</script>
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "WebSite",
  "@id": "https://xerj.org/#website",
  "name": "XERJ.ai",
  "url": "https://xerj.org/",
  "inLanguage": "en",
  "publisher": {
    "@id": "https://xerj.org/#organization"
  }
}
</script>
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "TechArticle",
  "headline": "XERJ · Docs · Recipe · Agent memory",
  "description": "Give an AI agent long-term memory with XERJ's namespaced offline memory API — store, recall by vector or text, filter by metadata, forget, stay isolated.",
  "url": "https://xerj.org/docs/recipes/agentic-memory",
  "image": "https://xerj.org/og/xerj-card.png",
  "inLanguage": "en",
  "datePublished": "2026-07-08",
  "dateModified": "2026-09-08",
  "author": {
    "@id": "https://xerj.org/#organization"
  },
  "publisher": {
    "@id": "https://xerj.org/#organization"
  },
  "isPartOf": {
    "@id": "https://xerj.org/#website"
  },
  "mainEntityOfPage": {
    "@type": "WebPage",
    "@id": "https://xerj.org/docs/recipes/agentic-memory"
  }
}
</script>
<script type="application/ld+json">
{
  "@context": "https://schema.org",
  "@type": "BreadcrumbList",
  "itemListElement": [
    {
      "@type": "ListItem",
      "position": 1,
      "name": "Home",
      "item": "https://xerj.org/"
    },
    {
      "@type": "ListItem",
      "position": 2,
      "name": "Docs",
      "item": "https://xerj.org/docs/"
    },
    {
      "@type": "ListItem",
      "position": 3,
      "name": "Recipes",
      "item": "https://xerj.org/docs/recipes/"
    },
    {
      "@type": "ListItem",
      "position": 4,
      "name": "Agent memory"
    }
  ]
}
</script>
<!-- seo:end -->
</head>
<body>


<nav class="nav" aria-label="Primary">
  <a href="/" class="brand">XERJ</a>
  <a href="/product">PRODUCT</a>
  <a href="/use-cases">USE CASES</a>
  <a href="/answers/">ANSWERS</a>
  <a href="/benchmarks/">BENCHMARKS</a>
  <a href="/blog/">BLOG</a>
  <span class="spacer"></span>
  <a href="/for-agents">FOR AGENTS</a>
  <a href="/docs/" class="active">DOCS</a>
  <a href="/playground">DASHBOARDS</a>
  <a href="https://hub.xerj.org" rel="external">HUB</a>
  <a class="nav-stars" href="https://github.com/xerj-org/xerj" rel="external" title="XERJ on GitHub" aria-label="XERJ on GitHub, 14,225 stars"><svg viewBox="0 0 16 16" width="14" height="14" aria-hidden="true"><path d="M8 0C3.58 0 0 3.58 0 8c0 3.54 2.29 6.53 5.47 7.59.4.07.55-.17.55-.38 0-.19-.01-.82-.01-1.49-2.01.37-2.53-.49-2.69-.94-.09-.23-.48-.94-.82-1.13-.28-.15-.68-.52-.01-.53.63-.01 1.08.58 1.23.82.72 1.21 1.87.87 2.33.66.07-.52.28-.87.51-1.07-1.78-.2-3.64-.89-3.64-3.95 0-.87.31-1.59.82-2.15-.08-.2-.36-1.02.08-2.12 0 0 .67-.21 2.2.82.64-.18 1.32-.27 2-.27.68 0 1.36.09 2 .27 1.53-1.04 2.2-.82 2.2-.82.44 1.1.16 1.92.08 2.12.51.56.82 1.27.82 2.15 0 3.07-1.87 3.75-3.65 3.95.29.25.54.73.54 1.48 0 1.07-.01 1.93-.01 2.2 0 .21.15.46.55.38A8.013 8.013 0 0016 8c0-4.42-3.58-8-8-8z"/></svg><span class="nav-stars-n" data-gh-stars="14225">★ 14.2k</span></a>
  <a href="https://x.com/xerj_org" rel="external" title="XERJ on X" style="display:inline-flex;align-items:center"><svg width="15" height="15" viewBox="0 0 24 24" fill="currentColor" aria-hidden="true"><path d="M18.244 2.25h3.308l-7.227 8.26 8.502 11.24H16.17l-5.214-6.817L4.99 21.75H1.68l7.73-8.835L1.254 2.25H8.08l4.713 6.231zm-1.161 17.52h1.833L7.084 4.126H5.117z"/></svg></a>
  <a href="https://discord.com/invite/yetXFusXfs" rel="external" title="XERJ on Discord" style="display:inline-flex;align-items:center"><svg width="15" height="15" viewBox="0 0 24 24" fill="currentColor" aria-hidden="true"><path d="M20.317 4.37a19.79 19.79 0 00-4.885-1.515.074.074 0 00-.079.037c-.21.375-.444.864-.608 1.25a18.27 18.27 0 00-5.487 0 12.64 12.64 0 00-.617-1.25.077.077 0 00-.079-.037A19.74 19.74 0 003.677 4.37a.07.07 0 00-.032.027C.533 9.046-.32 13.58.099 18.057a.082.082 0 00.031.057 19.9 19.9 0 005.993 3.03.078.078 0 00.084-.028c.462-.63.874-1.295 1.226-1.994a.076.076 0 00-.041-.106 13.1 13.1 0 01-1.872-.892.077.077 0 01-.008-.128 10.2 10.2 0 00.372-.292.074.074 0 01.077-.01c3.928 1.793 8.18 1.793 12.062 0a.074.074 0 01.078.01c.12.098.246.198.373.292a.077.077 0 01-.006.127 12.3 12.3 0 01-1.873.892.077.077 0 00-.041.107c.36.698.772 1.362 1.225 1.993a.076.076 0 00.084.028 19.84 19.84 0 006.002-3.03.077.077 0 00.032-.054c.5-5.177-.838-9.674-3.549-13.66a.061.061 0 00-.031-.03zM8.02 15.33c-1.183 0-2.157-1.085-2.157-2.419 0-1.333.956-2.419 2.157-2.419 1.21 0 2.176 1.096 2.157 2.42 0 1.333-.956 2.418-2.157 2.418zm7.975 0c-1.183 0-2.157-1.085-2.157-2.419 0-1.333.955-2.419 2.157-2.419 1.21 0 2.176 1.096 2.157 2.42 0 1.333-.946 2.418-2.157 2.418z"/></svg></a>
  <span class="theme">
    <button type="button" data-theme="day" class="active">DAY</button>
    <span class="dash">·</span>
    <button type="button" data-theme="night">NIGHT</button>
  </span>
</nav>

<div class="docs-search" id="docsSearch">
  <input type="text" id="docsSearchInput" placeholder="Search docs · keys, flags, endpoints, playbooks…" aria-label="Search docs" autocomplete="off" spellcheck="false">
  <span class="hint">⌘ K</span>
  <div class="results" id="docsSearchResults" aria-live="polite"></div>
</div>

<div class="docs-shell">
<nav class="docs-nav" aria-label="Docs">  <div class="group">
    <span class="label">START</span>
    <a href="/docs/" class=""><span class="eyebrow">01</span>Overview</a>
      <a href="/docs/quickstart" class=""><span class="eyebrow">02</span>Quickstart</a>
      <a href="/docs/install" class=""><span class="eyebrow">03</span>Install</a>
      <a href="/docs/migration-from-es" class=""><span class="eyebrow">04</span>Migrate from ES</a>
  </div>
  <div class="group">
    <span class="label">REFERENCE</span>
    <a href="/docs/cli" class=""><span class="eyebrow">01</span>CLI</a>
      <a href="/docs/config" class=""><span class="eyebrow">02</span>Config TOML</a>
      <a href="/docs/env" class=""><span class="eyebrow">03</span>Environment</a>
      <a href="/docs/api-native" class=""><span class="eyebrow">04</span>Native REST API</a>
      <a href="/docs/api-es-compat" class=""><span class="eyebrow">05</span>ES-compat API</a>
  </div>
  <div class="group">
    <span class="label">DATA MODEL</span>
    <a href="/docs/queries" class=""><span class="eyebrow">01</span>Query types</a>
      <a href="/docs/aggregations" class=""><span class="eyebrow">02</span>Aggregations</a>
      <a href="/docs/analyzers" class=""><span class="eyebrow">03</span>Analyzers</a>
      <a href="/docs/vectors" class=""><span class="eyebrow">04</span>Vectors & kNN</a>
  </div>
  <div class="group">
    <span class="label">ENGINE</span>
    <a href="/docs/ingest" class=""><span class="eyebrow">01</span>Ingest pipelines</a>
      <a href="/docs/storage" class=""><span class="eyebrow">02</span>Storage & WAL</a>
      <a href="/docs/compression" class=""><span class="eyebrow">03</span>Compression</a>
      <a href="/docs/clustering" class=""><span class="eyebrow">04</span>Clustering</a>
      <a href="/docs/metrics" class=""><span class="eyebrow">05</span>Metrics</a>
      <a href="/docs/security" class=""><span class="eyebrow">06</span>Auth & TLS</a>
  </div>
  <div class="group">
    <span class="label">OPERATIONS</span>
    <a href="/docs/operations" class=""><span class="eyebrow">01</span>Running in production</a>
      <a href="/docs/backup-restore" class=""><span class="eyebrow">02</span>Backup & restore</a>
      <a href="/docs/upgrades" class=""><span class="eyebrow">03</span>Upgrades</a>
      <a href="/docs/troubleshooting" class=""><span class="eyebrow">04</span>Troubleshooting</a>
  </div>
  <div class="group">
    <span class="label">PLAYBOOKS</span>
    <a href="/docs/playbooks/siem" class=""><span class="eyebrow">01</span>SIEM</a>
      <a href="/docs/playbooks/log-analytics" class=""><span class="eyebrow">02</span>Log analytics</a>
      <a href="/docs/playbooks/vector-search" class=""><span class="eyebrow">03</span>Vector · RAG</a>
      <a href="/docs/playbooks/full-text" class=""><span class="eyebrow">04</span>Full-text search</a>
      <a href="/docs/playbooks/observability" class=""><span class="eyebrow">05</span>Observability</a>
  </div>
  <div class="group">
    <span class="label">RECIPES</span>
    <a href="/docs/recipes/" class=""><span class="eyebrow">00</span>All recipes</a>
    <a href="/docs/recipes/semantic-search-rag" class=""><span class="eyebrow">01</span>Semantic search &amp; RAG</a>
    <a href="/docs/recipes/agentic-memory" class="active"><span class="eyebrow">02</span>Agent memory</a>
    <a href="/docs/recipes/passage-retrieval" class=""><span class="eyebrow">03</span>Passage retrieval</a>
    <a href="/docs/recipes/vector-search-knn" class=""><span class="eyebrow">04</span>Vector search &middot; kNN</a>
    <a href="/docs/recipes/vector-quantization" class=""><span class="eyebrow">05</span>Vector quantization</a>
    <a href="/docs/recipes/hybrid-search" class=""><span class="eyebrow">06</span>Hybrid search</a>
    <a href="/docs/recipes/log-analytics" class=""><span class="eyebrow">07</span>Log analytics</a>
    <a href="/docs/recipes/anomaly-detection" class=""><span class="eyebrow">08</span>Anomaly detection</a>
    <a href="/docs/recipes/continuous-anomaly-datafeeds" class=""><span class="eyebrow">09</span>Anomaly datafeeds</a>
    <a href="/docs/recipes/migrate-from-elasticsearch" class=""><span class="eyebrow">10</span>Migrate from ES</a>
  </div>
</nav>

<main class="docs-content">
  <!-- Print-only branded header (hidden on screen, visible in PDF) -->
  <div class="print-header" style="display:none;">XERJ</div>

  <div class="breadcrumb"><a href="/docs/">DOCS</a><span class="sep">·</span><a href="/docs/recipes/">RECIPES</a><span class="sep">·</span>AGENT MEMORY
    <button class="pdf-btn" onclick="window.print()" title="Save this page as PDF">
      <svg viewBox="0 0 24 24"><path d="M12 3v12M12 15l-4-4M12 15l4-4"/><path d="M4 17v2a2 2 0 002 2h12a2 2 0 002-2v-2"/></svg>
      PDF
    </button>
  </div>
  <div class="kicker"><span>RECIPE · 02</span></div>
  <h1>Agent memory</h1>


  <p><strong>Use case:</strong> your agent learns things during a session — user preferences, project facts, decisions — and needs to <em>remember</em> them on a later turn to answer well. The usual answer is to bolt a separate vector database next to your search stack. XERJ folds that job into the engine you already run: a namespaced, offline <strong>agent-memory API</strong> (<code class="inline">/_memory/{ns}</code>) that stores text + an embedding + metadata and recalls the most relevant memories by <strong>vector similarity</strong> or <strong>plain text</strong>, with metadata filtering, forgetting, and hard per-agent isolation.</p>

  <p>This recipe builds a tiny simulated coding-assistant agent (no LLM required) that:</p>
  <ol>
    <li><strong>stores</strong> what it learns as memories (<code class="inline">POST /_memory/{ns}</code>),</li>
    <li><strong>recalls</strong> the right memory to answer a new question — semantically (kNN) and lexically (BM25),</li>
    <li>does a <strong>metadata-filtered</strong> recall ("only surface team <em>preferences</em>"),</li>
    <li><strong>forgets</strong> a memory that is no longer true, and</li>
    <li>proves <strong>multi-namespace isolation</strong>: a second agent can never see the first agent's memories.</li>
  </ol>

  <p>Everything below was run end-to-end against a live XERJ. No pip installs — the example is Python 3 stdlib only.</p>

  <h2>Why this replaces a bolt-on vector DB</h2>
  <p>A memory namespace is just a reserved XERJ index (<code class="inline">.xerj-memory-{ns}</code>) under the hood, so recall reuses the same exact <code class="inline">dense_vector</code> kNN, BM25, and metadata-filter paths that serve the rest of the engine. That means:</p>
  <ul>
    <li><strong>One system to run and back up.</strong> No second datastore, no sync job keeping a vector index consistent with your source of truth.</li>
    <li><strong>kNN <em>and</em> BM25 in the same store.</strong> Recall by embedding similarity, by keyword, or filter both by metadata — no glue code.</li>
    <li><strong>Namespaces are physical isolation.</strong> Each agent (or tenant, or user) gets its own backing index; a recall in namespace <code class="inline">A</code> literally cannot read namespace <code class="inline">B</code>.</li>
    <li><strong>Offline and zero-config.</strong> You supply the vectors (from your own model); XERJ never phones out to embed.</li>
  </ul>

  <h2>The memory API, in four calls</h2>
  <table>
    <thead>
      <tr><th>Verb + path</th><th>What it does</th><th>Body / result</th></tr>
    </thead>
    <tbody>
      <tr><td><code class="inline">POST /_memory/{ns}</code></td><td>store a memory</td><td><code class="inline">{text, vector?, metadata?, id?}</code> → <code class="inline">{id, namespace, created}</code></td></tr>
      <tr><td><code class="inline">POST /_memory/{ns}/_recall</code></td><td>recall top-k</td><td><code class="inline">{vector? | query?, k?, filter?}</code> → <code class="inline">{hits:[{id,text,metadata,score}]}</code></td></tr>
      <tr><td><code class="inline">GET /_memory/{ns}</code></td><td>list recent (bounded)</td><td>→ <code class="inline">{count, entries:[…]}</code></td></tr>
      <tr><td><code class="inline">DELETE /_memory/{ns}/{id}</code></td><td>forget one</td><td>→ <code class="inline">{id, forgotten}</code></td></tr>
      <tr><td><code class="inline">DELETE /_memory/{ns}</code></td><td>drop the namespace</td><td>→ <code class="inline">{namespace, dropped}</code></td></tr>
    </tbody>
  </table>
  <p><code class="inline">_recall</code> picks the path from the body: a <code class="inline">vector</code> runs kNN; otherwise <code class="inline">query</code> runs BM25 over the text; an empty body returns recent memories (<code class="inline">match_all</code>). <code class="inline">filter</code> is a normal ES query clause applied as a <code class="inline">bool</code> filter, so it narrows without affecting the score. Recall of an unknown namespace returns <code class="inline">{"hits": []}</code> — that is how isolation stays clean.</p>

  <h3>Storing a memory (raw wire)</h3>
<pre class="code">curl -XPOST localhost:9200/_memory/agent-ada -H 'content-type: application/json' -d '{
  "text": "The team prefers Rust over Go for new backend services.",
  "vector": [0.12, 0.0, 0.34],
  "metadata": {"kind": "preference", "topic": "language"},
  "id": "demo-1"
}'</pre>
<pre class="code">{"created":true,"id":"demo-1","namespace":"agent-ada"}</pre>

  <h3>Recalling by vector, filtered to preferences (raw wire)</h3>
<pre class="code">curl -XPOST localhost:9200/_memory/agent-ada/_recall -H 'content-type: application/json' -d '{
  "vector": [0.12, 0.0, 0.34],
  "k": 2,
  "filter": {"term": {"metadata.kind": "preference"}}
}'</pre>
<pre class="code">{"hits":[{"id":"demo-1",
          "metadata":{"kind":"preference","topic":"language"},
          "score":1.0,
          "text":"The team prefers Rust over Go for new backend services."}],
 "namespace":"agent-ada"}</pre>

  <p>Note <code class="inline">metadata.kind</code> filters straight out of the box — nested metadata subfields are queryable with a normal <code class="inline">term</code> clause, no explicit mapping needed.</p>

  <h2>The full agent</h2>
  <p>The script below is the whole example. The one thing to understand about it: it carries a <strong>tiny self-contained "embedder"</strong> (deterministic hashed character trigrams) so the demo needs no model download and cosine similarity is meaningful.</p>
  <p><strong>Honesty check.</strong> That toy embedder is <em>lexical</em>, not neural: it rewards shared character n-grams (so <code class="inline">service</code>/<code class="inline">services</code>, <code class="inline">prefer</code>/<code class="inline">prefers</code> land near each other), but it does not know that "car" ≈ "automobile". For real semantic recall you pass vectors from your own model (OpenAI/Cohere/a local model) — the store and recall calls are byte-for-byte the same. XERJ's own built-in <code class="inline">semantic_text</code> embedder is likewise lexical unless you configure an external <code class="inline">/v1/embeddings</code> endpoint; the memory API deliberately takes <em>your</em> vectors so quality is your choice.</p>
  <p>The full example script is on GitHub — <a href="https://github.com/xerj-org/xerj/blob/main/docs/examples/agentic-memory/agent_memory.py">docs/examples/agentic-memory/agent_memory.py</a> (Python 3 stdlib only, no dependencies).</p>

  <h2>Reproduce it yourself</h2>
  <p>Start XERJ (single node, no TLS/auth) on its default port and run the script. Nothing to install — the client is Python 3 <strong>stdlib only</strong>.</p>
<pre class="code"># 1. boot XERJ on the default port 9200
./engine/target/release/xerj --insecure --data-dir /tmp/xerj-mem &amp;

# 2. run the agent (defaults to http://localhost:9200)
python3 docs/examples/agentic-memory/agent_memory.py

# Point it at a different node/port with XERJ_URL if you like:
#   XERJ_URL=http://localhost:9481 python3 docs/examples/agentic-memory/agent_memory.py</pre>

  <p><strong>What a customer should see.</strong> The client exits <code class="inline">0</code> and prints <code class="inline">All assertions passed.</code>. The recall-quality line is the headline number, and it is <strong>deterministic</strong> — the toy embedder and BM25 are fixed functions of this corpus, so the values below reproduce exactly, run to run:</p>
  <ul>
    <li><strong>Semantic (kNN):</strong> recall@1 = <strong>83.3% (5/6)</strong>, recall@3 = <strong>100% (6/6)</strong></li>
    <li><strong>Lexical (BM25):</strong> recall@1 = <strong>100% (6/6)</strong>, recall@3 = <strong>100% (6/6)</strong></li>
  </ul>
  <p>The single semantic recall@1 miss is honest and instructive: on <em>"Which region is production deployed in?"</em> the toy trigram embedder scores <em>"<strong>Deploys</strong> are gated on CI…"</em> (0.790) a hair above the correct <em>"<strong>Production</strong> runs on AWS…"</em> (0.780) — shared <code class="inline">deploy</code>/<code class="inline">Deploys</code> character trigrams. The right memory is still in the top 3, and <strong>BM25 recalls it first (6/6)</strong>, which is exactly why the recipe offers both recall paths.</p>

  <h3>Real output</h3>
<pre class="code">== Storing what the agent learned this session ==
  stored ada-1    created=True
  stored ada-2    created=True
  stored ada-3    created=True
  stored ada-4    created=True
  stored ada-5    created=True
  stored ada-6    created=True

== Semantic recall for: "What programming language should I use for a new backend service?"
  [0.779] (preference) The team prefers Rust over Go for new backend services.
  [0.741] (fact      ) The payments service owns the Postgres 'ledger' database.
  [0.731] (preference) The team prefers tabs over spaces and a 100-column line limit.

== Lexical (BM25) recall for: "Which region is production deployed in?"
  [3.746] (fact      ) Production runs on AWS in us-east-1; staging is us-west-2.
  [1.010] (fact      ) Deploys are gated on the full integration suite passing in CI.

== Preference-only recall for: "How should I set up the new service?"
  [0.765] (preference) The team prefers Rust over Go for new backend services.
  [0.692] (preference) The team prefers tabs over spaces and a 100-column line limit.
  [0.666] (preference) The team prefers OpenTelemetry for tracing, not vendor SDKs.

== Recall quality over 6 labeled probe questions ==
  semantic (kNN):  recall@1 =  83.3%  (5/6)   recall@3 = 100.0%  (6/6)
  lexical  (BM25): recall@1 = 100.0%  (6/6)   recall@3 = 100.0%  (6/6)

== Namespace isolation ==
  agent-billing recall for a coding question -&gt; ['Invoices over $10k require CFO approval.']
  OK: agent-billing only ever sees its own namespace.

== Forgetting a memory that is no longer true ==

  before forget:
  [0.755] (preference) The team prefers OpenTelemetry for tracing, not vendor SDKs.
  [0.692] (fact      ) Production runs on AWS in us-east-1; staging is us-west-2.
  [0.683] (preference) The team prefers Rust over Go for new backend services.
  forget -&gt; {'forgotten': True, 'id': 'ada-6', 'namespace': 'agent-ada'}

  after forget:
  [0.692] (fact      ) Production runs on AWS in us-east-1; staging is us-west-2.
  [0.683] (preference) The team prefers Rust over Go for new backend services.
  [0.661] (preference) The team prefers tabs over spaces and a 100-column line limit.
  OK: the retracted memory is gone.

All assertions passed.</pre>

  <p>Read the run top to bottom:</p>
  <ul>
    <li><strong>Semantic recall</strong> put the Rust preference on top for a question that never says "Rust" or "prefer" — trigram overlap on <em>backend / service / new</em> pulled it up over unrelated facts.</li>
    <li><strong>BM25 recall</strong> answered the region question exactly, no vectors involved.</li>
    <li><strong>Preference-only recall</strong> returned three memories, <em>all</em> <code class="inline">kind:"preference"</code> — the <code class="inline">metadata.kind</code> filter did its job.</li>
    <li><strong>Recall quality</strong> over the 6 labeled probes is measured, not asserted by hand: BM25 recall@1 = 6/6, semantic recall@1 = 5/6, both recall@3 = 6/6 — a real number this run computed and printed.</li>
    <li><strong>Isolation</strong> held: <code class="inline">agent-billing</code>, asked a coding question, only saw its own single finance memory.</li>
    <li><strong>Forgetting</strong> removed <code class="inline">ada-6</code> (the tracing preference); the follow-up recall no longer surfaces it.</li>
  </ul>

  <h2>Notes, limits, and going to production</h2>
  <ul>
    <li><strong>Bring your own embeddings.</strong> The memory API stores whatever vector you pass and kNN-searches it with cosine similarity. Swap the toy <code class="inline">embed()</code> for a call to your model's embeddings endpoint and everything else is unchanged. Keep the vector dimension consistent within a namespace — the backing index's <code class="inline">dense_vector</code> field is sized from the first stored vector.</li>
    <li><strong>Vector <em>or</em> text.</strong> A memory needs at least <code class="inline">text</code> or a <code class="inline">vector</code>. Text is always BM25-indexed; a vector additionally enables kNN recall. Recall with <code class="inline">vector</code> uses kNN; without it, <code class="inline">query</code> uses BM25.</li>
    <li><strong>Recency (opt-in).</strong> Recall is pure relevance by default. Pass <code class="inline">recency_weight</code> in <code class="inline">[0, 1]</code> on <code class="inline">_recall</code> to blend recency: the engine over-fetches candidates and re-ranks by <code class="inline">(1 - w) * norm_relevance + w * norm_recency</code> (both min-max normalized across the candidate set), so <code class="inline">w = 1</code> surfaces the newest memory and <code class="inline">w = 0</code> is exactly the default relevance order. Verified: two same-vector memories tie on relevance; with <code class="inline">recency_weight: 1.0</code> the newer one is returned first. Useful when an agent's latest observation should win ties.</li>
    <li><strong>Isolation is structural, not a filter.</strong> Namespaces are separate physical indices, so there is no query you can write in one namespace that reads another. Use one namespace per agent/tenant/user.</li>
    <li><strong><code class="inline">GET /_memory/{ns}</code></strong> lists up to 100 most-recent entries for debugging/audit; it is bounded and recent-first, not a full export.</li>
    <li><strong>Wire-compatible.</strong> Because a namespace is a real index, you can inspect or operate on it with the standard ES-compatible surface if you ever need to — the memory API is a thin, honest adapter over the same machinery.</li>
  </ul>

  <h3>Related reading</h3>
  <ul>
    <li><a href="/answers/what-is-an-mcp-memory-server">What is an MCP memory server?</a></li>
    <li><a href="/answers/agentic-memory-vs-rag">Agentic memory compared with RAG</a></li>
    <li><a href="/answers/claude-code-memory">How Claude Code memory works</a></li>
    <li><a href="/compare/xerj-vs-agent-memory-tools">XERJ compared with Mem0, Zep, Letta and Cognee</a></li>
    <li><a href="/compare/xerj-vs-mem0">XERJ compared with Mem0</a></li>
  </ul>

  <h3>Before you trust a recall number</h3>
  <p>The recall@1 and recall@3 above come from six hand-labelled probes. That is a smoke test. Before you use a recall number to choose an embedder, a chunking scheme or a recall path for your own agent, apply these field lessons. They were contributed by Chunxiao Wang (Assay / <a href="https://github.com/chunxiaoxx/nautilus-compass">nautilus-compass</a>) in <a href="https://github.com/xerj-org/xerj/issues/1138">#1138</a>, from a production agent-memory system, and are written up with their measurements in <a href="/blog/eval-scars-from-agent-memory">Six eval scars from a production agent-memory system</a>.</p>
  <ul>
    <li><strong>Probe with situations, not questions.</strong> An agent recalls from memory using its current state (<em>"deploy failed again with 401"</em>), not an information need (<em>"how do 401s work?"</em>). This recipe's probes are questions. Add probes taken from your own agent's transcripts.</li>
    <li><strong>Score on your own memories.</strong> In their runs, slicing memories by topic instead of by token window moved recall@1 by double digits, a bigger effect than moving from dense-only to hybrid retrieval. Their own data also ranked two embedding models in the opposite order to their MTEB scores. Freeze a sample of your own memories and probes, and compare candidates on that sample.</li>
    <li><strong>If every probe scores zero, check the harness first.</strong> A masking bug that drops the gold item from the candidate pool turns an eval into a constant zero, or a constant pass. In this recipe, the equivalent check is that each probe's expected id was stored in the namespace you recall from.</li>
    <li><strong>Small n needs a baseline and an interval.</strong> With six probes, 5/6 against 6/6 is a single probe. Report counts next to percentages. Compute what a trivial strategy scores, for example <code class="inline">recency_weight: 1.0</code> (always the newest memory). Below about 100 probes, give a confidence interval.</li>
    <li><strong>Compare recall paths on the same probes.</strong> kNN against BM25 on one probe set is a paired comparison. Count the probes where the two disagree (McNemar) instead of comparing two percentages.</li>
    <li><strong>Freeze the criteria before the run.</strong> Write down the metric, the threshold and the artifact that will prove each claim before you run anything. Criteria may tighten mid-run but must not loosen.</li>
  </ul>
  <p>The same team donated a corpus of real multi-session agent trajectories, with the failure turns kept, to the corpus hub as <a href="https://github.com/xerj-org/xerj/releases/tag/pack-agent-session-trajectories-2026-10-09"><code class="inline">agent-session-trajectories</code></a>. It is a starting point for testing recall on agent history instead of reference docs.</p>
</main>
</div>

<!-- seo:footer-begin — generated by scripts/seo/fix_links.py · do not hand-edit -->
<nav class="footer-map" aria-label="Site map">
  <div class="fm-col">
    <span class="fm-h">PRODUCT</span>
    <a href="/product">PRODUCT</a>
    <a href="/architecture/">ARCHITECTURE</a>
    <a href="/benchmarks/">BENCHMARKS</a>
    <a href="/pricing/">PRICING</a>
    <a href="/security/">SECURITY</a>
    <a href="/playground">DASHBOARDS</a>
    <a href="/demo/">REAL-DATA DEMO</a>
    <a href="/aise-demo">PRODUCT TOUR</a>
  </div>
  <div class="fm-col">
    <span class="fm-h">LEARN</span>
    <a href="/answers/">ANSWERS</a>
    <a href="/compare/">COMPARE</a>
    <a href="/use-cases">USE CASES</a>
    <a href="/case-studies">CASE STUDIES</a>
    <a href="/industries/">INDUSTRIES</a>
    <a href="/blog/">BLOG</a>
  </div>
  <div class="fm-col">
    <span class="fm-h">SOLUTIONS</span>
    <a href="/solutions/">SOLUTIONS</a>
    <a href="/industries/finserv">FINANCIAL SERVICES</a>
    <a href="/industries/public-sector">PUBLIC SECTOR</a>
    <a href="/industries/healthcare">HEALTHCARE</a>
    <a href="/industries/retail">RETAIL</a>
    <a href="/use-cases/code-security-audit">AI CODE-SECURITY AUDIT</a>
    <a href="/use-cases/semantic-analytics">SEMANTIC ANALYTICS</a>
    <a href="/resources/">SALES KIT</a>
  </div>
  <div class="fm-col">
    <span class="fm-h">DOCS</span>
    <a href="/docs/">DOCUMENTATION</a>
    <a href="/docs/quickstart">QUICKSTART</a>
    <a href="/docs/install">INSTALL</a>
    <a href="/docs/migration-from-es">MIGRATE FROM ELASTICSEARCH</a>
    <a href="/docs/recipes/">RECIPES</a>
    <a href="/docs/agents/">DOCS FOR AI</a>
    <a href="/for-agents">FOR AI AGENTS</a>
    <a href="/agent-search/">AGENTIC SEARCH DEMO</a>
    <a href="/llms.txt">LLMS.TXT</a>
    <a href="https://github.com/xerj-org/xerj" rel="external">GITHUB</a>
  </div>
</nav>
<!-- seo:footer-end -->
<footer class="footer">
  <span>XERJ.AI · DOCS · V1.0.0-RC.100</span>
  <span>READ THE SOURCE</span>
  <span><a href="/">HOME</a> · <a href="/product">PRODUCT</a> · <a href="/answers/">ANSWERS</a> · <a href="/playground">DASHBOARDS</a> &middot; <a href="/llms.txt">LLMS.TXT</a></span>
</footer>

<script type="application/json" id="docs-index">[{"title":"Overview","slug":"/docs/","kicker":"START","snip":""},{"title":"Quickstart","slug":"/docs/quickstart","kicker":"START","snip":""},{"title":"Install","slug":"/docs/install","kicker":"START","snip":""},{"title":"Migrate from ES","slug":"/docs/migration-from-es","kicker":"START","snip":""},{"title":"CLI","slug":"/docs/cli","kicker":"REFERENCE","snip":""},{"title":"Config TOML","slug":"/docs/config","kicker":"REFERENCE","snip":""},{"title":"Environment","slug":"/docs/env","kicker":"REFERENCE","snip":""},{"title":"Native REST API","slug":"/docs/api-native","kicker":"REFERENCE","snip":""},{"title":"ES-compat API","slug":"/docs/api-es-compat","kicker":"REFERENCE","snip":""},{"title":"Query types","slug":"/docs/queries","kicker":"DATA MODEL","snip":""},{"title":"Aggregations","slug":"/docs/aggregations","kicker":"DATA MODEL","snip":""},{"title":"Analyzers","slug":"/docs/analyzers","kicker":"DATA MODEL","snip":""},{"title":"Vectors & kNN","slug":"/docs/vectors","kicker":"DATA MODEL","snip":""},{"title":"Ingest pipelines","slug":"/docs/ingest","kicker":"ENGINE","snip":""},{"title":"Storage & WAL","slug":"/docs/storage","kicker":"ENGINE","snip":""},{"title":"Compression","slug":"/docs/compression","kicker":"ENGINE","snip":""},{"title":"Clustering","slug":"/docs/clustering","kicker":"ENGINE","snip":""},{"title":"Metrics","slug":"/docs/metrics","kicker":"ENGINE","snip":""},{"title":"Auth & TLS","slug":"/docs/security","kicker":"ENGINE","snip":""},{"title":"Running in production","slug":"/docs/operations","kicker":"OPERATIONS","snip":""},{"title":"Backup & restore","slug":"/docs/backup-restore","kicker":"OPERATIONS","snip":""},{"title":"Upgrades","slug":"/docs/upgrades","kicker":"OPERATIONS","snip":""},{"title":"Troubleshooting","slug":"/docs/troubleshooting","kicker":"OPERATIONS","snip":""},{"title":"SIEM","slug":"/docs/playbooks/siem","kicker":"PLAYBOOKS","snip":""},{"title":"Log analytics","slug":"/docs/playbooks/log-analytics","kicker":"PLAYBOOKS","snip":""},{"title":"Vector · RAG","slug":"/docs/playbooks/vector-search","kicker":"PLAYBOOKS","snip":""},{"title":"Full-text search","slug":"/docs/playbooks/full-text","kicker":"PLAYBOOKS","snip":""},{"title":"Observability","slug":"/docs/playbooks/observability","kicker":"PLAYBOOKS","snip":""},{"title":"CLI · --config","slug":"/docs/cli#config-flag","kicker":"CLI","snip":"-c, --config <PATH>  Path to the TOML config file."},{"title":"CLI · --data-dir","slug":"/docs/cli#data-dir-flag","kicker":"CLI","snip":"-d, --data-dir <PATH>  Override [server] data_dir at the command line."},{"title":"CLI · --insecure","slug":"/docs/cli#insecure-flag","kicker":"CLI","snip":"-k, --insecure  Disable TLS and API-key auth. Dev only."},{"title":"[server] rest_port","slug":"/docs/config#server","kicker":"CONFIG","snip":"u16 · default 8080 · Native REST API listener port."},{"title":"[server] es_compat_port","slug":"/docs/config#server","kicker":"CONFIG","snip":"u16 · default 9200 · Elasticsearch-wire-format port. 0 = OS-assigned port."},{"title":"[storage] wal_sync","slug":"/docs/config#storage","kicker":"CONFIG","snip":"enum · default batched · sync · batched · async. Durability knob."},{"title":"[vector] hnsw_ef_search","slug":"/docs/config#vector","kicker":"CONFIG","snip":"u32 · default 100 · Accepted but not wired — the beam width comes from the request num_candidates."},{"title":"[vector] default_quantization","slug":"/docs/config#vector","kicker":"CONFIG","snip":"enum · none (default) · scalar8. binary is refused at startup; scalar4 is not accepted."},{"title":"GET /v1/health","slug":"/docs/api-native#health","kicker":"NATIVE API","snip":"Liveness + readiness. {status, uptime_s, version}."},{"title":"GET /v1/metrics","slug":"/docs/api-native#metrics","kicker":"NATIVE API","snip":"Prometheus text format. Scrape-friendly."},{"title":"POST /v1/indices/:name/turbo-ingest","slug":"/docs/api-native#turbo-ingest","kicker":"NATIVE API","snip":"NDJSON bulk path with parallel tokenization."},{"title":"POST /v1/indices/:name/search","slug":"/docs/api-native#search","kicker":"NATIVE API","snip":"Unified query endpoint — full-text, term, range, KNN, hybrid."},{"title":"POST /:index/_bulk","slug":"/docs/api-es-compat#bulk","kicker":"ES API","snip":"NDJSON bulk. Actions: index, create, update, delete."},{"title":"POST /:index/_search","slug":"/docs/api-es-compat#search","kicker":"ES API","snip":"ES query DSL. Drop-in migration path."},{"title":"match_all","slug":"/docs/queries#match-all","kicker":"QUERY","snip":"Everything in the index."},{"title":"bool","slug":"/docs/queries#bool","kicker":"QUERY","snip":"must / should / must_not / filter clauses."},{"title":"knn","slug":"/docs/queries#knn","kicker":"QUERY","snip":"Vector kNN — HNSW-served when unfiltered (exact-rescored); filtered runs the exact scan."},{"title":"semantic","slug":"/docs/queries#semantic","kicker":"QUERY","snip":"Embed the query text at query time and search the vector field."},{"title":"hybrid","slug":"/docs/queries#hybrid","kicker":"QUERY","snip":"RRF / linear fusion of sub-queries."},{"title":"terms aggregation","slug":"/docs/aggregations#terms","kicker":"AGG","snip":"Exact, pre-computed terms histograms over keyword fields."},{"title":"date_histogram","slug":"/docs/aggregations#date-histogram","kicker":"AGG","snip":"Time-bucketed counts."},{"title":"kNN serving","slug":"/docs/vectors#knn-query","kicker":"VECTORS","snip":"Unfiltered kNN via persisted HNSW + exact rescore; filtered kNN exact scan."},{"title":"SQ8 quantization","slug":"/docs/vectors#quantization","kicker":"VECTORS","snip":"8-bit scalar quantization. int8 scoring precision, not a memory saving (#392)."},{"title":"turbo-ingest","slug":"/docs/ingest#turbo","kicker":"INGEST","snip":"NDJSON bulk path. Parallel tokenization across cores."},{"title":"delta-of-delta","slug":"/docs/compression#delta-of-delta","kicker":"COMPRESSION","snip":"Timestamp encoding. Apache/Nginx/ISO auto-detect."},{"title":"dictionary encoding","slug":"/docs/compression#dictionary","kicker":"COMPRESSION","snip":"Up to 256 unique strings per column."},{"title":"api_key","slug":"/docs/security#api-key","kicker":"SECURITY","snip":"Static admin key. Auto-generated on first run."},{"title":"SIEM playbook","slug":"/docs/playbooks/siem","kicker":"PLAYBOOK","snip":"Terms agg over source IPs for SIEM, with detection rules."},{"title":"RAG playbook","slug":"/docs/playbooks/vector-search","kicker":"PLAYBOOK","snip":"Embeddings, kNN, hybrid search, BM25 + kNN fusion."},{"title":"[cluster] enabled","slug":"/docs/config#cluster","kicker":"CONFIG","snip":"bool · default false · Enable multi-node Raft cluster mode."},{"title":"[cluster] peers","slug":"/docs/config#cluster","kicker":"CONFIG","snip":"array · \"host:port=host:port\" · label must equal address (#1171)."},{"title":"[cluster] port","slug":"/docs/config#cluster","kicker":"CONFIG","snip":"u16 · default 9300 · Intra-cluster Raft + search port."},{"title":"[embedding] default_endpoint","slug":"/docs/config#embedding","kicker":"CONFIG","snip":"string · OpenAI-compatible embeddings URL for auto-embedding on ingest."},{"title":"[embedding] default_model","slug":"/docs/config#embedding","kicker":"CONFIG","snip":"string · Model name (e.g. text-embedding-3-small, nomic-embed-text)."},{"title":"Raft","slug":"/docs/clustering","kicker":"CLUSTER","snip":"Embedded Raft — leader election, log replication, metadata only."},{"title":"Node draining","slug":"/docs/clustering","kicker":"CLUSTER","snip":"POST /v1/cluster/nodes/:id/drain before maintenance."},{"title":"Region split/merge","slug":"/docs/clustering","kicker":"CLUSTER","snip":"Range partitions auto-split at byte threshold, merge back when small."},{"title":"Systemd unit","slug":"/docs/operations","kicker":"OPS","snip":"Production unit file with hardening directives."},{"title":"Readiness probe","slug":"/docs/operations","kicker":"OPS","snip":"GET /v1/health/ready — only 200 once WAL replay is complete."},{"title":"Config reload","slug":"/docs/operations","kicker":"OPS","snip":"SIGHUP reloads most keys without restart."},{"title":"Hot backup","slug":"/docs/backup-restore","kicker":"OPS","snip":"rsync the data dir after POST /v1/admin/flush."},{"title":"S3 backup","slug":"/docs/backup-restore","kicker":"OPS","snip":"POST /v1/admin/backup → any S3-compatible endpoint."},{"title":"Rolling upgrade","slug":"/docs/upgrades","kicker":"OPS","snip":"Drain → stop → upgrade → start → activate, per node."},{"title":"Merge pressure","slug":"/docs/troubleshooting","kicker":"OPS","snip":"Segment count climbing → raise merge.max_concurrent + io_rate."},{"title":"Cluster flapping","slug":"/docs/troubleshooting","kicker":"OPS","snip":"Leader changes every few seconds → raise cluster.tick_ms."},{"title":"WAL warnings after a crash","slug":"/docs/troubleshooting","kicker":"OPS","snip":"Repaired on startup; no manual repair command."},{"title": "Recipes", "slug": "/docs/recipes/", "kicker": "RECIPES", "snip": "Task-oriented AI & search recipes, each verified end-to-end against a live XERJ."},{"title": "Semantic search & RAG", "slug": "/docs/recipes/semantic-search-rag", "kicker": "RECIPE", "snip": "Retrieval by meaning with semantic_text — auto-embed on ingest, no separate vector DB."},{"title": "Agent memory", "slug": "/docs/recipes/agentic-memory", "kicker": "RECIPE", "snip": "Long-term memory for agents via /_memory — store, recall, forget, per-agent isolation."},{"title": "Passage retrieval", "slug": "/docs/recipes/passage-retrieval", "kicker": "RECIPE", "snip": "Auto-chunk long docs; best-passage (max-sim) scoring — 98% top-3."},{"title": "Vector search · kNN", "slug": "/docs/recipes/vector-search-knn", "kicker": "RECIPE", "snip": "Nearest-neighbor over dense_vector — HNSW-served unfiltered, exact with filters."},{"title": "Vector quantization", "slug": "/docs/recipes/vector-quantization", "kicker": "RECIPE", "snip": "scalar8 (accepts ES int8_hnsw; exact code scan) — int8 scoring at recall@10 = 0.998. Precision, not memory."},{"title": "Hybrid search", "slug": "/docs/recipes/hybrid-search", "kicker": "RECIPE", "snip": "Keyword + vector in one query — RRF fusion."},{"title": "Log analytics", "slug": "/docs/recipes/log-analytics", "kicker": "RECIPE", "snip": "Raw logs to dashboards — error rates, p95 latency, top services."},{"title": "Anomaly detection", "slug": "/docs/recipes/anomaly-detection", "kicker": "RECIPE", "snip": "Statistical _ml detectors that flag spikes in metrics/logs."},{"title": "Anomaly datafeeds", "slug": "/docs/recipes/continuous-anomaly-datafeeds", "kicker": "RECIPE", "snip": "Live _ml datafeed re-scores an index on a timer, stores anomaly records."},{"title": "Migrate from ES", "slug": "/docs/recipes/migrate-from-elasticsearch", "kicker": "RECIPE", "snip": "Point your existing ES client at XERJ — same wire, change the URL."}]</script>

<script>
  // IIFE — js-auth navigates between decrypted pages in the same
  // JS realm, so const bindings must be lexically scoped.
  (function () {
    const root = document.documentElement;

    // Client-side docs search — filters the inline JSON index.
    const searchEl    = document.getElementById('docsSearch');
    const inputEl     = document.getElementById('docsSearchInput');
    const resultsEl   = document.getElementById('docsSearchResults');
    const indexJson   = document.getElementById('docs-index');
    if (searchEl && inputEl && resultsEl && indexJson) {
      let index = [];
      try { index = JSON.parse(indexJson.textContent || '[]'); } catch {}
      const render = (q) => {
        const query = q.trim().toLowerCase();
        if (!query) { searchEl.classList.remove('active'); resultsEl.innerHTML = ''; return; }
        const scored = [];
        for (const e of index) {
          const hay = (e.title + ' ' + (e.kicker || '') + ' ' + (e.snip || '')).toLowerCase();
          if (hay.includes(query)) {
            let score = 0;
            if (e.title.toLowerCase().includes(query)) score += 10;
            if ((e.kicker || '').toLowerCase() === query.toUpperCase()) score += 5;
            if ((e.snip || '').toLowerCase().includes(query)) score += 1;
            scored.push({ e, score });
          }
        }
        scored.sort((a, b) => b.score - a.score);
        const top = scored.slice(0, 20);
        if (top.length === 0) {
          resultsEl.innerHTML = '<div class="no-results">NO MATCHES — TRY A SHORTER QUERY</div>';
        } else {
          resultsEl.innerHTML = top.map(({e}) => {
            const kicker = e.kicker || '';
            const snip = e.snip ? '<div class="snip">' + escHtml(e.snip) + '</div>' : '';
            return '<a href="' + e.slug + '">' +
              '<span class="kicker">' + escHtml(kicker) + '</span>' +
              '<span class="title">' + escHtml(e.title) + '</span>' +
              snip + '</a>';
          }).join('');
        }
        searchEl.classList.add('active');
      };
      const escHtml = (s) => String(s).replace(/&/g, '&amp;').replace(/</g, '&lt;').replace(/>/g, '&gt;').replace(/"/g, '&quot;');
      inputEl.addEventListener('input', (e) => render(e.target.value));
      inputEl.addEventListener('focus', () => { if (inputEl.value) render(inputEl.value); });
      document.addEventListener('keydown', (e) => {
        if ((e.key === 'k' || e.key === 'K') && (e.metaKey || e.ctrlKey)) {
          e.preventDefault();
          inputEl.focus();
          inputEl.select();
        }
        if (e.key === 'Escape' && document.activeElement === inputEl) {
          inputEl.value = '';
          searchEl.classList.remove('active');
          resultsEl.innerHTML = '';
          inputEl.blur();
        }
      });
    }
  })();
</script>

</body>
</html>