Deterministic, build-time content extraction for Astro sites, designed for Large Language Model (LLM) usage.

Supports Astro 4, 5, and 6.

astro-llm generates a single, clean, static context file from your built HTML — suitable for:

• Retrieval-Augmented Generation (RAG)
• Chat grounding
• Offline LLM training
• Search indexing
• Auditable documentation snapshots

No runtime JavaScript.
No servers.
No magic.

• Build-time only – runs after astro build
• Deterministic output – same input, same output
• Config-first – behaviour controlled by llm.config.json
• Safety by default – sensitive data stripped
• LLM-friendly – readable, predictable structure

After your site is built:

1. Scans generated .html files in /dist
2. Extracts readable content in DOM order
3. Applies safety rules (email / phone / scripts)
4. Applies include/exclude rules
5. Writes a single output file (e.g. llm.txt or llm.json)

• ❌ No runtime DOM mutation
• ❌ No network requests
• ❌ No environment variables
• ❌ No telemetry or analytics
• ❌ No automatic crawling or discovery

Everything is explicit.

On first run (dev or build), astro-llm will:

• Create llm.config.json in the project root
• Populate it with explicit defaults
• Never overwrite it again

If the file already exists, it is left untouched.

This file is the single source of truth.

{
  "enabled": true,
  "output": {
    "format": "txt",
    "filename": "llm.txt"
  },
  "include": {
    "pages": true,
    "headings": true,
    "paragraphs": true,
    "lists": true,
    "tables": true,
    "codeBlocks": true,
    "meta": {
      "title": true,
      "description": true,
      "keywords": true
    }
  },
  "exclude": {
    "paths": [],
    "selectors": []
  },
  "safety": {
    "stripEmails": true,
    "stripPhoneNumbers": true,
    "stripForms": true,
    "stripScripts": true
  },
  "purpose": {
    "llmTraining": true,
    "ragIndexing": true,
    "chatGrounding": true
  }
}

---
PATH: /index.html
---
Page title
Section heading
Paragraph content here
[email removed]

{
  "documents": [
    "---\nPATH: /index.html\n---\nPage title Section heading Paragraph content here"
  ]
}

When enabled, the plugin removes:

• Email addresses → [email removed]
• Phone numbers → [phone removed]
• <script>, <style>, <form> blocks
• Inline JavaScript content

Already-encoded entities are preserved.

"exclude": {
  "paths": ["/admin", "/api"]
}

"exclude": {
  "selectors": [".llm-ignore", "#internal"]
}

Given:

• Same HTML output
• Same config
• Same plugin version

You will always get identical output.

• RAG pipelines
• Static knowledge bases
• LLM prompt grounding
• Offline semantic indexing
• Compliance-safe extraction

MIT © Velohost