Guides

AI and Agent Support

Make your Barodoc site friendly to AI assistants and agents: llms.txt, manifest, chat, and RAG

You can make a Barodoc documentation site consumable by AI assistants and agentic tools in several ways: built-in artifacts, third-party chat widgets, and custom RAG or APIs.

Built-in: llms.txt and docs manifest

llms.txt (plugin)

The LLMs.txt plugin generates llms.txt files at build time:

• /llms.txt — Summary with page titles, descriptions, and links (spec-compliant).
• /llms-full.txt — Full text of all docs when full: true (optional).

Many AI tools and crawlers can use these URLs to understand and cite your docs. Enable the plugin in barodoc.config.json:

{
  "plugins": ["@barodoc/plugin-llms-txt"]
}

docs-manifest.json (CLI)

The barodoc manifest command generates a structured JSON file (e.g. for CI or tooling) with metadata and optional content chunks:

barodoc manifest
# Output: docs-manifest.json (default)

barodoc manifest --output manifest.json --chunks
# With --chunks: includes section-level content for RAG or indexing

Use this to feed custom pipelines, MCP servers, or internal agent tools. The manifest is not deployed by default; you run the command and use the file where needed.

Adding a chat or “ask docs” widget

Barodoc does not ship a built-in chat UI. You can add one in two ways:

1. Third-party embed

Services like Inline, Mintlify, or provider-specific “docs chat” widgets can be embedded in your layout. Typically you:

• Sign up and get a snippet or script URL.
• In Full custom mode, add the script or component to your layout (e.g. in @barodoc/theme-docs overrides or a custom layout).
• In Quick mode, use overrides to inject a custom layout or component that includes the widget.

The provider usually handles indexing your public docs (often by crawling or using an API) and serves the chat backend.

2. Custom widget + your backend

You can build your own “Ask docs” UI and backend:

• Frontend: Add a React (or Astro) component that talks to your API (e.g. “send question → get answer”). Mount it via overrides or a custom layout in Full mode.
• Backend: Implement a RAG or LLM API that uses your docs (e.g. content from llms-full.txt, barodoc manifest --chunks, or your own index). Host this separately (serverless or server).

Barodoc stays a static site; the chat backend is your service.

RAG and custom APIs

To power RAG (retrieval-augmented generation) or an API that answers questions from your docs:

1. Content source
   Use one or more of:

   • llms-full.txt — Single flat file with full text (good for simple ingestion).
   • barodoc manifest --chunks — Structured chunks with metadata (slug, section, locale, tags).
   • Crawling — Crawl the deployed site (HTML or raw Markdown if you use a raw-MD endpoint).

2. Indexing
   Ingest that content into your vector store or search engine (e.g. OpenAI embeddings, Pinecone, Algolia, or a simple keyword index).

3. API
   Expose an endpoint that accepts a question, retrieves relevant chunks, and returns an answer (e.g. via an LLM). Your Barodoc site can then call this API from a custom chat component.

This is all outside Barodoc’s scope; Barodoc only provides the content and optional artifacts (llms.txt, manifest).

Summary

For plugin setup, see LLMs.txt plugin and Recommended plugins.