Legible
Chatbots & RAG

Legible: Turn Your Website into a Chatbot & AI Knowledge Layer

Empower your website with Legible's RAG pipeline. Easily create AI chatbots, customer support assistants, and external AI applications that leverage your website's content for intelligent responses.

8 min readUpdated 2026-03-21Chatbots & RAG
Why this matters

Legible does more than publish AI-readable pages. It also turns your content into a retrieval-ready layer for chatbots and AI assistants.

Internally, Legible chunks content and generates embeddings for search and retrieval. Externally, the product exposes that knowledge through the AI Export API, so customers can use it in Intercom, Zendesk, custom chatbots, and other RAG workflows.

What 'Chat With Your Website' Means In Legible

Legible's Content Chat is the product-facing version of the same retrieval system customers can use in their own support and assistant experiences. Instead of asking a model to guess from raw pages, Legible gives it clean content, chunked context, and source-aware retrieval.

That means the chatbot can answer from your synced website content rather than from a static FAQ export or a manually curated spreadsheet of answers.

The Legible RAG Pipeline

  • The same cleaned content can serve AI crawlers, internal testing, and customer-facing chatbots.
  • Chunking preserves structure so answers can stay grounded in specific sections of the source content.
  • Embeddings let Legible retrieve semantically relevant content, not just keyword matches.
1. Sync website content into Legible
2. Clean and normalize the content into Markdown
3. Split documents into chunks with heading context
4. Generate embeddings for those chunks
5. Retrieve the most relevant chunks for a query
6. Pass the retrieved context into Content Chat or your external assistant

Vector Database vs Chunk API

Customers often describe this as a hosted vector database, and that is a useful mental model. Legible does maintain embeddings and similarity search internally.

More precisely, the customer-facing product surface is an AI Export API that serves documents and chunks. So when you connect Legible to another system, you are usually integrating with a vector-backed retrieval and chunk-serving API, not managing the database layer yourself.

  • Internally: Legible stores chunk embeddings and runs retrieval search.
  • Externally: customers consume `ai-index`, `ai-content`, and `ai-chunks` endpoints.
  • This keeps the integration simple because teams do not have to build their own ingestion and embedding pipeline first.

What The AI Export API Gives You

  • `ai-index` for a paginated content inventory with document and chunk URLs.
  • `ai-content` for full document-level Markdown and structure.
  • `ai-chunks` for chunk-based ingestion and bulk retrieval use cases.
  • Policy information, hashes, and update timestamps for incremental syncs.

Who benefits and how

  • Marketing and content teams can improve site content once and have that benefit both AI discoverability and chat quality.
  • Support and product teams can test the knowledge layer in Content Chat before wiring it into external systems.
  • Engineering teams can integrate with the export API instead of building a full crawl, clean, chunk, embed, and sync pipeline from scratch.