Retrieval-augmented generation
Files are chunked, embedded, and stored in PostgreSQL with pgvector. Questions retrieve relevant passages and are answered with the configured language model.
Document index
Select PDF, Word, or plain text files to add to the index.
| File | Status | Chunks | Size | Actions |
|---|---|---|---|---|
| Loading… | ||||
Grounded chat
Model Context Protocol
Open a server-sent events channel, then invoke tools that delegate to the same RAG and document services. Use the spotlight tools to show how an MCP client can orchestrate structured actions (list → read text → keyword search) alongside semantic RAG.
Suggested live demo
- Upload at least one document above and wait until status is PROCESSED.
- Connect MCP, run get_documents — point out JSON your agent receives.
- Run get_document_text with that document’s id — “open the book” via a tool call.
- Run spotlight_search with a word you know appears in the file — exact phrase vs vector chat.
- Run rag_query with the same topic — contrast semantic retrieval with keyword spotlight.
Idle
Last structured result
Connect and run a tool — the latest tool_result JSON appears here.