Conversational AI over your documents.
Chat uses your knowledge bases to answer questions with citations. RAG retrieval finds relevant chunks and the model responds using your content. Integrate via API or use the hosted UI — answers stay grounded and traceable.
Chat
Conversational AI over your documents.
Chat uses your knowledge bases to answer questions with citations. RAG retrieval finds relevant chunks and the model responds using your content, so answers stay grounded and traceable.
Integrate chat via API or use the hosted UI. Multi-document and multi-session conversations make it easy to add document Q&A to your product.
Knowledge bases & retrieval
Add documents, embeddings, and hybrid search.
Add processed documents to a knowledge base via upload or pipeline. DocLD embeds chunks (e.g. llama-text-embed-v2) and indexes them in Pinecone. When you send a question, the API runs similarity search and retrieves top-k relevant chunks to ground the answer.
Each chat session is tied to a knowledge_base_id so responses stay scoped to your data. Use the same knowledge base across many sessions or create separate bases for different document sets.
Response modes
Quick, deep, or code — tune for the question.
Choose a response mode per request: quick for fast, focused answers with minimal context; deep for comprehensive analysis with more sources; code for code-related questions in your docs. Each mode adjusts retrieval depth and generation style.
Pass mode in the request body when calling POST /api/chat. Use quick for simple Q&A and deep when you need thorough, multi-source answers.
Sessions & API
Create a session, then send messages; stream-only response.
Create a session with POST /api/chat/sessions and knowledge_base_id; you get a session_id. Send messages with POST /api/chat using that session_id and the same knowledge_base_id. The API expects a single message (AI SDK UIMessage format) and returns a streamed UIMessage response (SSE).
Get conversation history with GET /api/chat?session_id=.... List, get, patch, and delete sessions via the sessions endpoints. All chat responses are stream-only for real-time UX.
Citations & tools
Source citations, spreadsheet Q&A, charts, and more.
Every answer can include source citations so users see which passages support the response. The assistant can also query spreadsheets (e.g. "top 10 by price"), generate bar/line/pie charts from data, summarize a cited chunk on request, and search only within documents from a previous answer (scoped search).
Shareable links let you expose a read-only view of a conversation. Tools such as search_documents, query_spreadsheet, and share_link appear as tool-output-available chunks in the stream.
Share & export
Share sessions with a link; export as JSON, Markdown, or HTML.
Share a session with others via POST /api/chat/sessions/:id/share; you get a shareable link. Revoke with DELETE .../share. Optional approval flow controls who can view the conversation.
Export a session with GET /api/chat/sessions/:id/export?format=json|markdown|html for backup, compliance, or integration into other tools.
How it works
| Step | Description |
|---|---|
| Knowledge base | Add processed documents to a knowledge base; chunks are embedded and indexed. |
| Query | User asks a question in natural language. |
| Search | Query is embedded and similarity search runs (e.g. Pinecone). |
| Retrieve | Top-k relevant chunks are fetched (count depends on mode). |
| Generate | LLM produces an answer using the retrieved context. |
| Cite | Sources are linked to the response so users can verify. |
Response modes
| Mode | Description | Best for |
|---|---|---|
| quick | Fast, focused responses with minimal context | Simple questions, quick facts |
| deep | Comprehensive analysis with multiple sources | Complex questions, detailed explanations |
| code | Optimized for finding and explaining code | Code in docs, APIs, snippets |
API overview
| Endpoint | Description |
|---|---|
| POST /api/chat/sessions | Create session (pass knowledge_base_id); returns session_id |
| POST /api/chat | Send message (message, session_id, knowledge_base_id); streamed UIMessage response |
| GET /api/chat?session_id=... | Get conversation history for a session |
| GET /api/chat/sessions | List sessions (optional knowledge_base_id filter) |
| GET /api/chat/sessions/:id | Get session details |
| PATCH /api/chat/sessions/:id | Update session (e.g. title) |
| DELETE /api/chat/sessions/:id | Delete session |
Chat: Questions & Answers
Ready to add chat to your documents?
Get started with a knowledge base and the Chat API in minutes. Sign up for free or read the full API reference for sessions, streaming, and tools.