RAG Chat

Chat with your documents using retrieval-augmented generation (RAG).

How It Works


Question → Search (Pinecone) → Retrieve → Generate → Respond with Citations

Knowledge Base — Add processed documents to a knowledge base
Query — Ask a question in natural language
Search — Query is sent to Pinecone; embeddings and similarity search happen server-side (llama-text-embed-v2)
Retrieve — Get top-k relevant chunks
Generate — LLM creates answer using context
Cite — Sources are linked to the response

The assistant can also query spreadsheets (e.g. “top 10 by price”), summarize cited chunks, search only within documents from a previous answer, create charts (bar, line, pie), and generate shareable links for the conversation.

Features

Feature	Description
Context-aware	Answers grounded in your documents
Source citations	See which passages support each answer
Confidence scores	Know how reliable the answer is
Session history	Continue conversations across messages
Streaming	Real-time response streaming
Multi-language	Ask questions in any language
Export	Download chat sessions
Sharing	Share sessions with others
Spreadsheet Q&A	Query spreadsheets with sort/limit (e.g. cheapest, top N by column)
Charts	Display bar, line, or pie charts from data
Summarize citation	Summarize a cited document or chunk on request
Scoped search	Search only within documents from a previous answer
Share link	Generate a shareable link for the conversation (with optional approval)

Getting Started

1. Create a Knowledge Base


curl -X POST "/api/knowledge-bases" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{"name": "Company Policies"}'

2. Add Documents


curl -X POST "/api/upload" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@handbook.pdf" \
  -F "knowledge_base_id=kb-uuid"

3. Chat

Create a session first (POST /api/chat/sessions with knowledge_base_id), then send a message. The API expects a single message object with role and parts (AI SDK UIMessage format):


curl -X POST "/api/chat" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "message": { "role": "user", "parts": [{ "type": "text", "text": "What is the vacation policy?" }] },
    "session_id": "session-uuid",
    "knowledge_base_id": "kb-uuid"
  }'

See the Chat API for the full request shape and streaming response.

Response Format

The chat endpoint is stream-only. The response is an AI SDK UIMessage stream (SSE): you will see text-delta chunks for the assistant’s reply and tool-output-available chunks when tools (e.g. search_documents, query_spreadsheet) return context and citations. For a non-streaming style response, use the Embed API with action: 'chat'; see the Chat API for full stream and tool details.

Response Modes

Mode	Description	Best For
`quick`	Fast response, fewer chunks	Simple questions
`deep`	Deeper search, more context	Complex questions
`code`	Optimized for code-related questions	Code in docs


curl -X POST "/api/chat" \
  -d '{
    "message": { "role": "user", "parts": [{ "type": "text", "text": "Explain all the benefit options in detail" }] },
    "session_id": "session-uuid",
    "knowledge_base_id": "kb-uuid",
    "mode": "deep"
  }'

Sessions

Session Management

Conversations are organized into sessions:


# List sessions
curl -X GET "/api/chat/sessions?knowledge_base_id=kb-uuid"
 
# Get session details
curl -X GET "/api/chat/sessions/{session_id}"
 
# Update session title
curl -X PATCH "/api/chat/sessions/{session_id}" \
  -d '{"title": "Benefits Discussion"}'
 
# Delete session
curl -X DELETE "/api/chat/sessions/{session_id}"

Continuing Conversations

Pass session_id to continue a conversation:


curl -X POST "/api/chat" \
  -d '{
    "message": { "role": "user", "parts": [{ "type": "text", "text": "What about sick leave?" }] },
    "session_id": "existing-session-uuid",
    "knowledge_base_id": "kb-uuid"
  }'


curl -X POST "/api/chat/sessions/{id}/share"

Response includes a shareable URL:


{
  "share_url": "https://your-domain.com/chat/share/token123"
}


curl -X DELETE "/api/chat/sessions/{id}/share"

Export

Export sessions in various formats:


# JSON
curl -X GET "/api/chat/sessions/{id}/export?format=json"
 
# Markdown
curl -X GET "/api/chat/sessions/{id}/export?format=markdown"
 
# HTML
curl -X GET "/api/chat/sessions/{id}/export?format=html"

Multi-Language Support

Ask questions in any language. Use the same UIMessage shape and include session_id and knowledge_base_id; optionally set language (e.g. fr) for response language:


curl -X POST "/api/chat" \
  -d '{
    "message": { "role": "user", "parts": [{ "type": "text", "text": "Quelle est la politique de vacances?" }] },
    "session_id": "session-uuid",
    "knowledge_base_id": "kb-uuid",
    "language": "fr"
  }'

The response will be in the specified language when supported.

Streaming Example (JavaScript)

The endpoint is stream-only and returns an AI SDK UIMessage stream (e.g. text-delta, tool-output-available). Create a session first, then send a message with the UIMessage shape:


// 1. Create session
const sessionRes = await fetch('/api/chat/sessions', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({ knowledge_base_id: 'kb-uuid' })
});
const { session_id } = await sessionRes.json();
 
// 2. Send message and read stream
const response = await fetch('/api/chat', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    message: { role: 'user', parts: [{ type: 'text', text: 'What are the payment terms?' }] },
    session_id,
    knowledge_base_id: 'kb-uuid'
  })
});
 
const reader = response.body.getReader();
const decoder = new TextDecoder();
let buffer = '';
while (true) {
  const { done, value } = await reader.read();
  if (done) break;
  buffer += decoder.decode(value, { stream: true });
  // Parse SSE lines (data: {...}) and handle text-delta / tool-output-available
}

Best Practices

Curate knowledge bases — Include relevant, quality documents
Be specific — Clear questions get better answers
Check citations — Verify answers against sources
Use sessions — Continue conversations for follow-ups
Organize by topic — Separate knowledge bases by domain
Review confidence — Low scores may need human review

Credit Usage

Operation	Credits
Chat query	0.1 per query

API Reference

See the Chat API for complete endpoint documentation.

RAG Chat

Chat with your documents using retrieval-augmented generation (RAG).

How It Works


Question → Search (Pinecone) → Retrieve → Generate → Respond with Citations

Knowledge Base — Add processed documents to a knowledge base
Query — Ask a question in natural language
Search — Query is sent to Pinecone; embeddings and similarity search happen server-side (llama-text-embed-v2)
Retrieve — Get top-k relevant chunks
Generate — LLM creates answer using context
Cite — Sources are linked to the response

Features

Feature	Description
Context-aware	Answers grounded in your documents
Source citations	See which passages support each answer
Confidence scores	Know how reliable the answer is
Session history	Continue conversations across messages
Streaming	Real-time response streaming
Multi-language	Ask questions in any language
Export	Download chat sessions
Sharing	Share sessions with others
Spreadsheet Q&A	Query spreadsheets with sort/limit (e.g. cheapest, top N by column)
Charts	Display bar, line, or pie charts from data
Summarize citation	Summarize a cited document or chunk on request
Scoped search	Search only within documents from a previous answer
Share link	Generate a shareable link for the conversation (with optional approval)

Getting Started

1. Create a Knowledge Base


curl -X POST "/api/knowledge-bases" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{"name": "Company Policies"}'

2. Add Documents


curl -X POST "/api/upload" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -F "file=@handbook.pdf" \
  -F "knowledge_base_id=kb-uuid"

3. Chat

Create a session first (POST /api/chat/sessions with knowledge_base_id), then send a message. The API expects a single message object with role and parts (AI SDK UIMessage format):


curl -X POST "/api/chat" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "message": { "role": "user", "parts": [{ "type": "text", "text": "What is the vacation policy?" }] },
    "session_id": "session-uuid",
    "knowledge_base_id": "kb-uuid"
  }'

See the Chat API for the full request shape and streaming response.

Response Format

Response Modes

Mode	Description	Best For
`quick`	Fast response, fewer chunks	Simple questions
`deep`	Deeper search, more context	Complex questions
`code`	Optimized for code-related questions	Code in docs


curl -X POST "/api/chat" \
  -d '{
    "message": { "role": "user", "parts": [{ "type": "text", "text": "Explain all the benefit options in detail" }] },
    "session_id": "session-uuid",
    "knowledge_base_id": "kb-uuid",
    "mode": "deep"
  }'

Sessions

Session Management

Conversations are organized into sessions:


# List sessions
curl -X GET "/api/chat/sessions?knowledge_base_id=kb-uuid"
 
# Get session details
curl -X GET "/api/chat/sessions/{session_id}"
 
# Update session title
curl -X PATCH "/api/chat/sessions/{session_id}" \
  -d '{"title": "Benefits Discussion"}'
 
# Delete session
curl -X DELETE "/api/chat/sessions/{session_id}"

Continuing Conversations

Pass session_id to continue a conversation:


curl -X POST "/api/chat" \
  -d '{
    "message": { "role": "user", "parts": [{ "type": "text", "text": "What about sick leave?" }] },
    "session_id": "existing-session-uuid",
    "knowledge_base_id": "kb-uuid"
  }'


curl -X POST "/api/chat/sessions/{id}/share"

Response includes a shareable URL:


{
  "share_url": "https://your-domain.com/chat/share/token123"
}


curl -X DELETE "/api/chat/sessions/{id}/share"

Export

Export sessions in various formats:


# JSON
curl -X GET "/api/chat/sessions/{id}/export?format=json"
 
# Markdown
curl -X GET "/api/chat/sessions/{id}/export?format=markdown"
 
# HTML
curl -X GET "/api/chat/sessions/{id}/export?format=html"

Multi-Language Support

Ask questions in any language. Use the same UIMessage shape and include session_id and knowledge_base_id; optionally set language (e.g. fr) for response language:


curl -X POST "/api/chat" \
  -d '{
    "message": { "role": "user", "parts": [{ "type": "text", "text": "Quelle est la politique de vacances?" }] },
    "session_id": "session-uuid",
    "knowledge_base_id": "kb-uuid",
    "language": "fr"
  }'

The response will be in the specified language when supported.

Streaming Example (JavaScript)

The endpoint is stream-only and returns an AI SDK UIMessage stream (e.g. text-delta, tool-output-available). Create a session first, then send a message with the UIMessage shape:


// 1. Create session
const sessionRes = await fetch('/api/chat/sessions', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({ knowledge_base_id: 'kb-uuid' })
});
const { session_id } = await sessionRes.json();
 
// 2. Send message and read stream
const response = await fetch('/api/chat', {
  method: 'POST',
  headers: {
    'Authorization': 'Bearer YOUR_API_KEY',
    'Content-Type': 'application/json'
  },
  body: JSON.stringify({
    message: { role: 'user', parts: [{ type: 'text', text: 'What are the payment terms?' }] },
    session_id,
    knowledge_base_id: 'kb-uuid'
  })
});
 
const reader = response.body.getReader();
const decoder = new TextDecoder();
let buffer = '';
while (true) {
  const { done, value } = await reader.read();
  if (done) break;
  buffer += decoder.decode(value, { stream: true });
  // Parse SSE lines (data: {...}) and handle text-delta / tool-output-available
}

Best Practices

Curate knowledge bases — Include relevant, quality documents
Be specific — Clear questions get better answers
Check citations — Verify answers against sources
Use sessions — Continue conversations for follow-ups
Organize by topic — Separate knowledge bases by domain
Review confidence — Low scores may need human review

Credit Usage

Operation	Credits
Chat query	0.1 per query

API Reference

See the Chat API for complete endpoint documentation.

RAG Chat

How It Works

Features

Getting Started

1. Create a Knowledge Base

2. Add Documents

3. Chat

Response Format

Response Modes

Sessions

Session Management

Continuing Conversations

Sharing

Enable Sharing

Disable Sharing

Export

Multi-Language Support

Streaming Example (JavaScript)

Best Practices

Credit Usage

API Reference

RAG Chat

How It Works

Features

Getting Started

1. Create a Knowledge Base

2. Add Documents

3. Chat

Response Format

Response Modes

Sessions

Session Management

Continuing Conversations

Sharing

Enable Sharing

Disable Sharing

Export

Multi-Language Support

Streaming Example (JavaScript)

Best Practices

Credit Usage

API Reference