Q&A

Ask natural language questions about your files. Roset uses Retrieval Augmented Generation (RAG) to find relevant documents via vector search, then generates an answer with source citations.

How It Works#

1. Your question is embedded using OpenAI text-embedding-3-small
2. Vector search finds the most relevant document chunks
3. The matched content is sent as context to gpt-4o-mini
4. You get an answer with citations back to specific files

Quick Start#

Parameters#

Response#

{
  "answer": "The payment terms are net 30 days from the invoice date...",
  "sources": [
    {
      "fileId": "abc-123",
      "filename": "contract.pdf",
      "snippet": "Payment shall be made within thirty (30) days...",
      "score": 0.89
    }
  ],
  "question": "What are the payment terms?"
}

Streaming#

When stream: true, the response is an SSE stream with three event types:

data: {"type": "chunk", "content": "The payment"}
data: {"type": "chunk", "content": " terms are"}
data: {"type": "chunk", "content": " net 30 days..."}
data: {"type": "sources", "sources": [...]}
data: [DONE]

Next Steps#

• Search — Lower-level search without LLM answer generation
• TypeScript SDK — Full SDK reference