RAG Query

Execute a RAG (Retrieval-Augmented Generation) query.

This endpoint combines search results with language model generation to produce accurate, contextually-relevant responses based on your document corpus.

Features:

• Combines vector search, optional knowledge graph integration, and LLM generation
• Automatically cites sources with unique citation identifiers
• Supports both streaming and non-streaming responses
• Compatible with various LLM providers (OpenAI, Anthropic, etc.)
• Web search integration for up-to-date information

Search Configuration: All search parameters from the search endpoint apply here, including filters, hybrid search, and graph-enhanced search.

Generation Configuration: Fine-tune the language model’s behavior with rag_generation_config:

1
{
2
    "model": "openai/gpt-4o-mini",  // Model to use
3
    "temperature": 0.7,              // Control randomness (0-1)
4
    "max_tokens": 1500,              // Maximum output length
5
    "stream": true                   // Enable token streaming
6
}

Model Support:

• OpenAI models (default)
• Anthropic Claude models (requires ANTHROPIC_API_KEY)
• Local models via Ollama
• Any provider supported by LiteLLM

Streaming Responses: When stream: true is set, the endpoint returns Server-Sent Events with the following types:

• search_results: Initial search results from your documents
• message: Partial tokens as they’re generated
• citation: Citation metadata when sources are referenced
• final_answer: Complete answer with structured citations

Example Response:

1
{
2
"generated_answer": "DeepSeek-R1 is a model that demonstrates impressive performance...[1]",
3
"search_results": { ... },
4
"citations": [
5
    {
6
        "id": "cit.123456",
7
        "object": "citation",
8
        "payload": { ... }
9
    }
10
]
11
}