Research Agent

The Research agent searches stored documents and synthesizes answers with citations. It is typically invoked standalone for ad-hoc queries, or conditionally after QA failure.

Tools

Output: ResearchResult

class ResearchResult(BaseModel):
    query: str
    answer: str
    citations: list[Citation]
    documents_searched: int
    reasoning: str
 
class Citation(BaseModel):
    document_id: str
    source_filename: str
    chunk_index: int
    text_snippet: str
    score: float

How It Works

1. Receives a research query (either from QA failure context or direct user query)
2. Calls search_store with the query embedding to find relevant chunks
3. Calls get_document_info to enrich results with document metadata
4. Optionally calls get_store_stats for context on corpus size
5. Synthesizes an answer grounded in the retrieved chunks
6. Returns the answer with full citation chain

Standalone Usage

async with Orchestrator() as orch:
    result = await orch.research("What are the payment terms in our contracts?")
    print(result.answer)
    for cite in result.citations:
        print(f"  [{cite.source_filename}] p.{cite.chunk_index}: {cite.text_snippet[:80]}")