Scholar RAG Agent
A local-first Python toolkit and FastAPI service for inspectable retrieval-augmented literature review, showing the retrieved passages and run history behind each answer.
Scholar RAG Agent lets a user load a small corpus of papers, ask comparison or hypothesis questions, and inspect the exact passages and run history behind the response, rather than trusting an opaque summary. It runs without model credentials by default, using a fake adapter that demonstrates the workflow without producing a real scientific summary.
It persists data in SQLite and supports optional model-provider adapters for real inference. Its API lets a user ingest text and query it, browse and filter a catalog of ingested documents, save and reuse named paper collections, restrict a query to selected documents, preview the exact chunks and retrieval plan before generation, require a minimum number of evidence documents before generating an answer, and limit how many passages one paper contributes.
It is a local-first Python toolkit and FastAPI service; retrieval uses hybrid and graph-based methods with provenance kept alongside saved evidence.