ragsage

ragsage

A sage that only speaks from your corpus.

ragsage answers questions from your documents, and says so honestly when it can't. Every answer is either grounded prose with citations that resolve back to a chunk and a page, or an explicit not found — the engine never invents a confident answer out of thin retrieval.

pip install ragsage

It owns the whole RAG domain — parsing, chunking, contextualizing, embedding, hybrid retrieval, reranking, grounded generation and citations — behind thirteen typing.Protocol ports, so any model or store is swapped by implementing an interface rather than by editing the engine. It never imports web, auth or tenant concepts: callers pass an opaque Scope, and the engine treats it as an untyped label.

New here?

Quickstart runs the full ingest-and-query loop offline in about sixty seconds, with no database and no API key.

Where to go next

Status

Alpha. The public surface is stable in shape, but minor versions may still break it before 1.0. Source, issues and pull requests live on GitHub; the package is on PyPI. Licensed MIT.

On this page