RAGStack indexes your documents four ways at once — keyword, vector, knowledge-graph and SQL — then routes every question through whichever strategy suffices. Every answer is split into claims, checked against its citations, and withheld when support is thin.
Nothing hidden behind a marketing page — every component, decision and control is written up in plain English.
The two pipelines end to end: how a document becomes four indexes, and how a question becomes a verified answer.
Read →Every module explained: parsers, chunker, three stores, graph engine, eleven agent tools, verifier, caches.
Browse →Install, configure, every CLI command, the HTTP API with its SSE event catalog, sessions, auth, Docker.
Use →Evidence grading, claim verification, abstention policy, injection defense, audit traces.
Trust →What we measure, how to run the harness, and the published research that shaped the design.
Measure →This page talks to 127.0.0.1:8000 — nothing is sent to us. Start the server locally and this console goes live.
# install pip install ragstack # index a folder of documents ragstack index ./docs # ask — or `ragstack serve` for the console ragstack query "how does fusion rank results?"
from ragstack import RAGStack svc = RAGStack("ragstack.yaml") svc.ingest(["./docs"]) answer = svc.query( "which components depend on auth?", mode="agentic", session_id="team", ) print(answer.text) print(answer.confidence)
MIT licensed. Local-first. Built to be read.