Local-first RAG · Open Source

Private RAG infrastructure for teams that need reliable answers

Keep your documents on your own infrastructure. Retrieve grounded evidence. Generate answers with local LLMs.

Recall@8 1.000 · MRR 0.875 on public gold set

Keep your knowledge base private

Data and indexes stay on your infrastructure. No document ever leaves your network. Suitable for regulated industries and EU data-residency requirements.

Separate evidence from generation

The /retrieve endpoint returns ranked excerpts before any generation. Auditable, citation-friendly, and fully transparent about where each answer comes from.

Tune and measure quality

Built-in eval reports Recall@k and MRR. Adjust BM25, vector weights, and fusion settings until quality meets your standard — no black-box guessing.

Engine

rag-agent

The retrieval and generation engine. Ingest Markdown or HTML corpora, run hybrid BM25 + vector search, and answer with any OpenAI-compatible LLM endpoint.

Learn more
Orchestration

rag-router

Multi-shelf orchestration. Routes questions to the right rag-agent instance, proxies SSE streaming, and exposes MCP tools for Cursor IDE.

Learn more

Support the project

From €2/month — help keep rag-agent maintained and documented.

See pricing

Need a pilot?

2–4 weeks · one corpus · deployed on your infrastructure.

Book a pilot