Ragz

Open source · Self-hosted · AGPL-3.0

Your private, agentic RAG platform.

Self-hosted, multi-tenant retrieval — your documents, your keys, your infrastructure. Ask questions, get cited answers, and never worry about where your data went.

Read the docsGitHub

Everything a production RAG deployment needs

One stack — FastAPI, React, Qdrant, LiteLLM, Postgres — with the access control and grounding guarantees enterprise deployments actually require.

Grounded, cited chat

Streaming answers with citations — document, version, section, and page. When retrieval lacks sufficient grounding, Ragz says so instead of guessing.

Rich generative UI

Answers render as more than text: visual cards, charts, tables, clickable source cards, follow-up chips, and accordions, driven by what the question needs.

Web search, on demand

Toggle web search per turn. Answers can cite full public web pages via DuckDuckGo by default, or Tavily if you configure it.

Multi-tenant RBAC

Orgs, workspaces, and groups with superadmin/admin/user tiers plus custom role templates. Document ACLs are enforced inside the vector query, never post-filtered — an answer can never cite a document you cannot open.

Bring any model

OpenAI, Anthropic, Gemini, or any LiteLLM provider — or run fully local with Ollama/vLLM and local embeddings and rerank for air-gapped installs.

Real documents

PPTX, DOCX, PDF and more, with OCR for scanned pages and version-aware retrieval — the latest approved version wins, superseded versions stay out of answers.

API + chat bots

An OpenAI-compatible API endpoint with per-user or per-workspace keys, plus inbound bots for Telegram, Discord, and Slack.

Secrets, encrypted

Every provider key is AES-256-GCM envelope-encrypted in Postgres under a single out-of-DB key (the KEK). Passwords are Argon2id-hashed. Keys are write-only and never logged.

Own everything.

Your documents never have to leave your infrastructure. Run Ragz on a single VM, point it at a cloud LLM or keep it fully air-gapped with local models — the choice is yours, and it can change later without switching platforms.

Your data

Documents, embeddings, and chat history stay on infrastructure you control.

Your keys

Provider keys are encrypted under a KEK only you hold — never sent anywhere else.

Your license

AGPL-3.0, fully open source. Read the code, fork it, run it forever.

Get started in about 10 minutes.

Bring up infrastructure with Docker Compose, start the backend and worker, then the frontend — you'll be chatting with your first document shortly after.

Step 1

Start infrastructure

docker compose up -d brings up Postgres, Qdrant, Redis, and MinIO.

Step 2

Run the backend

uv sync, migrate, bootstrap the first superadmin, then start the API, worker, and beat scheduler.

Step 3

Run the frontend

pnpm install && pnpm dev, sign in, add a model, and upload your first document.

Getting Started guideLinux install guide

Run your own RAG platform, on your own terms.

No usage-based pricing, no data-sharing terms to read twice. Clone it, deploy it, and keep every answer grounded in documents only your org can see.

Read the docsStar on GitHub