About Ragmux
Updated:
Ragmux is a self-hosted AI gateway. It sits between your applications and the model providers you use, exposes a single OpenAI-compatible API, and can add retrieval from your own documents to every request it forwards.
Why it exists
Most teams reach the same three problems at once: provider keys spread across services, no honest answer to “who spent what, on which project”, and a retrieval stack that arrives as a second database to operate.
Ragmux answers all three in one place. Provider credentials stay server-side and encrypted. Each project gets its own key, its own model connection, its own system prompt and its own budget. Documents, chunks and vectors live in the same PostgreSQL database as everything else — so what you back up is one database and one secret.
What it is, concretely
- A single Go binary, statically linked, on a distroless base image.
- Two containers: the gateway and PostgreSQL with pgvector. No Redis, no separate vector database.
- An OpenAI-compatible
/v1surface, JSON and SSE, that the official SDKs speak by changing two lines. - Adapters for OpenAI, Anthropic, Gemini, DeepSeek, Ollama, and anything OpenAI-compatible you run yourself.
Who builds it
Ragmux is developed in the open under the Tunedness umbrella. The source, the issue tracker and the documentation are all in the project repository.
It is still being built. The changelog is the honest record of what works today, and the roadmap section on the home page lists what is deliberately not in the current release.
Licence
Ragmux is free software licensed under the GNU Affero General Public License v3.0 or later (AGPL-3.0-or-later). The AGPL’s network-service clause applies: if you run a modified version as a service that users interact with over a network, you must offer those users the corresponding source of your modified version.
This website’s terms are separate from the software’s licence — see Terms of Use.