Skip to content
<- All products
// product

Ragmux

One API, many models, your own documents.

Brand: Tunedness

Visit siteragmux.com

Ragmux is an AI gateway you run on your own infrastructure. Your applications talk to a single OpenAI-compatible API; Ragmux forwards each request to the provider behind it and, when asked, adds context retrieved from the documents you have uploaded. It is built by Tunedness.

One API, many providers

The /v1/chat/completions (JSON and SSE), /v1/models, /v1/embeddings, /v1/retrieve and /v1/rerank endpoints are OpenAI-compatible. Anthropic and Gemini translation covers tool calls too. Anthropic SDKs and Claude Code connect through /v1/messages by changing only the base URL. A single reasoning_effort setting maps onto each provider’s own thinking controls.

Retrieval from your documents

You upload PDF, DOCX, HTML, TXT and Markdown files. Ragmux offers section-aware chunking, hybrid search that combines pgvector with PostgreSQL full-text search or ParadeDB BM25, reranking and metadata filtering. The RAG store can also be opened to agents as a search tool through the MCP server at /mcp.

Routing and control

Targets can be prioritized or weighted; requests are retried before the first byte and each connection has a circuit breaker. Each project gets an sk-proj- key, and there are admin, editor and viewer roles, login lockout and an audit log. Per-project rate limits and token budgets are enforced, and costs are estimated from an editable price table. For observability there is a request log, an authenticated Prometheus /metrics endpoint and OpenTelemetry traces.

Setup and limits

Ragmux is a static Go binary that ships with PostgreSQL and pgvector in a single container. No separate vector database or Redis is needed. Not available yet: OCR for scanned PDFs, SSO/OIDC login and budgets in money terms.

License

AGPL-3.0-or-later. A commercial license is also available. The source is developed on GitHub.

// Products

Key features

  • OpenAI, Azure OpenAI, Anthropic, Gemini, DeepSeek, Ollama and custom OpenAI-compatible endpoints behind one OpenAI-compatible API
  • Hybrid search and reranking over uploaded documents (PDF, DOCX, HTML, TXT, Markdown)
  • Anthropic-compatible /v1/messages and an MCP server that exposes the RAG store as a search tool
  • Per-project keys, role-based access, rate limits and token budgets
  • Priority and weighted routing, retries, per-connection circuit breaker
  • One Docker container: no separate vector database and no Redis
// faq

Frequently asked questions

What does Ragmux actually do?+

It is a self-hosted AI gateway that exposes one OpenAI-compatible API for several LLM providers and adds retrieval from uploaded documents.

Which providers does it work with?+

OpenAI, Azure OpenAI, Anthropic, Gemini, DeepSeek and Ollama are supported directly. OpenAI-compatible endpoints such as vLLM, LM Studio or LiteLLM can be added as custom providers. Cohere and Voyage are available for reranking.

How is it installed?+

With Docker. There is an all-in-one container image, a gateway-only image and an image with ParadeDB BM25 search, plus a docker-compose file. The images are signed.

Does my data leave my servers?+

The project states that nothing phones home: it sends no telemetry. Provider keys are stored encrypted with AES-256-GCM. Requests go only to the providers you configure.

Does it handle scanned PDFs?+

No. OCR is not available yet. SSO/OIDC login and budgets in money terms are not shipped either.

What are the license and the price?+

Ragmux is free software under AGPL-3.0-or-later. A commercial license is also available; its price is not stated in public sources.

Let’s build your next project together

Tell us your idea and we’ll design the right solution for you.

Get in touch