> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://respan.ai/docs/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://respan.ai/docs/_mcp/server.

# Open-source gateway

[respan-gateway](https://github.com/respanai/respan-gateway) is an open-source LLM gateway you run on your own machine or server. Point any official SDK at it, name any model, and it translates the request to that model's provider: chat, images, speech, video and realtime voice.

> **Note**
>
> respan-gateway is pre-alpha. Its config and APIs may still change. It is MIT or Apache-2.0 licensed.

## What it does

| Capability            | For example                                                                                                                                            |
| --------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------ |
| Any SDK, any model    | The Anthropic SDK calls `gpt-5.5`; the OpenAI SDK calls Claude; the Google GenAI SDK calls GPT. Tools, images, streaming and reasoning are translated. |
| Images, speech, video | One OpenAI client generates images, speech and transcriptions, and `client.videos.create` makes a Google Veo video.                                    |
| Realtime voice        | An OpenAI Realtime client talks to Gemini Live, and a Gemini Live client talks to GPT Realtime.                                                        |
| Routing               | Clients ask for `smart`; the gateway picks the cheapest or fastest model, falls back on errors, or races two providers.                                |
| MCP tools             | Put MCP servers behind one key, or let any model use their tools mid-request, with no client code.                                                     |
| Request log           | See every request, its provider, latency, tokens and cost in a built-in web UI, and send each one to Respan as a trace.                                |

It ships as one binary with the web UI built in. 16 providers are built in, including OpenAI, Anthropic, Gemini, Groq, DeepSeek, Mistral, xAI, OpenRouter and Ollama. Any OpenAI-compatible endpoint works with a base URL.

## How it fits with Respan

|               | Respan gateway                         | Open-source gateway            |
| ------------- | -------------------------------------- | ------------------------------ |
| Runs          | Hosted at `https://api.respan.ai/api/` | On your machine or server      |
| Provider keys | Respan credits, or your keys in Respan | Your keys, in your environment |
| Setup         | Change your client's base URL          | Install a binary, add keys     |

You can combine them: add your `RESPAN_API_KEY` as a provider to reach 1000+ models with one key, and turn on the `respan` plugin to trace every request in Respan. See [Configure plugins](/docs/gateway/open-source/configure).

To start, follow the [Quickstart](/docs/gateway/open-source/quickstart).