LLM Bridge brings Claude to tools that only know OpenAI

Claude in any tool that requires an OpenAI provider

You want to use a Claude model in MacWhisper, but the tool only knows “OpenAI-compatible” providers. Many dictation and writing tools have an input field like this. LLM Bridge solves the problem on your own computer. The small proxy accepts requests in the OpenAI format, translates them into the Anthropic format, and sends them to LangDock or directly to Anthropic. Your tool thinks it’s an ordinary OpenAI provider.

What the bridge translates

LLM Bridge supports Chat Completions with and without streaming, system prompts, images, function calling, and JSON mode. Responses come back in the OpenAI format, including error messages. If you enter an incorrect model ID, the error tells you what’s wrong.

In the tool, enter an alias instead of the long model ID, such as haiku. You set the specific model ID behind it in LLM Bridge, and if you’ve configured multiple upstreams, you can assign the right one to each alias.

Switching models is then a single entry in one place instead of a change in every tool.

See what’s being used

Every request is logged in a local SQLite database: number of requests, input, output, and cache tokens, and response times, broken down by model and client. A CSV export is included. LLM Bridge identifies the client by its User-Agent, so you can see which tool uses how much. LLM Bridge doesn’t store the contents of your prompts or responses, only this metadata.

As a tray app or on the server

The tray app runs on macOS, Windows, and Linux and can launch at sign-in if you want. The interface is available in German and English and automatically detects the system language. A command-line program is available for servers and scripts; it can run under systemd and shares configuration and statistics with the app.

The app installs new versions automatically. It downloads an update in the background, verifies its signature, and only restarts after no requests have run for five minutes. On the server, just run llm-bridge update.

Set up in four steps

1. Start the app. On first launch, it creates a configuration with the LangDock EU upstream and the alias haiku.
2. Under “Upstreams,” enter your API key and save.
3. Click “Fetch models” and select the model ID your workspace actually offers. The default ID is only an assumption.
4. Add an OpenAI-compatible provider in the tool:

LLM Bridge generates the access token itself the first time it launches. Select “Test” under “Models” to check the connection; the test costs a single token.

The proxy only listens on your machine

A program that manages your API key should have a small attack surface. LLM Bridge listens only on 127.0.0.1 and requires the access token. It rejects requests from websites in your browser so no open page can secretly use your proxy. Upstreams must use https. LLM Bridge stores your API key in your system’s keychain and never writes it to the configuration file, which only your user account can read anyway.

Download and source code

The macOS packages are signed with an Apple Developer ID and notarized. The Windows and Linux packages are unsigned, but include a file with SHA256 checksums that you can use to verify the download. The source code is open source and licensed under MIT.

You can find the packages at github.com/eqms/llm-bridge/releases, the code at github.com/eqms/llm-bridge. Download the package for your system, enter your key, and send your first request from MacWhisper to haiku.

Created by Martin Schmid, with support from Claude Sonnet 5.5 (revised with Claude Opus 5.5) and approved after my own content review. Our Notices and Disclaimer.