LMRouter is an open-source, all-in-one API router that gives AI developers a single, unified way to connect to the best model providers.
With LMRouter, you don’t need to create accounts on multiple platforms or juggle different API keys. A single API key unlocks access to any provider, any model, any service, all through one interface.
It’s an open-source alternative to OpenRouter, but goes far beyond language models. LMRouter supports multiple modalities and services out of the box:
- Language: Chat Completions, Responses API, Anthropic Messages
- Image: Image generation and editing
- Audio: Speech-to-text and audio models
- Video: Video generation (coming soon)
- Embeddings: Semantic search and RAG
- Search: Jina, Exa, and other web search APIs (coming soon)
- Code: Execution with interpreters such as e2b (coming soon)
LMRouter is still under active development. When the full service is launched, it will be available at lmrouter.com.
For now, you can try out LMRouter by connecting Claude Code to the demo server. This allows Claude Code to run on any model available through OpenRouter, not just Claude itself.
The demo environment proxies requests to OpenRouter and supports both the OpenAI Chat Completions API and the Anthropic Messages API. It runs in BYOK (Bring Your Own Key) mode, which means you’ll need to supply your own OpenRouter API key.
Connect Claude Code to LMRouter
ANTHROPIC_BASE_URL=https://api.lmrouter.com/v1/anthropic \
ANTHROPIC_AUTH_TOKEN=BYOK:<YOUR_OPENROUTER_API_KEY> \
ANTHROPIC_MODEL=<YOUR_MODEL_NAME> claudeTo setup LMRouter locally, make sure you have Node.js and pnpm installed, then follow these steps:
git clone https://github.com/LMRouter/lmrouter
cd lmrouter
pnpm i
cp config/config.example.yaml config/config.yamlEdit config/config.yaml to set up your own provider and model configurations.
# Run the server in development mode
pnpm dev
# Run the server in production mode
pnpm build
pnpm startAlternatively, you can use pnpx to run the server:
pnpx @lmrouter/cli /path/to/config.yamlLMRouter provides drop-in compatibility with existing APIs, so applications that already work with OpenAI or Anthropic can integrate without changes:
OpenAI API
/v1/openai/v1/chat/completions — Chat Completions
/v1/openai/v1/images/generations — Image generation
/v1/openai/v1/images/edits — Image editing
/v1/openai/v1/embeddings — Embeddings
/v1/openai/v1/responses — Responses
/v1/openai/v1/audio/speech — Audio speech
/v1/openai/v1/audio/transcriptions — Audio transcriptions
/v1/openai/v1/audio/translations — Audio translations
/v1/openai/v1/models — List available models
Anthropic API
/v1/anthropic/v1/messages — Messages
/v1/anthropic/v1/models — List available models
LMRouter connects to the broader AI ecosystem, with built-in support for:
- OpenAI-compatible — OpenAI, Google, Perplexity, Fireworks, Groq, Cerebras, etc.
- Anthropic-compatible — Anthropic, Amazon Bedrock, etc.
- Fireworks — FLUX.1 [schnell] FP8, FLUX.1 Kontext Pro, FLUX.1 Kontext Max
- Google — Imagen, Gemini image generation, etc.
LMRouter is built on Hono and runs in multiple environments:
- Node.js — Standard server deployment
- Cloudflare Workers — Lightweight edge deployment at scale
We believe LMRouter should be built by the community, for the community. Every contribution, whether it’s fixing a bug, adding a feature, improving documentation, or sharing feedback, helps make LMRouter stronger.
Checkout our Kanban for the latest roadmap.
- Fork the repository
- Create a feature branch (
git checkout -b feature/my-feature) - Build your changes, code, docs, or tests
- Verify everything works as expected
- Submit a pull request and join the conversation
No contribution is too small, even a typo fix helps. And if you have big ideas, we’d love to hear them.
Tip: If you’re not sure where to start, check out the open issues labeled
good first issue.
LMRouter is licensed under the MIT License. See the LICENSE file for details.