✨ Features

• Every Ollama endpoint: chat, generate, embeddings, tags, ps, show, pull, create, copy, delete, blobs. Each one is rewritten into the LM Studio call that matches it.
• Ollama-style model names: LM Studio ids appear under the names Ollama clients expect. /api/create and /api/copy add your own aliases on top.
• Streaming with real token counts: SSE, cancelled when the client hangs up. Non-streaming chat/generate report LM Studio's own eval_count / eval_duration / prompt_eval_*. Streaming reports them on the native chat path (--use-native-chat or --native-chat-streaming); the default path estimates them, because LM Studio's v0 stream carries no stats chunk.
• Reasoning: think / reasoning_effort map onto LM Studio's reasoning tiers, the model's reasoning comes back in thinking. A reasoning-capable model thinks by default when think is omitted, matching Ollama.
• Context window: options.num_ctx reloads the model at that size before inference, since LM Studio fixes context at load time. An instance already loaded at the right size is reused. --default-context-length / OLLAMA_CONTEXT_LENGTH covers requests that omit it.
• Passthrough: /v1/* and /api/v1/* go straight to LM Studio, so Anthropic Messages (/v1/messages) and OpenAI Responses (/v1/responses) work as they are. On the way through, the proxy remaps the model name and applies two compatibility shims: response_format: json_object, base64 embeddings.
• Web tools: /api/web_fetch returns {title, content, links} with HTML rendered to markdown. /api/web_search forwards to whichever provider --search-url names. Neither needs an ollama.com account.
• Auth: --api-key / OLLAMA_API_KEY puts a Bearer gate on every route, GET /api/version and CORS preflight aside. Leave it unset and the proxy stays open.

📦 Install

cargo install ollama-lmstudio-proxy

Pre-built binaries live on the Releases page. To build the tip of main: cargo install --git https://github.com/uwuclxdy/ollama-lmstudio-proxy.git.

🚀 Quick start

Requires LM Studio 0.4.0+, the release that shipped the /api/v1 REST API the proxy resolves models through.

# defaults: binds 0.0.0.0:11434, talks to LM Studio on :1234
ollama-lmstudio-proxy

# common overrides
ollama-lmstudio-proxy \
  --listen 0.0.0.0:11434 \
  --lmstudio-url http://localhost:1234 \
  --load-timeout-seconds 30

[!WARNING]
Stop any running Ollama server first. It holds the same port.

At most once a day, startup asks GitHub for the newest release tag and logs a line if yours is older. No flag turns that off.

Then set your client's Ollama host to http://localhost:11434. Anthropic and OpenAI clients can use that address too: LM Studio serves /v1/messages and /v1/responses itself, the proxy passes those through.

Every flag: Configuration.

🤝 Alternatives

Pick by the protocol your client speaks:

Use this one when your client speaks Ollama. For OpenAI-native clients, LiteLLM or llama-swap fit better.

❓ FAQ

How do I use LM Studio with the Ollama API?
Run this proxy on port 11434, start LM Studio on 1234, point your Ollama client at http://localhost:11434.

Can I use Claude Code with LM Studio?
Yes. Claude Code speaks the Anthropic Messages API, which LM Studio serves at /v1/messages. Set ANTHROPIC_BASE_URL=http://localhost:11434 and the proxy forwards it.

How do I connect VSCode Copilot or Open WebUI to LM Studio?
Set the client's Ollama host to http://localhost:11434. Model names show up in Ollama style; streaming, reasoning, tool calls all run when the model supports them.

Do I need to stop Ollama first?
Yes. Ollama and the proxy both want port 11434.

📚 Documentation

The wiki is the full technical reference:

🛠️ Development

cargo test            # run the test suite
cargo run -- --help   # see every flag

api-docs/ is the source of truth for upstream API behavior. Issues and pull requests are welcome.

📄 License

MIT or Apache-2.0, at your option. Contributions land under both.