Model Proxy

Model Proxy lets you run Claude Code and Codex on any AI provider, not just one. It is a tiny macOS menu bar app that starts a local HTTP proxy and transparently routes each request to the right vendor based on the model ID. Point your CLI at localhost once, and everything keeps working.

Why developers use it:

• Use cheaper or faster models. Route Claude Code traffic to DeepSeek, MiniMax, Kimi, GLM, or any Anthropic-compatible endpoint, and keep your exact CLI setup. • See what you save. Model Proxy tracks token usage per model and shows your real cost next to the equivalent official price, so you know exactly how much each session saved. • Web search keeps working everywhere. Most third-party vendors do not support Claude's built-in web search. The Web Search Bridge runs the search through Brave, Google, or Tavily and feeds the results back, so the web_search tool works no matter which model you route to. • Watch every request live. A real-time traffic monitor shows each call, its model, token counts, and request type as it happens. Nothing is buffered. • Route by rule, toggle on the fly. Build model-mapping rules, enable or disable them without deleting, and let Model Proxy suggest mappings for new model IDs it observes. • Private by design. The proxy listens on localhost only. It never stores or logs your request and response bodies. Your API keys stay in a local config file on your Mac.

Built on SwiftNIO for low-latency streaming. Native menu bar app, macOS 15.6 and later.