LLM Proxy

LLM Proxy is a local TypeScript proxy manager for routing Claude-compatible clients to multiple configured LLM providers.

It runs one resident management server and an independent child proxy for each provider. Clients keep using stable local Anthropic or OpenAI-compatible base paths while each provider can define its own upstream endpoints, credentials, model mappings, listen address, and log level.

What it provides

• Anthropic Messages, OpenAI Chat Completions, and OpenAI Responses routing;
• streaming-safe forwarding of upstream response headers and bodies;
• model-name rewriting when a provider mapping exists;
• HTTP and HTTPS upstream support;
• a browser-based management console for starting and stopping providers;
• provider logs, level filtering, and newest-first prompt history;
• local PID, log, and JSONL history storage under LLM_PROXY_HOME;
• macOS LaunchAgent installation, upgrades, rollback, and service checks.

The project is designed to keep provider-specific configuration local and explicit. A TOML file defines endpoints and model mappings, while the local console exposes the provider state and the client base URLs needed to connect tools such as Claude Code or Codex.

Repository

The source code and development instructions are available on GitHub:

• github.com/yuanstrong/llm-proxy