alange/tinyclaude

TinyClaude is a proxy server that sits between your coding tools and the Claude API, spreading requests across several accounts and providers so you do not hit rate limits. Instead of calling Claude directly, you point tools like Claude Code at this proxy, and it decides which account or provider handles each request. The proxy supports many backends: Claude through OAuth or the Claude Console API, Amazon Bedrock, Google Vertex AI, OpenAI-compatible services, and several other providers including OpenRouter, Ollama, and xAI. You can group these into named fallback chains, called combos, so if one account runs out of capacity the proxy automatically switches to the next one in the chain, then switches back once the original account's limits reset. To cut down on wasted tokens, the proxy compresses large tool-result payloads before sending them along, and it can replay cached requests on a schedule to keep the prompt cache active, which can save on repeated processing costs. It also tracks token usage, latency, and errors for every request and shows this information on a dashboard you can view in a browser at localhost:8080/dashboard. Getting started is meant to be simple: install it globally through bun or npm, or run it without installing using npx, then add an account with a single command and set an environment variable so your existing tools send requests through the proxy instead of directly to Claude. Pre-built binaries are available for people who do not want to install a package manager, and the project also supports running inside Docker or as a systemd service, along with optional API-key protection if you want to restrict who can use the proxy. The project builds on ideas from a few similar open-source tools and credits them directly in the README. It is licensed under MIT, so you are free to use, modify, and redistribute it with minimal restriction.