Drop-in OpenAI-compatible proxy. Compress before upstream. Meter every call.
Shrink prompts before they hit OpenAI / OpenRouter / Anthropic.
Original vs compressed tokens and savings % on every request.
Swap base URL: Cursor, Continue, Cline, Roo, Hermes, Aider, Open WebUI, SDKs, LangChain. Full matrix in the docs.
Lossless / balanced / aggressive. Your provider keys stay yours.