A CLI proxy that intercepts Warp terminal AI requests and routes them to a local LLM without cloud exposure

Warp terminal routes all AI inference through its own cloud servers. Even after adding a custom inference endpoint, Warp rejects localhost URLs and requires users to expose their local Ollama or LMStudio instance to the public internet via ngrok or Cloudflare Tunnel. This means every AI-assisted terminal session leaks shell context, command history, and file paths to Warp's servers and a public tunnel endpoint, with no opt-out. A local socket-level proxy intercepts Warp's AI requests before they leave the machine, rewrites the destination to a local model endpoint, and returns the response in the same format Warp expects. Zero cloud exposure, zero tunneling, full Warp UX intact. The proxy runs as a background daemon, requires no Warp modification, and works with any OpenAI-compatible local backend (Ollama, LMStudio, llama.cpp, vLLM). Targets privacy-conscious developers, enterprise teams with data-handling constraints, and air-gapped environments where routing terminal context to a third-party cloud is not acceptable.