• about 1 min read

Ollama Proxy — Give VSCode Copilot Any LLM Backend

A local proxy that pretends to be Ollama so tools locked to the Ollama protocol (VSCode Copilot, notably) can use cloud LLM providers instead — no local models burning VRAM. Forked from an existing open-source proxy and extended with several fixes: force-streaming upstream to dodge Cloudflare 524 timeouts (reassembled non-streaming for the client), polling-based config hot reload that actually works inside Docker, Docker packaging, and completion of the /api/version and related endpoints that newer VS Code builds require to even display the provider. A few changes have been merged back upstream. Details in the Chinese version.

jcbl1
jcbl1 Software Engineer