I just released v0.0.2 of jetbrains-ai-gemini-local-proxy.
It's a local proxy that helps JetBrains AI Assistant work with Google Gemini when you bring your own API key (BYOK).
Some background
Some time ago I shared the first version. It fixed the "Something went wrong" error that showed up because the IDE was adding an OpenAIAPI/ prefix to the model name, and Gemini rejected the request. I kept using it every day, and I found a few more things to fix.
What's new in v0.0.2
A new bug, fixed. On newer IDE builds, the AI Assistant's HTTP client (ktor) tries to upgrade the connection to HTTP/2 (Upgrade: h2c) on every request. Some servers drop the request body when this happens, so the proxy got an empty request and returned a 400 error. In the IDE this looked like the same "Something went wrong" again. Switching the server to the h11 parser fixed it.
Better logging. Now you can see every request, how long it takes, the response status, and the streamed chunks. So when something breaks, it's much easier to understand why.
Cleaner setup. I went back to a simple uvicorn config and removed a dependency I didn't need.
How to use it
You still just need two steps:
docker compose up -d
Then add your GEMINI_API_KEY and point the IDE's BYOK endpoint at http://localhost:5003.
Tested on IntelliJ IDEA 2026.1.3.
If you have the same problem, feel free to try it: https://github.com/agitrubard/jetbrains-ai-gemini-local-proxy
Top comments (0)