ciroautuori/openvidia — reverse-engineered prompt
Reverse engineered prompt
Build me a small desktop app called OpenVidia that runs a local proxy for the NVIDIA NIM API. I want to paste in several NVIDIA API keys once, have them saved in the app, and then have the proxy automatically rotate to the next key whenever one is rate limited, invalid, or errors out. It should keep using the last good key so my tools do not keep retrying dead ones. The proxy should be OpenAI compatible on localhost port 3940, support normal requests plus streaming responses, and optionally let me force one model for every request.
I also want a clean dark UI where I can add or remove keys, start the proxy, see request counts, rotations, successes, and a live log showing which key and status were used. Please keep it loopback only so it never exposes my keys to the network, and make it auto start from saved settings when the app opens. If it makes sense, include the opencode setup too. Look up current docs online if you need to.
Want more depth? Deep Reverse