levent-dogan/zcode-nvidia-nim-fix

This project is a small local fix for a specific compatibility problem between a code editor tool called ZCode and NVIDIA NIM, which is NVIDIA's service for running chat AI models through an OpenAI style API. Normally you would connect ZCode straight to NVIDIA NIM, but ZCode sends an extra field called extra_body that NVIDIA NIM does not accept, so requests fail with an error about an unsupported parameter. The fix works by running a small local proxy program on your own Windows computer. Instead of ZCode talking to NVIDIA directly, it talks to this proxy at a local web address, and the proxy strips out the fields NVIDIA does not support before forwarding a clean, compatible request on to NVIDIA NIM's real servers. Standard fields like the chosen model, the conversation messages, and streaming settings are kept and passed through untouched. The proxy supports two ways of handling API keys. In one mode, a single NVIDIA API key is set once and shared across every ZCode provider you configure. In the other mode, each ZCode provider can carry its own separate NVIDIA API key, and the proxy simply forwards whichever key ZCode sends, which is useful if you want to spread requests across several different NVIDIA accounts. The tool is not limited to one specific AI model, since it forwards whatever model name ZCode requests, as long as that model is available through your NVIDIA NIM account. Setup requires Windows 10 or 11, PowerShell, and Python 3.10 or newer. There is no extra Python package to install, since the proxy relies only on Python's built in libraries. Configuration is done through a provided launcher script, and the project includes a health check address and example requests for confirming everything is working correctly. The README is explicit that no NVIDIA API keys are stored in the repository itself, and the proxy avoids printing keys or full message content to its logs.