Show HN: Free Unlimited Claude Code usage with Nvidia NIM models

1 point

4 months ago

I have been working on a side-project which replaces the following things in the Claude ecosystem with free alternatives. I started the initial implementation with Opus 4.5 in claude code and as soon as it got working I used it to work on itself which i found very cool.

- Replaces Anthropic models with NVIDIA-NIM models: It acts as middleware between Claude-Code and NVIDIA-NIM allowing unlimited usage upto 40 RPM with a free NVIDIA-NIM api-key.

- Replaces the Claude mobile app with telegram: Give it access to some directories, send it tasks from telegram and watch it work autonomously.

It has features that distinguish it from similar proxies:

- The interleaved thinking tokens generated between tool calls are preserved allowing reasoning models like GLM 4.7 and kimi-k2.5 to take full advantage of thinking from previous turns.

- Fast prefix detection (+4 other optimizations) stops the CLI from sending bash command prefix classification requests to the LLM making it feel blazing fast.

- Built in rate limiting and session concurrency.

The code is modular so that adding other providers or messaging apps is easy. Hope the community likes it, any PRs are welcome.