Found a way to use Claude Code with NVIDIA Cloud models.
Posted by Amazing-Candle-1430@reddit | LocalLLaMA | View on Reddit | 3 comments
Just wanted to share a project I've been refining: "Arbiter." It's a local proxy that translates Anthropic's streaming format to OpenAI's format in real-time.
The main goal was to make Claude Code’s agentic features (tool use, disk access, terminal control) compatible with free cloud providers. I added some logic to handle the specific crash conditions that usually happen when people try to use LiteLLM or similar proxies directly with the Claude CLI.
Current status:
- Kimi K2 (0905) and Mistral Large 3 are confirmed stable.
- Llama 3.3 70B works for most logic tasks.
- Includes a parallel availability prober at startup.
I'll post the repository info in the comments for those who want to check it out
Worried_Drama151@reddit
So if I want a piece of shit like Kimi to destroy my codebase this is a way?! Pi + qwen3.6
Amazing-Candle-1430@reddit (OP)
You have access to other models too, not just kimi. evem qwen3.6
Shoddy_Cook_864@reddit
Github link: https://github.com/Ujwal397/Arbiter/