Arbiter banner
Ujwal397 Ujwal397

Arbiter

AI community

Description

A local proxy that lets Claude Code use NVIDIA NIM models instead of Anthropic's API. Features task-aware model routing, three-tier fallback, and an interactive model selection menu.

Installation

This entry records only its repository, not the path inside it, so there is no exact command to give. Open the source below and copy the folder into ~/.claude/skills/, or the file into ~/.claude/agents/.

README

⚖️ Arbiter

Run Claude Code on NVIDIA's free cloud models — zero cost, no hardware needed.

[![License: MIT](https://img.shields.io/badge/License-MIT-blue.svg)](LICENSE) [![Python](https://img.shields.io/badge/Python-3.10%2B-blue?logo=python&logoColor=white)](https://python.org) [![FastAPI](https://img.shields.io/badge/FastAPI-0.100%2B-009688?logo=fastapi&logoColor=white)](https://fastapi.tiangolo.com) [![Platform](https://img.shields.io/badge/Launcher-Windows-0078D6?logo=windows&logoColor=white)](https://github.com/Ujwal397/Arbiter)


Claude Code  ──▶  Arbiter (localhost:4005)  ──▶  NVIDIA NIM Cloud  ──▶  Free Models

Claude Code only works with Anthropic's paid API by default. Arbiter breaks that lock — it sits between Claude Code and NVIDIA's free cloud API, transparently routing every request to models like **Kimi K2 0905**, **Mistral Large 3**, and **Llama 3.3 70B** at **zero cost**.

[**Get started →**](#-quick-start) · [Model availability](#-known-model-availability) · [Configuration](#️-configuration)


✦ Why Arbiter?

Without Arbiter With Arbiter
Claude Code only works with Anthropic's paid API Routes through NVIDIA's free cloud models
You pay per token — costs stack up fast Completely free with an NVIDIA API key
Limited to models Anthropic offers Access models like Kimi K2, Mistral Large 3, Llama 3.3
Large models require enterprise hardware locally Runs in the cloud — nothing to install beyond Python

**NVIDIA NIM is free.** Get an API key at [build.nvidia.com](https://build.nvidia.com) — no credit card, no per-token charges.


✦ Features

🧭 Task-aware model routing

Arbiter reads your last two messages and classifies the task before dispatching. Heavy work goes to the most capable model; quick questions go to the fastest one — automatically.

| Task type | Triggers on | Model tier | |:---|:---