Claude Code, but make it someone else's bill

What it does

Free Claude Code is a local proxy that intercepts Claude Code’s Anthropic Messages API requests and translates them for 17 other providers. You keep using the official Claude Code CLI or VS Code extension, but the actual inference happens on NVIDIA NIM, OpenRouter, Gemini, Ollama, or another backend of your choice. It handles the protocol mismatch—translating between Anthropic’s Messages API and various OpenAI-compatible or provider-specific endpoints—so the client remains none the wiser.

The interesting bit

The per-model routing is the clever part: you can send Opus-tier requests to one provider, Sonnet to another, and Haiku to a third, all through the same proxy. It also exposes a local admin UI for swapping API keys without touching config files, and bundles optional Discord or Telegram bots plus voice transcription so you can run coding sessions from your phone.

Key highlights

• 17 provider backends including free tiers (OpenRouter, Gemini, DeepSeek, Ollama, and others)
• Per-tier model routing: MODEL_OPUS, MODEL_SONNET, and MODEL_HAIKU can each point to different providers
• Native /model picker integration in Claude Code via the proxy’s /v1/models endpoint
• Local admin UI at /admin for key management and validation (loopback-only)
• Optional chat bots and voice-note transcription via local Whisper or NVIDIA NIM

Caveats

• The admin UI is restricted to loopback, so remote management isn’t an option.
• Claude Code must opt in to Gateway model discovery for the native model picker to populate.
• Provider API fragmentation is your problem: some backends use Anthropic-compatible endpoints while others use OpenAI-compatible ones, and the README notes non-standard fields may need special handling.

Verdict

Good for developers who want Claude Code’s agentic workflow but prefer local, free, or enterprise-backed inference. Skip it if you’re happy paying Anthropic directly or don’t want to babysit another local proxy.