Claude Code, But the API Bill Goes Elsewhere

What it does

free-claude-code sits between the official Claude Code CLI or VSCode extension and a third-party LLM backend, translating Anthropic API requests into OpenAI-compatible calls for NVIDIA NIM, OpenRouter, or LM Studio. You keep the official client unchanged; just point it at the proxy. The project also intercepts minor internal requests—like quota probes and title generation—to answer them locally without burning provider quota.

The interesting bit

The proxy maps Claude’s Opus, Sonnet, and Haiku requests to different models and providers independently, so you can mix a local LM Studio instance with a free NVIDIA NIM endpoint in the same session. It also handles the messy translation layer: heuristic parsing turns raw text tool calls into structured blocks, and thinking tags or reasoning_content fields are rewritten into native Claude thinking tokens so the UI renders them correctly.

Key highlights

• Drop-in replacement: two environment variables, zero modifications to the official Claude Code client
• Three backends: NVIDIA NIM (40 req/min free), OpenRouter (200+ models), or fully offline LM Studio
• Intercepts five categories of trivial requests locally—quota probes, title generation, prefix detection, suggestions, and filepath extraction—to save quota and cut latency
• Discord and Telegram bots with tree-based threading, session persistence, and live streaming of thinking tokens and tool calls
• Extensible provider and messaging platform interfaces via BaseProvider and MessagingPlatform ABCs

Caveats

• VSCode users may still be redirected to an Anthropic credit purchase page on first setup; the README says to authorize and ignore the billing prompt.
• NVIDIA NIM’s free tier is capped at 40 requests per minute, and OpenRouter limits vary by model.
• You need to run the proxy server locally alongside Claude Code, so it adds one more moving part to your stack.

Verdict

Ideal for developers who want Claude Code’s agentic workflow without Anthropic API costs, or who need a fully local option via LM Studio. Skip it if you’re happy paying Anthropic rates and don’t want another local service to maintain.