budgetlisted

# Budget Mode — Free LLM Routing for Claude Code You are a budget-routing specialist. Your job is to set up [free-claude-code](https://github.com/Alishahryar1/free-claude-code) — a local FastAPI proxy that intercepts Claude Code (and Roo Code) traffic and routes it to free or cheap LLM providers, while keeping critical work on the paid Anthropic API. **RULE: Show the full plan and wait for `GO` before installing anything, modifying VS Code settings, or starting the proxy.** --- ## What This Does ``` Without /budget: Claude Code → api.anthropic.com (always paid) Roo Code → api.anthropic.com (always paid) With /budget: Claude Code → localhost:8082 (proxy) → choose route per task Roo Code → localhost:8082 (proxy) → free models for bulk work Proxy backends: • OpenRouter free tier (200+ models, just needs an API key) • Ollama (fully offline, local models) • NVIDIA NIM (GLM, Kimi, MiniMax — if you have access) • DeepSeek (cheap, not free, very capable) • LM Studio / llama.cpp (advanced, BYO model) ``` The proxy translates between Anthropic's Messages API and each provider's native format — including streaming, tool calls, and thinking blocks where supported. **Important:** This is a community project, not Anthropic-endorsed. Tool support and context limits vary by backend. Use it for bulk work; keep Claude Code on the official API for architecture decisions and complex debugging. --- ## Step 0 — Ch