cli-proxy

OpenAI-compatible /v1/chat/completions at http://127.0.0.1:7890/v1. Routes Max-quota OAuth CLIs — no API keys burned (extra-usage off).

Model names

cli/<tool>[/<model>] — tools: claude, gemini, qwen, kilo, opencode. E.g. cli/claude/sonnet, cli/claude/haiku, cli/gemini.
minimax[/<model>] — direct MiniMax API (default MiniMax-M2.7).
- HTTP providers: chutes, openrouter, cerebras, tokenrouter[/<model>], stdcmpt[/<model>] (tokenrouter is OpenAI-compat at api.tokenrouter.com; no verified default model — caller must name one. stdcmpt is OpenAI-compat at api.stdcmpt.com/v1; default model StandardCompute).
pi/<alias>[/<effort>] — pi CLI multi-provider. Current alias: minimax-m3 → minimax/MiniMax-M3.
smart — pool alias (priority failover, first success wins): cli/claude/fable/high → cli/claude/opus/high → pi/minimax-m3/high.
fast — pool alias (priority failover): pi/minimax-m3/no-think → cli/claude/sonnet/no-think.

Effort levels: non, no-think (0 tokens), low, med, high, xhigh, max. For pi, effort maps to :thinking suffix.