// the find
decolua/9router
Unlimited FREE AI coding. Connect Claude Code, Codex, Cursor, Cline, Copilot, Antigravity to FREE Claude/GPT/Gemini via 40+ providers. Auto-fallback, RTK -40% tokens, never hit limits.
A local OpenAI-compatible proxy that sits between AI coding tools (Claude Code, Cursor, Cline, etc.) and 40+ LLM providers, handling format translation, quota tracking, and automatic tier fallback. The genuinely useful piece is RTK, which compresses tool_result content (git diff, grep, ls output) before it reaches the LLM. For developers already paying for AI subscriptions who want fallback coverage or to stretch their quota without reconfiguring their tools.
RTK token compression is the real differentiator: it intercepts tool_result blocks, applies lossless filters per content type (git-diff, grep, tree, etc.), and fails open — if the filter breaks or makes output larger, it silently passes the original through. That's the right design. The executor-per-provider pattern under open-sse/executors/ is auditable: each provider's OAuth and API translation logic lives in one file, which makes debugging provider-specific breakage tractable. Format translation across OpenAI, Claude, Gemini, Cursor, Kiro, and Vertex at the SSE streaming layer is non-trivial work and it's done once here instead of per-tool. The 3-tier fallback with quota tracking solves a real problem: subscription quotas reset on different schedules (5-hour, daily, weekly, monthly) and this unifies them into one config.
The 'FREE unlimited' headline is collapsing: iFlow, Qwen Code, and Gemini CLI are all gone as of mid-2026, Kiro is now ~50 credits/month, OpenCode Free's model list 'fluctuates over time', and Vertex's $300 credit is one-time for new GCP accounts. The free tier story in the README and the actual free tier story are drifting apart. This is credential proxying through subscription OAuth flows for Claude Code, Codex, and Cursor — those providers' ToS almost certainly prohibit using extracted session tokens through third-party proxies; users adopt this risk without much warning. It's plain JavaScript with no TypeScript and no visible test suite, which matters here because SSE stream transformation bugs across a dozen provider formats are exactly the kind of thing that silently corrupts a response. The 'cost display' is a UX failure by construction: the README requires three separate FAQ entries explaining why the dashboard shows $290 when your actual bill is $0 — if you need that much text to explain a number, the metric is wrong.