A local proxy that squeezes your AI coding bill

What it does

9Router runs as a local OpenAI-compatible API endpoint at localhost:20128. Your AI coding tools—Claude Code, Cursor, Cline, Copilot, and others—point at it instead of a single provider. It forwards requests through a tiered chain (subscription accounts first, then cheap APIs, then free tiers) and translates between OpenAI and Claude formats on the fly.

The interesting bit

The RTK “Token Saver” is the quiet workhorse: it auto-compresses tool_result payloads like git diffs and grep output, which the README claims cuts 20–40% of tokens per request. That’s the kind of boring plumbing that actually moves the cost needle when you’re iterating in a tight loop.

Key highlights

• Auto-fallback chain: subscription → cheap → free providers with zero manual switching.
• Quota tracking and multi-account round-robin to squeeze value out of paid tiers before they reset.
• Format translation between OpenAI and Claude APIs so tools don’t care who’s behind the proxy.
• Bundled dashboard for provider management and key rotation.
• Supports 40+ providers including Kiro AI, OpenCode Free, Vertex AI, OpenRouter, GLM, and MiniMax.

Caveats

• The open-source repository is named 9router-app and marked private, while the public npm package is 9router; the README explicitly expects source or Docker execution for local development, so the repo and the distributed binary may not be in sync.
• Provider availability shifts: the README notes that iFlow, Qwen, and Gemini CLI free tiers were discontinued in 2026, leaving Kiro, OpenCode Free, and Vertex as the primary free options.

Verdict

If you’re running multiple AI coding tools and bleeding money on overlapping subscriptions, this is a pragmatic cost-control layer. If you already have a single unlimited enterprise API key, you don’t need another middleman.