9Router is a free, open-source local proxy server that connects AI coding tools to dozens of model providers through a single, unified endpoint. It is built for developers who juggle several AI coding assistants and want uninterrupted access without constantly hitting rate limits or overpaying for tokens. Instead of locking you into one vendor, 9Router sits on your machine and intelligently routes each request to whichever provider is available and most cost-effective at that moment.
How it works
9Router installs as a global npm package (npm install -g 9router) and runs locally on localhost:20128, exposing an OpenAI-compatible API endpoint at /v1 along with a built-in dashboard. Any tool that can point to an OpenAI-style base URL — Claude Code, Cursor, Cline, Continue, Copilot, Codex, Antigravity, and more — can send requests to 9Router. It translates those requests into each provider's native format, forwards them, and translates the responses back, handling formats across OpenAI, Claude, Gemini, Cursor, Kiro, Vertex, and Ollama.
Smart three-tier fallback
The core feature is a quota-aware fallback system organized into three tiers. The subscription tier uses premium plans you already pay for, such as Claude Pro/Max, Codex, and GitHub Copilot. The cheap tier covers low-cost options like GLM and MiniMax. The free tier taps no-cost services such as Kiro AI, OpenCode Free, and Vertex AI credits. When one tier exhausts its quota, requests automatically cascade to the next, giving effectively zero-downtime operation during long coding sessions or tight deadlines.
Token optimization with RTK
9Router includes RTK, a compression pipeline that automatically detects tool output such as git diffs, grep results, and file listings, then losslessly compresses that content before sending it to the model. This reduces token consumption by roughly 20–40% per request, lowering costs on paid tiers and stretching free quotas further, all transparently to the connected tool.
Key capabilities
Beyond routing and fallback, 9Router offers real-time quota tracking with reset countdowns, multi-account support per provider with round-robin routing, automatic OAuth token refresh, custom model combinations, request logging and debug mode, usage analytics with cost estimation, and cloud synchronization across devices. It supports 40+ providers and 100+ models in total, spanning Claude, GPT, Gemini, GLM, MiniMax, DeepSeek, Qwen, and region-specific variants. Deployment is flexible: localhost, a VPS, Docker, or Cloudflare Workers.
Who it is for
9Router targets developers who use multiple AI coding tools at once, need to optimize cost across existing subscriptions, require reliable access during critical work, or simply want to experiment with many models without managing separate accounts and API keys for each. Teams running coding agents in messaging platforms like Slack, Telegram, or WhatsApp can also route those integrations through 9Router.
Pricing and licensing
9Router itself is completely free and open-source under the MIT license, and it runs entirely on your own machine. It never charges you anything — you pay providers directly under their own pricing. Subscriptions stay the same, cheap tiers are billed per usage, and free providers genuinely cost nothing. Any cost figure shown in the dashboard is a reference comparison, not actual billing.
Limitations
Because 9Router depends on third-party providers, free-tier availability can change over time; for example, some previously free OAuth tiers have been discontinued or moved to paid plans. Certain free tiers also carry provider-side risks, such as account restrictions if used outside their intended client. As a local proxy it requires a one-time setup and basic familiarity with command-line tools, but once configured it runs quietly in the background.
