free · runs locally · open source

One Endpoint for
All AI Providers

9Router sits between your coding tools and 40+ AI providers. Its RTK token saver trims 20–40% of every request, and when a quota runs dry it falls back from subscription to cheap to free models — so you never stop coding.

Get startedStar on GitHub
  • 20–40% tokens saved
  • 40+ providers
  • 100+ models
  • $0 · MIT licensed
Claude CodeCodexCursorCline
9ROUTER
T1Subscription

Claude Max · Copilot · Codex

T2Cheap

GLM $0.6/1M · MiniMax $0.2/1M

T3Free

Kiro · OpenCode Free · Vertex

// quick start

Get started in 30 seconds

No cloud signup, no credit card. 9Router runs entirely on your machine.

  1. 1

    Run 9Router

    One command starts the router and local dashboard. No account, no config files.

  2. 2

    Connect providers

    Open the dashboard at localhost:20128 and link subscriptions, API keys, or free providers.

  3. 3

    Point your tools at it

    Set your CLI’s base URL to http://localhost:20128/v1 and keep coding.

~ terminal

$ npx 9router

› Starting 9Router…

› API ready on http://localhost:20128/v1

› Dashboard: http://localhost:20128/dashboard

✓ Ready to route

then point your tool at it:

# ~/.claude/settings.json
{
  "env": {
    "ANTHROPIC_BASE_URL": "http://localhost:20128",
    "ANTHROPIC_AUTH_TOKEN": "<key from dashboard>",
    "ANTHROPIC_MODEL": "kr/claude-sonnet-4.5"
  }
}

// how it works

Route every request to the cheapest model that can handle it

One local process between your tools and the providers. It saves tokens on the way in and never lets a rate limit stop you.

1. Your tools send requests

Claude Code, Codex, Cursor, Cline — anything that lets you change the base URL talks to 9Router instead of a provider.

2. 9Router compresses & translates

RTK trims bulky tool output (git diff, grep, ls) before it reaches the model and translates between OpenAI and Claude formats.

3. The best available tier answers

9Router tracks quota per provider and routes each request to the highest tier that still has room.

fallback cascade

zero downtime

TIER 1SubscriptionClaude Max · GitHub Copilot · Codexuse every bit of quota before it resets
quota exhausted → fall through to next tier
TIER 2CheapGLM $0.6/1M · MiniMax $0.2/1Mpennies per million tokens
budget limit → fall through to next tier
TIER 3FreeKiro · OpenCode Free · Vertex creditsalways-on safety net, $0

// features

Everything a router should do

Built for developers who live in AI coding tools and refuse to waste tokens.

Unified Endpoint
One OpenAI-compatible URL for every provider. Change the base URL once, switch models freely.
RTK Token Saver
Compresses tool output — git diff, grep, ls — before it reaches the model. 20–40% fewer tokens per request.
Model Fallback
Quota exhausted or provider down? Requests cascade from subscription to cheap to free tiers automatically.
Usage Tracking
Per-provider quota, cost, and token analytics so you know exactly where every request went.
OAuth & API Keys
Connect with OAuth or paste a key. Credentials live in one local vault on your machine.
Multi-Account
Round-robin across multiple accounts per provider to stretch every subscription’s quota.
CLI Support
Claude Code, Codex, Cursor, Cline, RooCode, OpenCode — anything with a configurable endpoint.
Dashboard
A local web dashboard for providers, keys, model routing, and live traffic at localhost:20128.

works with the tools you already use

  • Claude Code
  • Codex
  • Cursor
  • Cline
  • RooCode
  • OpenCode
  • GitHub Copilot
  • Gemini CLI
  • Antigravity
  • OpenClaw

// faq

Frequently asked questions

Stop paying for tokens
you don’t need to spend

One command, and every AI tool you use gets cheaper, more resilient, and quota-proof.

$npx 9router
Read the quick startView on GitHub