The 9router proxy is the trick that makes one local endpoint feel like every AI provider at once: your coding tool talks to http://localhost:20128/v1, and 9Router translates the request, picks the best available model across 40+ providers, and falls back automatically when one runs dry. Understanding the proxy layer is understanding the whole product — here’s how it works, per the repo and my own daily setup.
Short answer
- A proxy = software that sits between your tool and the model providers — 9Router runs it entirely on your machine.
- Your tools see one endpoint (localhost:20128); 9Router handles format translation to each provider behind it.
- Three-tier auto-fallback per the docs: Subscription → Cheap → Free — “never stop coding, zero downtime”.
- Built-in economics: RTK compresses input tokens 20–40%; Caveman Mode claims up to 65% output savings; quotas tracked live.
What the 9router proxy actually does
Every AI coding tool speaks some provider’s dialect — Claude Code speaks Anthropic’s API, Codex and Cursor speak OpenAI’s. A proxy exploits that: point the tool’s base URL at 9Router instead of the real provider, and the tool carries on happily while 9Router decides what actually serves the request. The README documents exactly this — Claude Code’s anthropic_api_base set to the local endpoint, Codex’s OPENAI_BASE_URL exported the same way, Cline and friends configured as “OpenAI compatible”.
Because it runs locally, the trust model stays sane: your connected logins and keys live on your machine, not on someone’s hosted middleman. That’s the line that separates a proxy you control from a service you have to trust.
Inside the routing: fallback, translation, quotas
The routing logic is the value. 9Router’s documented three-tier fallback tries your subscription providers first (the logins you already pay for — Claude, Codex, GitHub, Cursor), steps down to cheap API-key providers when quotas run out, and lands on genuinely free tiers (Kiro’s monthly credits, OpenCode Free, Vertex trial credits) as the floor. The dashboard tracks quotas in real time with reset countdowns, so the switching isn’t guesswork.
Layered on top are the token economics: RTK Token Saver compresses tool outputs to cut input tokens by a documented 20–40% per request — the setting I flagged in my setup guide as the one most people miss — and Caveman Mode pushes output savings up to a claimed 65% via prompt injection. Route through one proxy and every tool inherits the savings at once.
🔥 Want this set up without the guesswork? A proxy-powered AI stack that cuts your model bills is exactly the kind of thing we set up together inside the AI Profit Boardroom — 3,700+ members, four live calls a week, daily tutorials, done-for-you templates and a 30-day roadmap. Prefer 1-on-1 help? Book a free SEO strategy session and we’ll map it out for your business.
9router proxy patterns worth running
| Pattern | How | Why it works |
|---|---|---|
| One endpoint, many tools | Point Claude Code, Codex and Cursor all at localhost:20128 | Every tool shares the same fallback and savings |
| Subscription squeezer | Connect existing logins as tier one | Extracts full value from plans you already pay for |
| Never-blocked coding | Let the free tier be the floor | A rate limit downgrades the model instead of stopping you |
| Two-router safety net | Pair with OmniRoute as a dropdown | Router-level redundancy — my daily setup |
That last pattern is the one I actually run — the full comparison and pairing logic is in 9Router vs OmniRoute.
The bottom line on the 9router proxy
The 9router proxy layer is the whole product in one idea: your tools talk to one local address, and everything hard — format translation, provider choice, fallback, compression, quota accounting — happens behind it, on your machine, for free. Wire your tools to the endpoint once and the economics of your entire AI coding stack change quietly and permanently.
FAQ: 9router proxy
What is the 9router proxy?
A local proxy server: your coding tools point at http://localhost:20128, and 9Router translates each request and routes it across 40+ providers with automatic fallback.
Does the proxy send my keys anywhere?
It runs entirely on your machine — connected logins and keys stay local, which is exactly why the open-source, run-it-yourself model matters.
How does the fallback work?
Three documented tiers — Subscription first, then cheap API providers, then free tiers — with live quota tracking deciding the switches.
What is RTK in 9Router?
The RTK Token Saver — compression of tool outputs that the docs credit with saving 20–40% of input tokens per request; Caveman Mode adds claimed output savings up to 65%.
Which tools can use the proxy?
Sixteen per the README — Claude Code, Codex, Cursor, Cline, Continue, Copilot, OpenClaw, OpenCode, Antigravity, Qwen Code and more.
Is a local proxy slower?
It adds a hop on your own machine — negligible next to model latency; the fallback usually saves far more time than the hop costs.
Next step: if you want one endpoint feeding every tool you run working for you this week, join the AI Profit Boardroom for the full walkthroughs and live help — or book a free SEO strategy session and I’ll point you at the fastest path for your situation.
About Julian Goldie: SEO agency owner with 10+ years in SEO, 394K+ subscribers on YouTube, a 100% job-success score on Upwork, 75K+ members across his communities, and author of a best-selling SEO book. He runs the AI Profit Boardroom community and offers a free SEO strategy session.
Related reading
- How To Use 9Router: Step-By-Step Setup
- 9Router Claude Code Setup: Free Models In Minutes
- How To Reduce Claude Code Token Usage By 80%
Last updated September 2026. This is the living guide to 9router proxy — it gets updated as the tools change.
