How To Use 9Router: Step-By-Step Setup (2026)

9Router reaches 573 models through the logins you already pay for, runs entirely on your machine, and falls back automatically when one runs dry. Here’s how to set it up in five steps and the token settings most people miss.

What you’re setting up

9Router is a free, open-source router that runs locally and reaches 573 models across 40+ providers — through the logins you already have.

That last part is what makes it different from most routers. It wires into your existing Claude, Codex, GitHub, Copilot and Cursor OAuth sessions, so you’re using capacity you already pay for rather than buying new API keys.

If one model runs dry or rate-limits, it automatically switches to another — including free tiers. Your coding doesn’t stop.

For the overview of what it is and how it compares, see the 9Router guide. This page is the step-by-step.

Step 1: install and start it

1. Install 9Router from the project’s repository — it’s a local install, not a service you sign up for.
2. Start the router. It runs on localhost, on your machine.
3. Confirm it’s up before connecting anything to it.

Nothing here goes to a third party. That’s worth saying plainly because the usual objection to routers is “I’m handing my keys to a middleman” — you aren’t. It’s localhost.

Step 2: connect your existing logins

This is the step that unlocks the value, and the one people skip because it feels too easy.

Connect as many as you have. The more sessions attached, the more headroom before anything rate-limits — and 9Router is multi-account, so several of the same provider is fine.

You’re not buying anything new here. You’re pooling what you already own into one endpoint.

Step 3: point your agent at it

9Router is universal — it works with OpenClaw, Hermes, Claude Code, OpenCode and anything else that speaks a standard API.

1. Open your agent’s model or provider settings.
2. Point it at the local 9Router endpoint instead of a cloud provider.
3. Pick a model from the list — there’ll be a lot of them.
4. Run something trivial first to confirm the wiring before you hand over real work.

For Hermes specifically, this is a profile change — the same pattern covered in the OmniRoute Hermes guide. Your memory, workflows and dashboard all carry on unchanged.

Step 4: turn on the token savings

9Router isn’t only about reaching free models — it actively reduces what you burn.

• RTK compression saves roughly 20–40% of your tokens.
• Usage tracking per subscription so you can actually see where your tokens are going.
• Automatic fallback to free models, so the expensive capacity only gets used when it’s needed.

That third one is the quiet winner. Most people burn premium capacity on work that never needed it. Letting the router push routine jobs to free models is a bigger saving than any compression setting.

Step 5: run it as a two-coder setup

The way I actually use it: a free AI coder inside my Agent OS with a dropdown for two open-source routers — 9Router and OmniRoute. Coder on the left, live preview on the right, saved builds in the workspace, full conversation history.

If one gets rate-limited, you flick the dropdown to the other. Two free coders, zero downtime. Full detail in the OmniRoute install guide.

If something isn’t working

• No models listed: your logins probably aren’t connected yet. That’s step 2.
• Auth keeps dropping: OAuth sessions expire — reconnect the provider rather than reinstalling.
• Nothing responds: check the router is actually running and nothing else holds the port.
• Hitting limits fast: turn RTK on and confirm fallback is enabled rather than pinning one model.
• Output quality is poor: you’ve fallen back to a free model. Switch models — that’s a dropdown, not a reinstall.

Is free-model code actually good enough?

Honest answer: it’s not Fable 5. But 99% of the time you don’t need Fable 5 to build something good, and we’ve shipped plenty with this setup.

The sensible split is to let the router push routine and experimental work to free capacity, and keep your premium subscription for the jobs that genuinely justify it. That’s where the saving comes from — not from pretending free matches paid.

FAQ

How do I use 9Router?

Install and start it locally, connect your existing Claude, Codex, GitHub, Copilot and Cursor logins, point your agent at the local endpoint, then turn on RTK compression and automatic fallback.

Do I need to buy API keys?

No — that’s the point. It uses OAuth sessions you already have, so you’re pooling capacity you already pay for.

How many models does it reach?

Over 573 across 40+ providers once your logins are connected.

Is it safe to connect my logins?

It runs on localhost, on your machine, so nothing is handed to a third-party middleman. That’s the main design difference from hosted routers.

How much does it save on tokens?

RTK compression saves roughly 20–40%, and the automatic fallback to free models means premium capacity is only used when needed.

Which agents does it work with?

It’s universal — OpenClaw, Hermes, Claude Code, OpenCode and anything else speaking a standard API.

9Router or OmniRoute?

Both. 9Router is better when you’re using subscriptions you own; OmniRoute is better for pure free-API coding. Run them as a two-option dropdown.

What if a model rate-limits mid-task?

It falls back automatically, including to free tiers, so the work continues.

The bottom line

Learning how to use 9Router takes five steps: install it locally, connect the logins you already have, point your agent at the local endpoint, turn on RTK and fallback, then pair it with a second router so a rate limit never stops you. You’re not buying anything new — you’re pooling what you already own.

About Julian Goldie

I run Goldie Agency, a 7-figure SEO agency, and teach this daily on a 400K+ subscriber YouTube channel. 240+ client projects on Upwork at a 100% job-success score, 10+ years through every major Google update. My systems are in the AI Profit Boardroom; my link building book is free here.