Every AI tool wants its own setup. Claude Code wants one key, Cursor another, n8n a third, your agent a fourth, and when one provider rate-limits you, you’re editing four config files. It’s easier to give them all one endpoint and one key, and let a router decide who answers.
That’s what 9Router does: a self-hosted, OpenAI-compatible /v1 endpoint in front of all your providers.
What you get
- One base URL and key per tool. Every client talks to the same
/v1. - Automatic fallback. Subscriptions first, then cheap APIs, then free tiers. When one runs out or throttles, the next one answers.
- Named combos. Define
coding= Claude → GLM → free fallback, and tools callcodingas if it were one model. - Usage per key. See which tool burns what, and revoke a key without touching the others.
- Fewer tokens. Tool output like
git diffandgrepis compressed before it’s sent, typically saving 20–40% of input tokens.
Step 1: Connect your providers
In the 9Router dashboard, add the accounts you already have: API keys, OAuth subscriptions, OpenRouter, free tiers. Put them in tiers in the order you want them tried. (Check each provider’s terms for subscription plans; some are licensed for personal use only.)
Step 2: Create a combo and keys
Make a combo such as default or coding, then issue one API key per tool or teammate. Keys are cheap; one per client makes the usage chart useful.
Step 3: Point your tools at it
Each tool gets the same two values:
Base URL: https://your-9router-address/v1
API key: <key from the 9Router dashboard>
Model: coding
- Coding agents (Claude Code, Codex, Cursor, Cline, Continue, Roo, OpenCode): set the custom or OpenAI-compatible base URL.
- n8n: add an OpenAI credential with the custom base URL, and every AI node in every workflow uses the router.
- Hermes Agent: set it as a custom endpoint, and the agent inherits fallback for free.
Why the router should live on a server
A router on your laptop disappears when the lid closes, taking every tool with it. It needs to be always on and reachable over HTTPS from every machine.
On Cadesia, 9Router runs in its own Pod with an HTTPS Address and persistent storage for providers, combos, and usage history. Pick it in the App Library, or take a bundle: the AI Agent Router Solution pairs it with Hermes Agent, and the Workflow AI Router Solution pairs it with n8n. Comparing it with a hosted gateway? See 9Router vs OpenRouter.
Questions
Which tools work with a 9Router endpoint?
Anything that accepts a custom OpenAI-compatible base URL: Claude Code, Codex, Cursor, Cline, Continue, Roo, OpenCode, n8n's OpenAI nodes, Hermes Agent, and most SDKs.
What happens when a provider hits its rate limit?
9Router moves the request to the next provider in the chain, for example from a subscription to a paid API to a free tier, so the tool keeps working without a config change.
Can different tools use different keys?
Yes. The 9Router dashboard issues separate API keys and tracks usage per key, so you can see what each tool or teammate costs and revoke one without touching the others.