Connect your API keys
Sign in at app.fusr.ai and paste your own Anthropic, OpenAI, or xAI API key. It is stored encrypted in the FUSR.ai proxy, never in the app database, and never shown again. FUSR.ai keeps the last 4 characters so you can tell keys apart. Free covers 3 providers.
Your providers keep billing those keys directly, at their own prices. FUSR.ai adds nothing to token prices.
Run npx fusrai setup
One command puts your FUSR.ai key in your tools. It needs Node 18 or newer and nothing else, so there is nothing to install first.
npx fusrai setup01Approve the sign-in
It opens FUSR.ai in your browser with a short code. Sign in, check the code matches your terminal, and approve. Codes last ten minutes and work once.
02Pick your tools
Cursor, VS Code with Continue, VS Code with Cline, Aider, your terminal, or a Python and Node snippet. The ones it finds on your computer are ticked.
03It writes the settings
Every file it changes is copied to a .bak file first, and files that hold the key are readable only by you. Cursor and Cline keep settings in their own storage, so it prints the steps for those.
04It sends a test request
Then you're set. Your key is shown as its last four characters unless you ask to see it.
Add --dry-run to see what would change without signing in or writing anything.
Or point any OpenAI compatible tool at FUSR.ai
Every tool needs three things: the base URL https://proxy.fusr.ai/v1, your FUSR.ai key as the API key, and the model name fusr/auto. The setup command writes these for you; this is the same request by hand.
curl https://proxy.fusr.ai/v1/chat/completions \
-H "Authorization: Bearer $FUSR_API_KEY" \
-H "Content-Type: application/json" \
-d '{"model": "fusr/auto", "messages": [{"role": "user", "content": "Say hello"}]}'Chat completions are routed. Embeddings and image generation are not.
Ask for fusr/auto, or name a provider
With fusr/auto, FUSR.ai sorts each request into a task type, like coding, writing, long documents, or a quick question, and sends it to one of your connected models. If that model errors, the next one in line answers. A chat stays on the model whose prompt cache is warm, and prompt caching is on, on every plan.
For one provider every time, ask for fusr/claude, fusr/openai, or fusr/grok (xAI). Those go straight there, with no fallback. Every request and why it went where it did is in the routing log in the app. Prompts are not stored.
Rules and pins
On the Routing page in the app, rules decide which connected model a fusr/auto request goes to. They apply to everyone in your organization. Here is what wins, from the top.
01A model you name
fusr/claude, fusr/openai, or fusr/grok goes to that provider.
02A monthly token cap
Set a monthly token cap on a connected model, and once it is reached that model is skipped.
03A pin
Pin a task type, like coding, to one model and that model goes first for it.
04FUSR.ai's order for the task
Without a pin, FUSR.ai goes down its own list. Turn on Cheapest good enough and the lowest list price among models that handle the task well goes first.
Turn auto-routing off and every fusr/auto request goes to your default model, with no task routing and no fallback. On Pro and Team, fusr/auto also learns which model goes well for you, and skips a model whose context window is too small for the request. How the pick works is covered in the AI router guide.
Spend caps per key
A spend cap stops your FUSR.ai key once it has spent a set amount today or this month. Spend is what the proxy priced for each request, not a bill from your provider. Free sets one monthly cap. Pro and Team set daily and monthly caps on every key.
When a cap is hit, every request on that key gets an error that names the cap and when it resets. Nothing is queued or sent later. Days reset at 00:00 UTC and months at 00:00 UTC on the 1st. Raise or remove a cap and the key works again within about 30 seconds. You can also get one warning at the line you pick, in the app and by email.
The API
A small, read-only API for your own account, at https://app.fusr.ai/api/v1. Make a personal API token on the Developers page in the app and send it as a Bearer token. A token is separate from your FUSR.ai key: it reads your account and cannot call a model.
/usage: your logged usage events. Token counts only, never prompts./providers: the providers you connected, by name and status. Never keys./alerts: your alerts, on Pro and Team.
curl https://app.fusr.ai/api/v1/usage \
-H "Authorization: Bearer $FUSR_TOKEN"Each token gets 60 requests a minute. Webhooks send signed events to your own URL. The full reference, with every field, error, and webhook, is at app.fusr.ai/developers/docs once you sign in.