Vercel AI Gateway
One report per day of the window, grouped by model: the cost, the token counts and the request count. Never a prompt, never a response, never a generation's content.
Access
- Kind
- A read-only key you paste
- What to give
- AI Gateway API key
- Why that one
- An AI Gateway key on a Pro or Enterprise team; the reporting endpoint is not available on Hobby. Reading it is billed by Vercel — see the guide.
What it attributes on its own
- From the source
- day · model and the provider that served it · requests · input, output, cached and reasoning tokens · what Vercel charged, or the market price where you brought the key
- Needs a rule or a trace
- team, workflow and agent — unless your requests carry a Vercel reporting tag, which this source does not read yet · outcome · anything from a Hobby or Pro-trial team, where the endpoint is closed
The addresses it calls
- https://ai-gateway.vercel.sh/v1/report
Only these. A call anywhere else is refused before it leaves, and every call is logged.
Connecting it
- Create an AI Gateway key. AI Gateway → API keys. The team is read from the key, so nothing else is needed. Open ↗
- Know what reading it costs. Vercel bills this endpoint at $5 per thousand queries. The report groups by one dimension at a time, so a day-and-model ledger takes one query per day: thirty for a month, about $4.50 a month on a daily sync. Every run reports how many it made.
- Paste it and click Test and connect. Valistry reads the last seven days, one query each, reports what it found, then fills thirty.
Where a provider behind this gateway is also connected directly, connect one of the two: the same request would otherwise be counted twice.