Relay

Vibgrate Relay is the hosted-model gateway behind the Vibgrate CLI, the MCP server and agent runs. Your workspace's existing credential reaches it — there is no separate AI provider account to set up — and requests draw on prepaid credit.

What you'll see

• Overview — credit balance, spend, requests and tokens for the window you pick, plus the daily trend and the models you actually call.
• Models — the catalog available to this workspace, with the rate per 1M tokens, how much context each model takes, and whether it is currently serving.
• Activity — one row per request: model, whether it drew on credit or your own key, tokens in and out, what it was charged, and the repository the calling client reported.
• Access — the base URL and header a client uses, and which credentials exist for this workspace. Secrets are never shown here.
• Limits — the monthly ceiling, the per-request share of your balance, and auto top-up. Relay checks these before it reserves credit, so a breach refuses the request instead of overdrawing.

How to use it

Start on Overview to see whether spend matches what you expect. If a number surprises you, Activity shows the individual requests behind it and which repository each was attributed to. If you want spend bounded, set a monthly cap.

Things worth knowing

• Usage is recorded off the serving path, so the last few minutes of a busy period can still be settling.
• Buying credit and changing the caps happen in Billing ▸ Credits — this page reads them, it does not edit them.
• A model can show as temporarily unavailable when the upstream provider refuses it. Relay withholds it from clients until it retries, and the default model keeps serving.
• Requests made on your own provider key (BYOK) appear in Activity with no charge against credit.