Prepaid by default
Relay bills prepaid credits. Vibgrate is the principal on provider spend — tokens are bought from the model provider on Vibgrate's account, at Vibgrate's risk, the moment a request is served — so the rule is simple:
Nothing is served on Vibgrate's key unless the money is already ours.
A request that would exceed your workspace's balance is refused with HTTP 402 before anything is spent — never served-then-invoiced. There is no free trial credit: a new workspace's free path is BYOK or local Code Modes, both of which cost Vibgrate nothing to offer.
Enterprise customers on a signed contract can instead use an invoiced, post-pay lane with a contractual credit limit and spend cap — ask your account team.
Buying credit
Top-ups are offered at $25 / $50 / $100 / $250 / $500 / $1,000, by card, from dash.vibgrate.com → Settings → Billing → Relay. A processing fee applies on top, banded by size:
| Top-up | Processing fee |
|---|---|
| $100 and above | 3.9% |
| Under $100 | 4.9% |
The fee is a visible surcharge on the receipt — your card is charged credit + fee, but your workspace is granted the full round credit amount you selected, never a rounded-down figure with the fee silently absorbed into it.
Auto top-up is opt-in: when your balance falls below a threshold you set, Relay charges your saved card automatically. It's bounded three ways so a runaway process can't run up your card — a per-charge amount, a daily cap, and a consecutive-failure stop that disables auto top-up and emails you rather than retrying into a card decline.
Credit expiry
Each purchase is its own "lot" that expires 12 months after purchase. Consumption is FIFO by expiry — the credit closest to expiring is always spent first, the customer-favorable order. You're emailed at 30 and 7 days before any lot expires.
Spend protection
Because a runaway agent could otherwise spend quickly, Relay layers several caps beneath the balance check itself:
- Hard workspace spend ceiling (optional) — set a monthly cap; the workspace cannot spend past it regardless of what an agent tries.
- Per-request ceiling — a single request cannot reserve more than a configurable fraction of your balance, so one pathological prompt can't drain an account in one call.
- Anomaly alerts — spend that far exceeds your trailing 7-day average emails the workspace owner while it's happening, not in next month's statement.
How a request is billed (reserve → serve → settle)
The true cost of a streaming completion isn't known until the response ends, so Relay never serves first and bills after:
- Estimate — a conservative, rounds-up cost is estimated from your prompt size and requested max output tokens.
- Reserve — that estimate is atomically held against your balance. If the balance can't cover it, the request is refused before anything is sent upstream.
- Serve — the request is proxied to the model and streamed to you.
- Settle — once the provider reports actual usage, the reservation is corrected to the real cost and the unused hold is returned to your balance. A failed or errored request settles at zero — you are never charged for tokens you didn't receive.
Prompt caching
When a provider reports cache usage, Relay passes it through rather than billing at list price: cached reads are billed at the cheaper cache rate, and cache writes (which cost providers more) are billed at their real premium. When a provider reports no cache detail, tokens are billed at list — Relay never infers a cache hit it wasn't told about.
What you're never charged for
- A request that errors upstream before producing tokens.
- Tokens beyond what the provider actually reports as consumed.
- Anything on a card that fails — auto top-up disables itself rather than retrying into a decline.