Credits are a unit for reading your AI usage: one credit is a ten-thousandth of a US dollar of underlying cost, and your workspace can see its monthly usage in both dollars and credits, broken down by model, operation and bucket. Separately, each plan has a monthly spend ceiling, which Wetel checks before each chat turn, workflow task, direct model call and playground call, and new work is refused once the month’s recorded spend has reached it.
Think of a company card with a monthly limit and a monthly statement. The statement shows what was spent; the limit is checked when someone swipes. Neither is a prepaid card with a balance, and that difference matters.
What is metered?
Every successful AI operation your workspace runs is recorded against it. That covers:
- model calls: replies, embeddings for knowledge search, and conversation evaluations
- speech, in both directions (text to speech and speech to text)
- replies from the browser-rendered 3D avatar
- calls your AI employees make to tools, actions and webhooks
Three details:
- When one model provider fails and another answers, only the one that answered is billed.
- A failed tool, action or webhook call is not charged, and neither is a failed or refused test of a custom action.
- A custom action or a webhook can carry its own per-call price in credits, which you set and can see. Unset, the platform’s flat default applies.
How do I read my own usage?
Ask for your usage summary for a month; it returns dollars, credits and the ceiling you are measured against. The apiUsageSummary query takes an optional month (yearMonth, as YYYY-MM) and defaults to the current one. It returns:
- your total language-model cost in US dollars for that month
- the monthly spend ceiling your plan is measured against
- a breakdown by model and operation, each row with a call count, cost and credits
- four buckets: AI credits (model work plus tool and action calls), text to speech, speech to text, and avatar
It covers every source in the workspace but cannot be split by API key, and hosted video avatar usage is not tracked yet. See reading your usage on the pricing page.
How does the spend ceiling work?
Each plan has a monthly ceiling on language-model spend, checked before the work starts. The documented ceilings are $5 a month on DEV and $200 a month on PRO. When the month’s recorded spend has reached the ceiling, new work is refused until the calendar month rolls over or the plan changes; on the direct model API the error names the limit and the month’s spend so far. There is no self-serve plan change; a workspace owner who needs more asks the team.
What do credits not promise?
Credits are a planning tool, not an invoice, and several things people expect from credits are not built.
- Credit figures for model usage are an estimate. They are a blended conversion of the dollar cost. Compare and plan with them; do not reconcile against them.
- The ceiling is a guard, not an exact meter. It is checked before work starts, so work already under way can carry spend past it. It caps language-model spend, so speech and tool calls do not count toward it.
- A failed workflow run is not refunded. Steps that completed before the failure stay counted.
- There are no prepaid credits, credit balance, wallet or monthly credit allowance. Nothing stops at “zero credits”, because there is no balance to run down.
- There is no published rate card, and no credit figure per role or per AI employee. Wetel reports usage today; it does not bill it yet.
Related reference docs
- Pricing & Plans: plan limits and what is not built yet
- Tenant API:
apiUsageSummary: every field in the usage summary - Tenant API: usage buckets: what each bucket counts, and which figures are approximate
- LLM API: monthly spend cap: the error you get when the ceiling is reached
If you are planning real work, tell us which work you’d hand over first and we will walk through what it would use.