Billing & Credits
How credits, balances, and transactions work.
Available balance
The balance shown on the Billing page and the Dashboard is your available balance: total credits minus credits currently reserved for active voice sessions. Reserved credits are released or settled when the session ends, and the Credit allocation panel shows the reserved amount separately.
Credit buckets and spend order
Credits live in three buckets and are spent expiring-first: trial credits, then subscription (monthly plan) credits, then purchased credits.
- Monthly plan credits reset each billing period and do not roll over. The Billing page shows a meter with how many of the plan's granted monthly credits remain and when they reset.
- Purchased credit packages stay in your account until used.
- Trial credits are promotional and expire after the trial window.
Usage billing rules
- Voice sessions bill per started minute — a 65 second call costs 2 credits.
- Chat costs one credit when an assistant response is delivered. Failed calls before any output are not charged.
- Portal guest sessions debit the portal host account, not the guest member.
- Knowledge-base search costs one credit per search request; a search that fails is not charged. Searches performed by a voice agent during a call are covered by the per-minute voice charge and are not billed separately.
- Adding documents to a knowledge base costs one credit per indexed chunk (a chunk is roughly a page-sized piece of text). Larger documents produce more chunks and therefore cost more to index; the cost is charged as the document is processed. If indexing fails permanently, the charge is automatically refunded — you only pay for documents that actually get indexed.
AI workflow actions and model pricing
AI actions in workflows (Summarize, Custom prompt, Sentiment analysis, and similar) cost a base of one credit per run, scaled by the model that executes them:
| Model family | Credits per AI action |
|---|---|
| Claude Haiku | 1 |
| Claude Sonnet | 3 |
| Claude Opus | 15 |
| Other models | 5 |
- The Custom prompt action lets you choose the model. On the Free plan only Haiku models are available; every paid plan can select any model (billed per the table above).
- Actions that use your own connected AI account (bring-your-own-key) run on your provider account and are not charged platform credits.
Plan limits and usage
Each plan caps how many of certain resources you can have — agents, spaces, collections, workflows, API keys, and portals. Settings → Usage shows a bar for each resource with how much you've used against your plan limit (unlimited resources show no bar).
If you switch to a plan whose limit is below your current usage — or you're otherwise over a limit — the over-limit resources are flagged on the Usage tab ("over by N") and summarized on the Dashboard. While a resource is over its plan limit, creating new resources of that type (and starting new sessions or workflow runs) is blocked until you remove the excess or upgrade your plan. Your existing resources are never deleted when a plan changes — nothing is removed on your behalf, so you choose what to trim. Both the Usage tab and the Dashboard notice link straight to Billing to review or change your plan.
When generating an API key or adding text or a file to a collection, a billing block appears inside the form with Review plan. Your entries remain available while the form is open. Selecting Review plan takes you to Billing; leaving or cancelling the form does not promise to restore that draft.
Transactions
Recent transactions on the Billing page show the amount, the balance after the transaction, the operation that produced the charge or grant, and — where applicable — the resource it applies to (voice sessions and portals link straight to their pages). Use Previous / Next under the list to page through your full history.