Credits and billing
How credits work
One credit is one agent reply. What every action costs, the 200 free credits, and why your model provider bills you separately.
Concept2 min readUpdated 23 October 2026
SentientOne is prepaid. There are no plans, seats or monthly fees: you buy credits, and actions spend them. Every feature is open to every account, with unlimited agents and MCP servers.
What costs credits
| Action | Credits |
|---|---|
| An agent reply, from any channel (API, widget, Playground, workflow Agent step) | 1 |
| A PDF page added to knowledge | 5, once, when it's added |
| A website page crawled into knowledge | 5 per indexed page, each crawl |
| A snippet | 1 |
| A Q&A pair | 1 |
| A workflow Classify or Extract step | 1 each time it runs |
| A workflow Guardrail check step | 1, only when an AI-judged check is selected |
| A live workflow run that contains an HTTP, Webhook, Slack or Email block | 1 when the run starts |
Searching knowledge, calling MCP tools (however many), Knowledge lookup, Condition, Delay and Set value steps, tracing and analytics cost nothing extra. Test runs in the workflow designer don't pay the run credit, but their Agent and AI steps cost the same as live.
Free credits
New accounts get 200 free credits, once per email address, with no card. That's enough to add a short document, try your agent in the Playground and send your first API calls. Free credits never expire.
Credits and your model provider
Agents run on your own OpenAI, Anthropic, Google Gemini or Groq key, so your provider bills you for tokens at their prices, with no markup from us. A reply costs 1 credit whichever model it uses and however many tokens it takes. There is no shared or default model key: an agent only answers once you add yours. Credits pay for SentientOne itself: embedding and reranking your knowledge (on our own provider accounts, not your key), search, tool calls, workflows, streaming, hosting and tracing. See Add your LLM key.
When credits come back
- A reply that fails after its credit was taken is refunded automatically.
- A document or website page that fails to ingest is refunded.
- A cached answer (the same first question within 15 minutes) still costs 1 credit: the cache saves time and model tokens, not credits.
- Deleting knowledge doesn't refund the credits it cost.
Seeing where they went
Settings → Usage shows your balance and the credits spent on agent replies and on knowledge over the last 7, 30 or 90 days, with a daily chart. Billing → History lists every change to your balance, grouped by day (UTC).
Do credits expire?
No. Bought and free credits stay on your balance until you use them.
Is a long reply more expensive than a short one?
Not in credits: every reply is 1 credit. Your model provider charges by tokens, so long replies cost more there.
Does a rejected request cost a credit?
No. Rate limits, bad keys and validation errors are checked before a credit is spent.
Still stuck?
Email us with what you tried and any request id. A person replies.