Plans
Current prices are on the pricing page. Allowances may change: read the live values from the API instead of hard-coding them.
Concurrent sessions
A session holds a slot from the moment it leavesqueued until it reaches a terminal state, including when you cancel it. Queued sessions hold no slot. See Get quota for the fields.
Response
Queued sessions
At the limit, a create doesn’t fail: it returns201 with status queued, and the session starts on its own. Set queue: false on the create body to get a 429 instead. For batches, fire every task at once, let the platform pace them, and collect results with webhooks.
Idle sessions make room
At your limit, a new session, a message that restarts a finished one, or a queued session’s start first ends your own sessions that have beenidle for at least 120 seconds, longest idle first.
- They end
interrupted, with asession.status_updatedwebhook. A message picks one up again. - Sessions in any other state, and other users’ sessions, are never ended.
- In the local runtime, a new session at the runtime’s cap ends the longest-idle session right away, and it reads
completed.
Token usage
Agents consume tokens as they work. When the period’s allowance runs out, sessions still start: their model calls are billed to your organization’s credit, at model price plus 20%. Add credit from the billing page in the H Platform. With no credit left, the session fails witherror_code insufficient_credit (see Session failures). The allowance is checked when a session starts: a session that starts within the allowance stays on it until it ends.
Response
Per-organization limits
API requests aren’t rate-limited.