Quotas & usage
What the monthly token quota covers, what happens at the limit, and where to watch consumption.
What counts
Your plan sets a monthly token quota — the tokens the underlying language models consume in your tenant's conversations, input and output, across every channel and the API. Retrieval, document storage, and channel delivery do not consume quota.
What happens at the limit
When the quota is exhausted, chat requests return 429 with a human-readable message, and the widget
shows the customer a polite "temporarily unavailable" notice. Nothing is deleted; conversations resume
when the quota resets at the start of the next monthly cycle or when you upgrade.
Upgrades apply immediately — the new quota is live the moment your plan changes, so a busy month never has to end in silence.
Watching usage
The dashboard shows tenant-level consumption:
- Daily usage — tokens per day for the current cycle, so spikes are visible early.
- By backend — consumption split by model/provider when you run more than one.
The same numbers are available programmatically through the management API, so you can wire alerts into your own monitoring.
Cost hygiene
- Scope agents narrowly. A tight playbook produces shorter, cheaper conversations than a do-everything persona.
- Watch
k. Retrieving more chunks per question costs more input tokens; the default is tuned for typical documents. - Archive stale spaces for customers who churned — memory retrieval stays fast and focused.