One user turn plus the agent's reply counts as one message. Model tokens are billed by your own provider, not by us - we never proxy your inference or mark it up.

What counts as a message?

One user turn plus the agent's reply. Tool calls the agent makes while answering are part of that same message, however many it needs.

Does the price include model usage?

No, and that is deliberate. Conversation tokens are billed by your model provider through your own gateway key, in your own deployment. We never proxy your inference and never mark it up. What you pay here covers persistence, private attachments, knowledge, memory, and the dashboard.

What happens when I go over?

On Pro, additional messages are billed at the overage rate and nothing stops working. On Free, the agent pauses until the next month rather than surprising you with a charge. Either way this is about messages — your model bill is separate and always yours.

Do you charge per developer?

No. Seats are free on every plan. You are paying for conversation volume, not for who is allowed to open the dashboard.

Can I keep the data myself?

Yes. The widget is MIT licensed and the store contract is public, so you can point the same handler at your own Postgres and object storage and pay us nothing. That path is documented in full.

What happens to my conversations if I leave?

They are yours. Export them over the API, or move the handler onto your own store and keep running the identical widget.