Reply expectations
What one visitor reply includes, the token spend and time to expect, monthly conversation includes, and what people see when a reply cannot finish.
Last updated on
Each visitor message gets one reply. You choose the order in Agent Instructions. Another project can publish a different order.
The first part of that reply can appear while the rest of the same reply is still arriving. Visitors see one answer, not a series of separate messages. A short thinking state stays on screen before the reply so a fast answer does not flash in.
Tokens for one reply
One visitor reply is one model request. The platform sends that request once. It does not send another request to continue, retry, or finish the answer.
Cold reply starting point
A cold reply is the first answer in a visit, when an earlier reply in that visit is not reused. Use these figures as the starting point for a usual published project:
| Starting point | |
|---|---|
| Tokens the reply reads | about 9,000 |
| Tokens the reply writes | about 250 |
| Tokens in total | about 9,000 |
| Extra reasoning tokens | none |
| Cached read | none on that first reply |
A longer instruction document or more project content raises the read. A short answer can write fewer than 250 tokens. A usual first reply on a published project is in this range. The exact count moves with the project.
Maximum caps
These are the most one reply can use.
| Cap | Maximum |
|---|---|
| Requests for one reply | 1 |
| Tokens written | 2,048 |
| Extra reasoning tokens | none |
| Parts shown | 5 |
| Agent Instructions | 15,000 characters |
The editor blocks an instruction draft over 15,000 characters. Writing stops at 2,048 tokens. A sixth part is not shown. Shorten the answer in Agent Instructions, or continue the story on a later visitor message.
Reply time
On a usual reply, visitors should see the first part within about 2.5 seconds. The server is built to have that first part ready in about 1.2 seconds. The rest of that short wait is the thinking state and the page showing the reply. The remaining parts of the same reply follow after that.
A cold first reply can take about 2.7 seconds before that first part appears. Later replies in the same visit can be faster. A long instruction document, a large amount of project content, or a connected tool in that same reply can take longer.
What one reply can contain
A reply shows at most five parts. A part can be opening text, a saved section or a section composed for this turn, a follow-up question, or suggested next-step buttons. Leave out a part the turn does not need.
If the instructions ask for more than five parts in a single answer, the extra parts are not shown. Shorten the reply in Agent Instructions, or split the story across a later visitor message.
Connected tools (CRM, email delivery, and Agent tools) can run as part of that same reply. Their result does not start a second answer. The reply should not tell the visitor a tool succeeded unless the instructions say how to talk about that step.
Included conversations and messages
Open organization Billing to see the meters. Each plan includes conversations for the month, and about 2.5 messages per conversation.
| Plan | Conversations a month | Messages a month |
|---|---|---|
| Basic | 200 | 500 |
| Starter | 550 | 1,400 |
| Pro | 4,800 | 12,000 |
| Agency | 14,400 | 36,000 |
Higher plans include everything in the plans below them. See Plans and Limits.
When a reply stops
The visitor sees Conversation error and the reason. The same reason is in Conversation logs. The agent does not continue that answer.
| What happened | What the visitor sees |
|---|---|
| The organization has used its included messages for the billing period | The included-message sentence. Try again after the plan renews, or upgrade. |
| Too many messages in a short time | A wait time, then they can send again. |
| A required service did not respond | A short notice that the reply could not be completed. |
Dashboard Live preview can still send messages when the organization has used its included messages. Public visitors cannot.
A reply that cannot finish is not a content problem in Agent Instructions. Check Billing when the reason names the included message limit. Check Troubleshooting when the reason says a service did not respond.
© 2026 Cintesa. Guide text and screenshots are Cintesa product materials for personal and operator use. Do not scrape or republish them commercially without permission. Terms of Service · Privacy Policy.