AI Agent
Cost Controls
Cove AI charges per reply, not per conversation - here is what drives the cost and how to keep it predictable.
Cove AI is billed per reply, not per conversation or per seat. A “reply” is one bot response sent or drafted (in Suggest mode, a draft counts even if an agent rejects it).
Cost is tracked per interaction and visible in the AI agent analytics dashboard.
What drives the cost
Each reply involves:
- Embedding the customer’s question - converts the text to a vector for retrieval (small, fixed cost).
- Retrieving relevant knowledge chunks - no LLM involved, just a vector similarity search (negligible).
- Generating the answer - the main cost, driven by the model you choose and the length of the answer.
Escalated conversations where the bot decides to hand off are cheaper than full answers because the generation step is shorter.
Choosing a model
Go to Cove AI - Behavior - AI model to pick between:
| Model | Speed | Quality | Relative cost |
|---|---|---|---|
| Haiku (recommended) | Fast | Good for most questions | 1x |
| Sonnet | Slightly slower | Better on complex, nuanced questions | ~3x |
Haiku is the right choice for most apps. It handles the vast majority of support questions well and costs roughly a third as much as Sonnet. Switch to Sonnet only if you find that the bot regularly produces poor answers on complex questions and Haiku’s quality is not sufficient.
Keeping costs predictable
A few settings that directly affect per-reply cost:
- Mode - Suggest and Auto generate a reply for every eligible message. Auto when offline generates replies only during off-hours. If you want to limit bot activity, Auto when offline keeps volume lower.
- Max replies before handoff - setting a limit (for example, 5 replies) caps how many bot turns can happen in one conversation, so a long unresolved thread does not rack up charges.
- Confidence threshold - a higher threshold means more escalations and fewer generated replies, which reduces cost. A lower threshold means more answers (and more cost).
Seeing your spend
The Cost hero stat in the AI agent analytics dashboard shows total spend in dollars for the selected date range, with a comparison to the previous period so you can see whether it is trending up or down.
The daily activity chart also shows cost per day alongside answered and escalated counts.
Was this article helpful?
Thanks for your feedback!