Copilot
A grounded draft reply and a handoff summary for agents — what copilot writes, what it costs in credits, and why a human still presses send.
Copilot helps the agent, not the visitor. It has two modes, both inside a conversation in the Inbox.
Suggested reply
Copilot drafts an answer to the visitor's last message, grounded in your knowledge base — the same content the chatbot answers from.
The draft lands in the composer. Nothing is sent until you send it. That is deliberate: a human takeover exists precisely because the automatic answer was not enough, so the human stays in the loop.
Read the draft before sending. If it is wrong, that is a signal about your content — see Knowledge gaps.
Summarize for handoff
Copilot condenses a long thread into a short summary: what the customer wants, what has been tried, and where it stands. Use it when passing a conversation to a colleague, or when picking up a thread with 40 messages of history.
The summary is written once per conversation. When two agents open the same waiting thread at the same moment, the second one sees the first one's summary rather than paying for another.
What it costs
Copilot makes a real AI call, so:
- it draws down credits like any other AI usage,
- it is subject to your workspace's monthly spend cap,
- it is rate-limited.
It does not consume your plan's flat message quota — a draft is not a visitor message.
If your workspace is out of budget, copilot says so plainly rather than failing silently:
AI copilot is unavailable — this workspace is out of AI credits or hit its
monthly spend cap.See Credits and top-ups and Cost caps.
Where to go next
Last updated