AI Agent & Knowledge Base
Drafting a reply
Section titled “Drafting a reply”Open any ticket and click the ✨ button in the composer — the AI agent reads the conversation so far (and searches your knowledge base, if one’s connected) and drafts a reply. It fills the composer; nothing sends until you do. Editing before sending is expected and tracked (see Readiness, below).
Autonomy levels
Section titled “Autonomy levels”Every AI agent has one of three autonomy levels, and every organization starts on the safest one:
| Level | Behavior |
|---|---|
| Suggest (default) | Never sends anything on its own. A human always clicks send. |
| Auto-reply | Sends its draft automatically for messages a human agent hasn’t already picked up — unless it judges the message needs a human (frustration, a request to speak to a person, anything safety/legal-adjacent), in which case it escalates instead of replying. |
| Auto-resolve | Everything Auto-reply does, plus: after sending, it judges whether the exchange is genuinely finished and resolves the ticket itself if so. |
Auto-reply/auto-resolve never engage on a ticket that’s already assigned to a human, already resolved, or for a contact with automated replies turned off.
Readiness
Section titled “Readiness”Before turning up the autonomy level, check Settings → AI Agent’s readiness dashboard — it looks at your agent’s last 200 suggested replies and reports how many were sent exactly as drafted vs. edited vs. discarded, with a real recommendation once there’s enough sample size. This is real usage data, not a guess.
Knowledge base
Section titled “Knowledge base”Add sources under Settings → Knowledge Base — either a URL (crawled and indexed) or let it grow on its own: every resolved ticket transcript and every saved reply gets embedded automatically, so the knowledge base reflects how your team actually answers things, not just what you manually wrote down. Drafts retrieve from this automatically; nothing extra to configure per-reply.