Inside a ticket, the AI copilot docks in the right-hand context panel, right where you are already reading the conversation. It can summarize a long thread, draft a reply, and suggest relevant knowledge base articles to link. A drafted reply drops into the composer with one click, so you review and edit it before it ever sends. By default nothing the copilot proposes is sent or saved without an agent acting on it: you accept, edit, or reject every suggestion.

Autonomy is set in Service Settings → Autonomy (which needs the autonomy admin permission). Your organization default applies everywhere unless you add a more specific override. The four rungs, from least to most autonomous, are:

One autonomy level rarely fits every kind of ticket. From the Autonomy settings you can add an override for a specific queue, a specific intent (like “password reset” or “refund request”), or both. An override wins over the organization default for the tickets it matches, so you might leave billing questions at Draft + approve while letting a well-understood “where are my docs?” queue run at a higher level. Remove an override at any time to fall the queue back to the default.

Each policy has a confidence threshold. For the auto-send levels, the AI only acts unattended when its confidence in an action clears that threshold; otherwise it routes the action to a human. Raise the threshold to be more cautious; lower it (carefully) to let more through.

When the AI resolves a ticket autonomously, a verification gate runs first. It re-reads the ticket and confirms that whatever the resolution promised (for example “I have issued your refund”) actually happened. If the promised action did not fire, the AI refuses to resolve and routes the ticket to a human instead. A resolution that fails verification is never billed.

When a ticket is handed off to a person, autonomy is disabled for that ticket permanently: the AI will not try to auto-resolve it again. The handoff pings the right owner with full thread context (the assigned agent, else the queue's default assignee, else the account's owning rep), and a Human handoff requested badge appears on the ticket with a matching queue filter so the warm handoff is actually seen.

Every action the AI takes autonomously is written to a signed, append-only resolution log. Each entry is cryptographically chained to the one before it, so the record cannot be altered after the fact. The Autonomy settings page shows this log with a CSV export, giving you a complete, auditable history of what the AI did and why.

The CSV export covers the last 31 days. The on-page log can show older entries, so use the CSV as a recent audit window rather than a complete lifetime archive. A result of up to 10,000 rows and 50 MiB downloads immediately. A larger row set or file runs as a private background export and may download as several numbered CSV files, preserving the signed chain order across those parts.