AI test lab and admin copilot

Both are part of Cliedesk Pro. Tests never count in your reports and are labelled “test” in AI cost.

Test lab

Open Settings → AI → Test lab (needs the Manage AI permission).

• Preview: write what a customer says, or give a conversation ID to replay it. You see the answer, the actions the AI would take and a link to its trace. Actions such as refunds are dry runs: nothing changes. To let one test run real actions, turn on live actions for it and confirm. This is recorded in the audit log.
• Test sets: add a set, then add questions or paste a CSV with the columns question and expected. You can also take questions from resolved conversations. Click Run test set: it runs in the background. Above 500 questions you see a cost warning, but it still runs.
• Grades: each answer gets Good, Acceptable or Poor, with a reason, a cause and a fix to try. You can change a grade. Export the results as CSV.
• Compare models: list up to 3 models (one “connection model” per line) and run. You see the answers, grades, speed and cost side by side. Use … for this task sets that model for the task only.
• Regression sets run again when knowledge, policies or models change. If answers that passed now fail, AI changes are not published automatically until an admin confirms the run.

Shadow and rollout

Open Settings → AI → Shadow and rollout. A new AI agent starts in Shadow mode: it writes what it would answer as a private note and sends nothing to customers. You see how often agents sent its draft unchanged. Choose Staged rollout to let it answer a share of conversations (for example 10 %), and plan the next step. The report compares the AI group with the others. Choose Live to turn it on for everyone.

Admin copilot

People with the Use the admin copilot permission (Admins and Supervisors) open it on any screen with Ctrl + Shift + K (Mac: Cmd + Shift + K), the command palette, or the Admin copilot button.

• Ask in your own words: “Why did CSAT drop last week?”, “VIP refund requests over €100 go to Tier 2 with a 1-hour SLA”, “Close all thanks replies older than 3 days”, “Which help articles are out of date?”.
• The copilot can only do what you are allowed to do. It never changes anything by itself: every change is a proposal with the details. Approve, Reject or later Roll back it. New rules start paused and new SLA policies start off.
• Bulk changes show the matching conversations first. Article changes are drafts; nothing is published.
• Memory: the copilot remembers what you ask it to (“always use a formal tone”). You can see and delete it.
• Chats are saved. You can search them and share one with teammates who use the copilot.

Related: AI settings, AI triage.