Skip to main content
Enterprise Edition Mention a preset agent when you want it to investigate a case or respond in a comment thread. Tracecat posts the agent’s final answer as a reply and keeps its full run in case chat history.

​
Before you start

Your organization needs the agent add-ons and case add-ons entitlements. Your workspace role also needs:
  • agent:read to find preset agents in autocomplete.
  • agent:execute to invoke the selected agent.
  • case:update to create the comment.
Without agent:read and agent:execute, @ behaves as regular comment text and never starts an agent run. Without either entitlement, @ still opens the list, showing an Enterprise notice in place of agent names.

​
Mention an agent

  1. Open a case and start a new comment or reply.
  2. Type @, then enter part of the preset agent’s name.
  3. Select the agent from the autocomplete list.
  4. Add your request and submit the comment.
Selecting the autocomplete result creates the mention. Typing an agent name as plain text does not invoke it. Tracecat invokes agents only when you create a comment. Editing an existing comment to add a mention never starts a run; create a new comment instead.

​
Follow the response

The comment thread shows when each mentioned agent is preparing, running, or unable to finish. Multiple mentions run independently and can finish in any order. When the run succeeds, Tracecat posts the final answer as an attributed reply in the same thread. If a run fails, mention the agent in a new comment to try again. Agent replies have the same 25,000-character limit as other comments. Tracecat adds [Agent output truncated] when a longer response reaches that limit.

​
Agent context

The agent receives the current case context and the comment thread that contains the mention. The context identifies the mentioning user and asks the agent to write a self-contained reply. Each mention starts a new agent session. Later mentions receive the updated comment thread, but they do not receive another session’s private working transcript.

​
View the full session

Select View session from the agent activity or attributed reply to open the exact case chat session. The session contains the full transcript, including tool calls, errors, and approval pauses that do not appear in the comment reply. Comment-created sessions have a From comment badge in the case chat history. Use the history search to find these sessions after the comment thread is no longer visible.

​
Run workflows from an agent

A mentioned agent can run workflows configured as tools on its preset. When you need fixed order, deterministic branching, or guaranteed steps, use a / command to run the workflow directly.
  • See Case management to configure and manage case records.
  • See AI agent to configure agent tools, outputs, and approvals.
  • See Comment triggers to run a workflow from a case comment instead of invoking a preset agent.