The task detail view's Trajectory tab lays out a lane-grouped timeline of every tool call and message in a run — Input, Agent, Thinking, and Tools. Search it, click into any entry, and see exactly what ran, what it returned, and whether it was allowed or denied.
Live capture from the AgentRQ app — opening the Trajectory tab, then inspecting a denied and an allowed tool call and its payload.
One tab, every action the agent took, in order.
Every task detail view has CHAT and TRAJECTORY tabs. Trajectory switches from the conversation thread to a minimap of Input, Agent, Thinking, and Tools lanes — with a live count badge of how many tool calls the run made.
Search tool calls and messages by keyword, narrow the list to one lane with the filter chips, or scan the color-coded minimap dots to jump straight to a specific moment in a long-running task.
Click a tool call to see its tool name, status (auto-allowed, allowed, or denied), a human-readable summary, and the full input/payload that was sent — a complete audit trail for every risky action.
The lane strip at the top groups every event into Input, Agent, Thinking, and Tools rows, so you can see the shape of a run at a glance before diving in — a burst of tool calls, a long silence, a human reply in the middle.
Below it, the search box filters the full list of tool calls and messages by keyword, and the list itself is ordered chronologically — reads, bash commands, edits, writes, and the messages that surrounded them, exactly as they happened.