Open /agent-activity from the sidebar. The page surfaces every action the AI agent has taken in your workspace over the last 30 days, broken down by tool. Three top-line stats (total actions, distinct agent identities, most-used tool), a horizontal bar chart of the top tools by call volume, and a chronological recent-actions timeline that links to the affected documents.
This is a deliberate split from /audit. The full audit log is an auditor's tool — exhaustive, configurable, slow to scan. Agent activity is a manager's tool — "did the AI do anything weird this week?" — and it's scoped, summarized, and visual. The two coexist: a "Open the full audit log →" link sits at the bottom of /agent-activity for deep filtering or older history.
The data underneath is the same hash-chained, permission-trimmed audit log that backs /audit. Agent calls always run with the requesting user's permission scope; an agent action you see here always corresponds to something a human asked the agent to do.