Use Agent Approve

Agents and activity

Find an agent instance, read its event timeline, and understand how commands and tool data are displayed.

Agent Approve groups activity by agent instance. Two sessions from the same coding tool can appear separately when the integration provides different instance or profile information.

Find an agent

Open Agents to see your agents under Active Agents and Inactive Agents. Active Agents lists agents that currently hold an active-session slot, the same count shown in Plan & Usage. An inactive agent that reported activity within about 30 minutes can still show Ready; older entries show Offline until new activity arrives. See Active and inactive agents.

Use search when you have several agents or profiles. For Hermes, each enabled profile is reported as its own agent instance.

Renaming an agent or assigning an image changes how it appears in Agent Approve. It does not rename the coding agent, computer, or Hermes profile. See Agents, names, and images.

Read the timeline

Tap an agent to open its event timeline. Events are grouped by date, and the newest meaningful event opens first. You can search the timeline by event name, tool name, or summary.

The details view formats content according to what the integration reports:

  • Shell commands use a monospaced command view.
  • Structured tool input and output can appear as formatted JSON.
  • Responses and plans can render as Markdown.
  • Input and output appear separately when the event includes both.
  • Subagent starts, stops, and related activity appear when the integration reports those events.

Filter Background Activity, under Settings > Agent Activity, keeps recognized automatic background activity out of the timeline. It is on by default and currently recognizes only Codex's automatic suggested-prompt requests. Tool calls in those sessions still go through normal approval. See Settings reference.

Feature coverage differs by agent. An integration can support approvals without reporting every conversation, plan, subagent, or MCP event. Check the capability matrix for the current limits.

Event details

Tap an event, then swipe left or right to read adjacent events. The content appears first, followed by status and the coding agent's Model, when recorded. Policy Evaluation then explains the decision: Decided By, any list or rule match, evaluator details, timing, AI credits, and the policy.

Rule Evaluation Model names the model that checked the policy, separately from the coding agent's model. AI Credits Used shows the recorded charge; a missing charge is not assumed to be zero. See Read a rule check in event details.

Apple Watch follows the same field order in a compact view. On iPad, event details and forms use more of the available screen width. Content still depends on the integration, privacy level, retained history, and available encryption keys.

Agents page actions

Open the more menu on the Agents page for these list-wide actions:

ActionWhat it does
Reset All StatusSets recently seen agents to Ready. Offline agents remain Offline.
Clear Offline AgentsRemoves the offline entries listed in the confirmation. A running integration can appear again when it reports activity.

These actions change what is shown in your app. They do not uninstall an integration from your computer.

To rename, reset, clear, or remove one agent, open that agent and use its more menu. See Agents, names, and images.

Search agents and events

Use search on the Agents page to find an agent by its displayed name or status. Open an agent and use event search to match event names, tool names, summaries, commands, and other displayed event text.

Search only covers history still available under your data-retention setting.

To control how much event detail is available and how long it remains, review Privacy levels and Data Retention.

Go out and grab a coffee

We'll ping you if your agents need you. Try Agent Approve free for 7 days.