Monitor agent activity
Connect agent instances, choose an observation mode, and review the activity each integration reports.
Use Agent Approve to bring supported activity from local, remote, scheduled, and profile-based agents into one mobile view.
Choose how much control you need
- Choose Observe Mode when you want encrypted activity history without approval hooks.
- Choose Approval Mode when you also want supported actions evaluated by policy before they run.
Run npx agentapprove install, choose the mode, and select every integration running on that computer. Repeat the installer on each additional computer or server. Each computer has its own pairing and local configuration.
Verify each agent
- Start a normal session in one installed agent.
- Ask it to perform a harmless action.
- Open Agents in the iOS app.
- Find the agent instance and open its timeline.
- Confirm that the expected command, tool call, response, or lifecycle event appears.
- Repeat the check for each important agent, computer, or profile.
Hermes profiles appear as separate agent instances. Other integrations can also report separate instances when they provide enough session or profile information.
Read the feed
The timeline can show commands, tool input and output, responses, completion events, MCP activity, and subagent events. Coverage depends on the hooks or plugin surface exposed by the coding agent.
Use timeline search to find an event by its name, tool, or summary. Open the event to see formatted shell content, JSON, Markdown, and separated input or output when those fields were reported.
See Agents and activity for status and history controls, and check Integration capabilities before expecting a specific event type.
Know what the feed does not decide
Agent Approve shows the activity an integration reports. It does not automatically classify every long run as a loop, prove that an agent has drifted from its goal, or replace provider billing and spending alerts. Use the feed to investigate unusual timing, repeated actions, errors, or an unexpected active agent.
Your Privacy Level controls how much event content remains in history. Data Retention controls how long you can review it.