← Back to Agent
Agent Activity Log
A chronological record of everything this agent has done — events, tool calls, and journal entries with rollback support.
Overview
For Everyone
The Activity Log is your agent's diary. Every action the agent takes — sending a message, calling an API, completing a task, or encountering an error — is recorded here with a timestamp. Think of it as a security camera for your AI agent: you can always go back and see exactly what happened and when.
For Developers
The Activity section aggregates three data sources into a single unified view:
- Events — fetched from
GET /activity/events?agentId=…&limit=200. General lifecycle events (messages, tasks, errors, guardrail triggers, system events).
- Tool Calls — fetched from
GET /activity/tool-calls?agentId=…&limit=200. Every external tool or API the agent invoked, with duration and success/failure status.
- Journal Entries — fetched from
GET /journal?agentId=…&orgId=…&limit=200. Reversible action log with rollback capability via POST /journal/:id/rollback.
All three are loaded in parallel on mount via Promise.all. Client-side filtering and pagination (25 items/page) are applied after fetch.
How It Works
- Data Loading — When you open the Activity tab, all three data sources (events, tool calls, journal) are fetched simultaneously from the engine API.
- Tab Selection — Switch between Events, Tool Calls, and Journal using the tab bar. Each tab shows its own count in parentheses.
- Filtering — Use the filter bar to narrow results by type (dropdown of unique types), free-text search (searches the full JSON), and date range.
- Pagination — Results are paginated at 25 items per page. Changing filters resets to page 1.
- Detail View — Click any row to open a detail modal showing the complete data for that item.
- Rollback — In the Journal tab, reversible entries show a "Rollback" button. Clicking it triggers a confirmation dialog, then calls the rollback API.
- Refresh — The refresh button reloads only the current tab's data source.
Key Concepts
Events Tab
The Events tab shows general agent lifecycle events. Each event has:
- Time — When the event occurred (from
timestamp or createdAt).
- Type — A badge showing the event type. Common types include:
| Type | Description |
|---|
| Message | Chat messages sent or received by the agent. |
| Task | Tasks started, completed, or failed. |
| Tool Use | External tools or APIs the agent called. |
| Error | Failures, timeouts, or API errors. |
| Guardrail | Policy violations or safety interventions. |
| System | Deployment, restart, or configuration changes. |
- Details — A truncated preview of the event data. Click the row to see full details.
The Tool Calls tab tracks every external tool invocation with performance data:
- Tool — The name of the tool or API called (displayed in monospace).
- Duration — How long the call took in milliseconds.
- Status — Color-coded badge:
- OK — Successful call.
- Failed — The call returned an error.
- Pending — Still in progress or unknown status.
Journal Tab
The Journal provides an auditable, reversible action log. Each entry includes:
- Tool — Which tool performed the action.
- Action Type — The specific action (e.g., "create", "delete", "update").
- Reversible — ✅ if the action can be rolled back, ❌ if not.
- Status — "Active" (current) or "Rolled Back" (reversed).
- Rollback Button — Available for reversible entries that haven't been rolled back. Shows a confirmation dialog before executing.
Tip: The Journal is especially useful for debugging. If an agent accidentally deleted something, find the entry and hit "Rollback" to reverse it.
Filtering & Search
The filter bar appears below the tabs and includes:
- Type Dropdown — Lists all unique types found in the current data source. Filters by type/eventType/tool/toolName/actionType fields.
- Search Box — Full-text search across the entire JSON representation of each item.
- Date Range — "From" and "To" date pickers. The "To" date is inclusive (extends to 23:59:59).
- Clear Button — Appears when any filter is active. Resets all filters at once.
Results are displayed 25 per page. Navigation controls include first/previous/next/last buttons and a "Page X / Y" indicator. Changing any filter automatically resets to page 1.
Detail Modal
Clicking any row opens a DetailModal with:
- A color-coded type badge (red for errors, green for deploy/start, yellow for stop, blue for others).
- The complete raw data for the item, formatted for readability.
- The
agentId field is excluded from display (it's redundant since you're already on the agent's page).
Best Practices
- Start with errors when debugging — Use the type filter to select "Error" events first. This narrows down to the exact failure points.
- Use date ranges for incident investigation — If you know when an issue occurred, set the date range to narrow down to the relevant window.
- Check tool call durations — Slow tool calls (high ms values) may indicate API issues or rate limiting. Monitor these to optimize agent performance.
- Review journal entries regularly — The journal provides an audit trail of all reversible actions. Use it to verify that the agent is behaving as expected.
- Use rollback cautiously — Rollback reverses a specific action. Make sure you understand what will be undone before confirming.
- Combine search with type filter — For example, filter by "error" type and search for a specific tool name to find all failures for that tool.
Troubleshooting
No events are showing
The agent may not have performed any actions yet. Deploy the agent and send it a task, then check back. Events are created as the agent operates.
Filters show "No events match filters" but data exists
Your filter combination is too restrictive. Try clearing individual filters one at a time to identify which one is excluding all results. Check that date ranges are correct (from date should be before to date).
Rollback fails
Not all actions can be rolled back — only entries marked with ✅ in the "Reversible" column support rollback. If rollback fails on a reversible entry, the underlying resource may have been modified by a subsequent action. Check the error message for details.
Tool calls show "Pending" status
This usually means the tool call is still in progress, or the agent didn't report a completion status. If it persists, the agent may have crashed mid-call. Check the Events tab for corresponding error events.
Data seems outdated
Activity data is fetched on page load and doesn't auto-refresh. Click the refresh button (↻) to reload the current tab's data.