Agent Executions
View execution history with metrics, costs, and filtering options.
Overview
The Agent Executions page is your central hub for monitoring agent activity. Every time an agent processes an email or responds to a chat message, it creates an execution record with detailed information about what happened, how long it took, and how much it cost.
Both email executions and chat executions from the Test Chat feature appear in this unified view.
Accessing Executions
Navigate to Monitor > Activity > Agent Activity to view the execution list.
Understanding the Execution List
List View
The execution list shows:
| Column | Description |
|---|---|
| Status | Execution outcome (Success, Failed, In Progress) |
| Agent | Which agent processed the email |
| Subject | Email subject line |
| From | Email sender |
| Duration | How long the execution took |
| Tokens | Total tokens used (input + output) |
| Cost | Calculated cost for this execution |
| Time | When the execution occurred |
Status Indicators
- Green (Success) - Execution completed without errors
- Red (Failed) - Execution encountered an error
- Yellow (In Progress) - Execution is still running
- Gray (Cancelled) - Execution was stopped
- Blue (Paused) - Long-running execution paused for later resumption
- Purple (Scheduled) - Long-running execution waiting to resume
- Purple Timer (Resume Pending) - Execution hit a transient error and is queued for automatic retry
Note: For agents with long-running tasks enabled, executions may span multiple sessions. Paused and Scheduled statuses indicate the execution will continue later.
Note: The Resume Pending indicator appears when an execution encounters a transient infrastructure failure (such as an LLM provider credit balance issue or rate limit) and has been automatically queued for retry. See Auto-Resume below for details.
Filtering Executions
Use the timeframe filter and other controls to find specific executions:
By Timeframe
The timeframe selector at the top of the page lets you quickly filter executions by time period:
| Preset | Description |
|---|---|
| Past Hour | Executions from the last 60 minutes (default) |
| Past Day | Executions from the last 24 hours |
| Last 7 Days | Executions from the past week |
| Custom Range | Select specific start and end dates |
When you select Custom Range, two date pickers appear where you can specify exact start and end dates.
Tip: The timeframe filter helps manage large execution volumes and improves page load times by limiting results.
By Status
Filter by execution outcome:
- All
- Completed (Success)
- Failed
- In Progress
By Agent
Select a specific agent to see only its executions.
By Search
Search for text in:
- Email subject
- Sender email
- Sender name
Combining Filters
Filters combine with AND logic. For example:
- Timeframe: Past Day + Status: Failed
- Shows all failed executions from the last 24 hours
Sorting and Pagination
Sorting
Click column headers to sort:
- Click once for ascending
- Click again for descending
- Sortable columns: Time, Duration, Cost, Tokens
Pagination
At the bottom of the list:
- Change rows per page (10, 25, 50, 100)
- Navigate between pages
- See total execution count
Key Metrics
Duration
How long the execution took from start to finish:
- Fast (
<5s): Simple processing - Normal (5-30s): Typical with tool calls
- Slow (
>30s): Complex reasoning or many tools
Token Usage
Combined input and output tokens:
- Input tokens: Prompt + email content
- Output tokens: Agent response
Understanding tokens:
- ~4 characters = 1 token (English)
- Higher tokens = higher cost
- Large emails or long responses increase tokens
Cost
Calculated based on:
- Model pricing (configured per model)
- Input token count × input price
- Output token count × output price
Note: Costs are estimates based on your configured pricing. Verify against provider billing for accuracy.
Common Tasks
Finding Failed Executions
- Set Status filter to "Failed"
- Review the list
- Click each to see error details
- Take corrective action
Analyzing Costs
- Sort by Cost (descending)
- Review high-cost executions
- Click to see why they were expensive
- Consider optimization (smaller model, shorter prompts)
Reviewing Agent Performance
- Filter by specific agent
- Look at success rate (failures vs. successes)
- Check average duration
- Review cost patterns
Finding Recent Activity
- Sort by Time (descending)
- Review latest executions
- Verify system is processing emails
- Check for any immediate issues
Export and Reporting
Exporting Data
Currently, executions can be:
- Viewed in the UI
- Accessed via API for custom reporting
- Individual execution details can be copied
Metrics to Track
| Metric | Purpose | How to Calculate |
|---|---|---|
| Success Rate | Agent reliability | Successes / Total × 100 |
| Average Cost | Budget planning | Sum of costs / Total executions |
| Average Duration | Performance monitoring | Sum of durations / Total |
| Daily Volume | Usage trends | Count by day |
Best Practices
Regular Monitoring
Check executions daily:
- Look for unexpected failures
- Monitor cost trends
- Verify agents are active
- Catch issues early
Investigating Anomalies
When you see unusual patterns:
- Spike in failures → Check provider status, recent changes
- High costs → Review specific executions
- No activity → Verify subscriptions active
Using Filters Effectively
- Start broad, then narrow down
- Save common filter combinations (browser bookmarks)
- Use date ranges for trend analysis
Long-Running Executions
For agents with long-running tasks enabled, executions behave differently:
Multi-Session Workflows
A single email might trigger an execution that spans multiple sessions:
- Session 1: Initial processing, agent pauses to wait for more information
- Session 2: Resumption 24 hours later, continues work
- Session 3: Final completion
Each session appears as part of the same execution, with:
- Session Count: Shows which session number this is
- Total Duration: Cumulative time across all sessions
- Checkpoint Info: Last saved state for recovery
Tracking Progress
For paused or scheduled executions:
- Look for "Paused" or "Scheduled" status
- Check the execution details for:
- Agent's notes about what it was doing
- When it will resume
- Number of sessions so far
- View related Scheduled Reminders for resumption timing
Understanding Costs
Long-running executions accumulate costs across sessions:
- Each session adds to the token count
- Total cost reflects all sessions combined
- Cost per session visible in execution details
Auto-Resume
When an agent execution encounters a transient infrastructure failure, the system can automatically queue a retry instead of marking the execution as failed. This prevents lost work when temporary issues like LLM provider credit balance exhaustion or rate limiting occur.
How It Works
- The agent detects a transient, resumable error during execution
- A checkpoint is saved with a summary of work completed so far
- The execution is queued for automatic retry with a backoff delay
- A purple timer icon appears in the execution list showing resume status
Resume Status Indicator
The purple timer icon in the execution list shows:
- Attempt number - Which retry attempt this is (e.g., 2/5)
- Max attempts - The maximum number of retries before the system gives up
- Error class - The type of transient error (e.g., balance_exhausted, rate_limit)
Hover over the icon to see the full status detail.
Retry Schedule
The system uses a fixed backoff schedule for retries:
| Attempt | Delay |
|---|---|
| 1 | 5 minutes |
| 2 | 15 minutes |
| 3 | 1 hour |
| 4 | 4 hours |
| 5 | 24 hours |
After 5 failed attempts, the execution is marked as abandoned and no further retries are attempted.
Resume States
| State | Meaning |
|---|---|
| Pending | Retry is queued and waiting for the next attempt |
| Resumed | The execution has been retried and is currently running |
| Completed | The retry succeeded and the execution finished |
| Abandoned | All retry attempts were exhausted without success |
What Gets Preserved
When an execution resumes after a transient failure:
- Previously completed read-only operations (searches, lookups) are re-executed safely
- Email duplicate-send guards are restored to prevent sending the same email twice
- The agent receives a summary of prior progress so it can continue where it left off
When to Investigate
- Repeated resume attempts for the same execution - The underlying issue may not be transient. Check your LLM provider account balance or service status.
- Abandoned executions - All retry attempts failed. Review the error class and original error in the Execution Details to determine if manual intervention is needed.