Skip to main content

​
Overview

The Agents Page provides a centralized view of all agents in your organization. Agents are automatically discovered when they send their first run - no manual configuration required.

​
Agent discovery

​
How it works

  1. Agent sends first run via SDK with agent_id
  2. Platform automatically creates agent record
  3. Metrics begin calculating in real-time
  4. Agent appears in Agents list
No setup required - just instrument your agents with the SDK and they self-register.

​
Agent list view

Main table shows all discovered agents:
ColumnDescription
Agent IDUnique identifier (click to view details)
Display NameHuman-readable name (editable)
DescriptionWhat the agent does (editable)
Total RunsLifetime run count
Total CostLifetime USD spent
Success Rate% of successful actions
Intervention Rate% of actions blocked by policies
Last SeenTimestamp of most recent run
StatusActive (green) or Inactive (gray)

​
Sorting

Sort by multiple fields:
  • Last Seen - Most recently active
  • First Seen - Oldest agents
  • Total Runs - Most active agents
  • Total Cost - Most expensive agents
  • Intervention Rate - Most blocked agents
  • Success Rate - Most/least reliable agents
Default: Last seen (descending).

​
Filtering

  • Agent ID - Partial match search
  • Status - Active vs inactive
  • Organization - (if multi-org enabled)

​
Agent status

  • Active (🟢) - Sent data in last 7 days
  • Inactive (⚫) - No data in 7+ days

​
Agent details view

Click any agent to see comprehensive statistics:

​
Overview section

Basic information:
  • Agent ID - Unique identifier (cannot be changed)
  • Display Name - Human-readable name (editable)
  • Description - What the agent does (editable)
  • Organization - Which org owns this agent
  • First Seen - When agent was first discovered
  • Last Seen - Most recent activity
  • Days Active - Count of unique days with runs
  • Status - Active/Inactive
Edit metadata:
  • Click “Edit” button
  • Update display name and description
  • Save changes
  • Note: Agent ID cannot be changed

​
Statistics overview

Activity metrics:
  • Total Runs - Lifetime run count
  • Total Actions - Lifetime action count
  • Total Cost - Lifetime USD spent
  • Avg Cost per Run - Mean cost per run
  • Avg Actions per Run - Mean actions per run
Performance metrics:
  • Success Rate - % of successful actions
  • Error Rate - % of failed actions
  • Avg Duration per Action - Mean execution time
Intervention metrics:
  • Intervention Rate - % of actions that triggered interventions
  • Interventions Count - Total interventions
  • High-Risk Blocks - Critical/high-risk interventions
Approval metrics:
  • Total Approvals - Lifetime approval requests
  • Pending Approvals - Currently awaiting decision
  • Approval Rate - % of requests approved
  • Avg Decision Time - Mean time from request to decision

​
Recent activity

Three time-window breakdowns: Last 24 hours:
  • Runs: X
  • Actions: Y
  • Cost: $Z
Last 7 days:
  • Runs: X
  • Actions: Y
  • Cost: $Z
Last 30 days:
  • Runs: X
  • Actions: Y
  • Cost: $Z

​
Activity timeline

Visual timeline showing:
  • Daily run counts (bar chart)
  • Daily costs (line chart overlaid)
  • 30-day rolling window
  • Hover for exact values
Insights:
  • Spot trends (increasing/decreasing activity)
  • Identify spikes or anomalies
  • Correlate with deployments

​
Recent runs list

Table of last 10 runs:
  • Run ID (link to run details)
  • Status (completed/failed/running)
  • Cost
  • Actions count
  • Duration
  • Timestamp
Click “View All Runs” to filter Runs page by this agent.

​
Top actions

List of most frequent action types:
  • Action name
  • Count (how many times called)
  • Total cost
  • Avg cost per call
  • Success rate
Sorted by: Count (descending)

​
Cost breakdown

Pie chart showing:
  • Cost distribution by action type
  • Hover for exact amounts
  • Click slice to filter ledger by that action

​
Interventions summary

If agent has interventions:
  • Total interventions
  • Breakdown by type (hard_block, rate_limited, etc.)
  • Recent interventions list (last 5)
  • Link to filter Interventions page by this agent

​
Approvals summary

If agent has approval requests:
  • Total approval requests
  • Pending count
  • Approval rate
  • Recent approvals list (last 5)
  • Link to filter Approvals page by this agent

​
Quick actions

From agent details: View filtered data:
  • View Runs - Opens Runs page filtered by this agent
  • View Actions - Opens Activity Ledger filtered by this agent
  • View Interventions - Opens Interventions page filtered by this agent
  • View Approvals - Opens Approvals page filtered by this agent
Edit agent:
  • Edit Metadata - Update display name and description
  • Archive Agent - Hide from active list (future feature)
Export data:
  • Export Stats - Download agent statistics as JSON
  • Export Activity - Download all actions for this agent

​
Common workflows

​
Identify high-cost agents

  1. Sort by “Total Cost” (descending)
  2. Review top 5 agents
  3. For each:
    • Click to view details
    • Check “Top Actions” section
    • Identify expensive action types
    • Review “Cost Breakdown” pie chart
  4. Optimize expensive actions:
    • Implement caching
    • Use cheaper models
    • Add rate limiting

​
Monitor agent health

  1. Sort by “Success Rate” (ascending)
  2. Find agents with low success rates (< 90%)
  3. For each:
    • Click to view details
    • Review recent runs
    • Filter errors in Activity Ledger
    • Identify common error patterns
  4. Fix root causes

​
Review intervention patterns

  1. Sort by “Intervention Rate” (descending)
  2. Find agents with high intervention rates (> 5%)
  3. For each:
    • View interventions summary
    • Check what’s being blocked
    • Determine if:
      • Agent is misbehaving - Fix agent logic
      • Policies are too strict - Adjust policies
  4. Take corrective action

​
Audit agent activity

  1. Select agent
  2. View “Recent Activity” section
  3. Check for anomalies:
    • Sudden cost spikes
    • Drop in success rate
    • Increase in intervention rate
  4. Investigate using filtered views (runs, actions, interventions)

​
Set up new agent

  1. Instrument agent code with SDK: 
    enable_remote_sync(
    platform_url="...",
    api_token="...",
    run_id="run-001"
)
  2. Run agent once
  3. Return to Agents page - see agent auto-discovered
  4. Click agent
  5. Edit metadata:
    • Display Name: “Customer Support Bot”
    • Description: “Answers customer questions via email”
  6. Save

​
Best practices

Use descriptive agent_id values: Choose meaningful IDs like “customer-support-prod” instead of “agent1”. Agent IDs cannot be changed.
Add display names immediately: When a new agent appears, edit it to add a human-readable display name and description.
Monitor intervention rates: Agents with intervention rates > 10% need attention - either agent logic or policies should be adjusted.
Low success rates indicate problems: If success rate drops below 80%, investigate immediately - the agent may be broken.
Review top actions regularly: Understanding which actions are most frequent helps optimize costs and identify bottlenecks.

​
Metrics explained

​
Success rate

Success Rate = (Successful Actions / Total Actions) × 100
  • 90-100%: Excellent - agent is reliable
  • 70-89%: Good - some errors expected
  • 50-69%: Poor - investigate errors
  • < 50%: Critical - agent is broken

​
Intervention rate

Intervention Rate = (Interventions / Total Actions) × 100
  • 0-2%: Normal - occasional policy blocks
  • 3-5%: Elevated - review policies
  • 6-10%: High - agent or policies need adjustment
  • > 10%: Critical - major mismatch between agent behavior and policies

​
Approval rate

Approval Rate = (Approved Requests / Total Requests) × 100
  • 80-100%: Good - agent requests appropriate actions
  • 60-79%: Moderate - some rejections expected
  • 40-59%: Poor - agent requesting inappropriate actions
  • < 40%: Critical - review agent logic

​
Troubleshooting

​
”Agent not appearing”

  • Verify agent has sent at least one run
  • Check agent_id in SDK configuration
  • Confirm enable_remote_sync() is called
  • Check organization selector (may be in different org)

​
“Stats not updating”

  • Stats update every 5 minutes
  • Refresh page to see latest data
  • Check if agent has recent runs (Last Seen)
  • Verify WebSocket connection (should show REALTIME)

​
“Can’t edit agent ID”

  • Agent IDs are permanent (set on first run)
  • Only display name and description are editable
  • To change ID, create a new agent with different agent_id

​
”Intervention rate seems high”

  • Review “Interventions Summary” section
  • Check which actions are being blocked
  • Determine if policies are too strict or agent needs fixes
  • Link to Interventions page for full details

​
See also