← Back to Agent
Agent Overview
The central dashboard for monitoring an agent's status, usage, and performing quick actions.
Overview
For Everyone
The Overview tab is the first thing you see when you open an agent's detail page. It provides a snapshot of the agent's current state — whether it's running, how much it's cost today, whether it's been onboarded, and whether its guardrails are active. Think of it as the agent's "home screen."
For Developers
The Overview section aggregates data from four engine endpoints (/agents/:id/usage, /onboarding/status/:id, /guardrails/status/:id, /workforce/status/:id) and subscribes to real-time updates via Server-Sent Events (SSE) on /api/engine/agent-status-stream. Actions like deploy, stop, restart, pause, resume, clock-in, and clock-out are POST calls to the engine. The 5-step delete flow goes through the bridge API (DELETE /bridge/agents/:id).
How It Works
When you navigate to an agent's detail page, the Overview tab loads automatically. It:
- Fetches the agent's usage data (tokens, cost, uptime, error rate, active sessions)
- Checks onboarding status — whether the agent has acknowledged your organization's policies
- Checks guardrail status — whether the agent is active or paused, and its intervention count
- Checks workforce status — whether the agent is clocked in and how many tasks are active
- Opens an SSE connection for live status updates (online/idle/busy/error)
All data refreshes automatically. The SSE connection keeps the live status updated without polling.
Agent Summary Card
The top card shows five key fields at a glance:
| Field | Description |
|---|
| Status | Current agent state — running, active, deploying, ready, degraded, error, stopped, or draft. Color-coded badges. |
| Role | The agent's assigned role (e.g., agent, assistant, manager, specialist). |
| Model | The LLM model the agent uses (e.g., claude-sonnet-4-20250514). Shown in monospace. |
| Reports To | The agent's manager — either another agent (clickable) or an external person with email. |
| Created | When the agent was created. |
Below the grid, the agent's description and personality traits are displayed as neutral badges.
Today's Usage Stats
A row of five stat cards showing real-time metrics for the current day (UTC):
- Tokens Today — Total input + output tokens consumed by the LLM.
- Cost Today — Estimated API cost based on token usage and model pricing.
- Uptime — How long since the agent's last restart.
- Error Rate — Percentage of failed requests. Turns red above 5%.
- Active Sessions — Currently open conversation sessions.
Tip: If cost is climbing faster than expected, check the Budget tab to set daily/monthly spending limits.
Status Indicators
Three cards in a row show the agent's operational status:
Onboarding
Shows whether the agent has acknowledged your organization's policies. States:
- Onboarded — Complete. The agent has read all policies.
- In Progress (X/Y) — Partially complete.
- Not Onboarded — Hasn't started. You can "Start Onboarding" or "Force Complete" to skip.
Guardrails
Shows whether the agent's safety rules are active:
- Active — All guardrails enforced.
- Paused — Agent paused manually or by an automated trigger.
Also shows the intervention count — how many times a guardrail rule has triggered.
Workforce
Shows whether the agent is "on duty":
- Clocked In — The agent processes incoming tasks.
- Clocked Out — Off duty. Tasks queue until the agent clocks back in.
Organization & Knowledge Access
Two side-by-side cards:
Organization
Shows which client organization the agent is assigned to. You can reassign via the dropdown. Selecting "No organization" unassigns the agent.
Knowledge Access
Lists all knowledge bases and the agent's access level for each:
- No access — Agent cannot use this knowledge base.
- Read — Agent can search and retrieve information.
- Contribute — Agent can add new entries.
- Both — Read + Contribute.
Live Status (SSE)
A real-time card that streams the agent's current activity via Server-Sent Events. Updates automatically without refreshing.
- Online — Running and processing requests.
- Idle — Running but waiting for tasks.
- Busy — Actively processing. Shows the current activity, tool being used, and start time.
- Offline — Agent is not running. Shows last activity time.
The stats row shows active session count, uptime, and last heartbeat time.
Quick Actions
A toolbar of common operations:
| Action | When Available | What It Does |
|---|
| Reset State | Error / Degraded / Draft | Resets agent state to "ready" |
| Deploy | Not running | Starts the agent |
| Stop | Running / Active | Stops the agent |
| Restart | Running / Active / Stopped | Restarts the engine |
| Pause / Resume | Guardrails available | Temporarily halts task processing (context preserved) |
| Clock In / Out | Workforce available | Toggles availability |
Tip: Pause is instant and safe — it doesn't lose any context. Use it when you need to make config changes.
Danger Zone
Contains the Delete Agent button. This triggers a 5-step confirmation flow:
- Step 1: Initial warning — "Are you sure?"
- Step 2: Data loss warning — emails, sessions, logs, config, scheduled jobs
- Step 3: Memory & knowledge loss — long-term memory, learned preferences, embeddings, personality
- Step 4: Communication & integration impact — email stops, APIs break, workflows disrupted
- Step 5: Type the agent's name to confirm deletion
Warning: Agent deletion is permanent and irreversible. There is no recycle bin. Export any data you need before proceeding.
Key Concepts
- Agent State — The lifecycle state: draft → ready → deploying → running/active → stopped/error/degraded.
- Hot-Update — When an agent is running, config changes use the hot-update endpoint to apply without restart.
- SSE (Server-Sent Events) — A persistent HTTP connection that streams real-time status updates from the engine.
- Onboarding — A process where the agent reads and acknowledges your organization's policies before it can work.
- Guardrail Interventions — Each time a safety rule triggers (blocks a response, flags content, or pauses the agent).
- Workforce Clock — An availability system — agents only process tasks when clocked in.
Best Practices
- Always onboard agents before deploying them to production. This ensures they know your policies.
- Monitor the error rate — anything above 5% warrants investigation.
- Use Pause instead of Stop when making config changes — it preserves session context.
- Set up workforce schedules to automatically clock agents in/out during business hours.
- Check the intervention count regularly. High counts may indicate the agent needs reconfiguration or better prompting.
- Assign agents to organizations to scope their knowledge access and policies correctly.
Troubleshooting
| Problem | Solution |
|---|
| Agent shows "offline" but should be running | Try Restart from Quick Actions. Check engine logs for errors. |
| High error rate | Check the Activity tab for specific errors. Common causes: expired API keys, rate limits, model unavailable. |
| Agent stuck in "deploying" | Use Reset State to return to "ready", then Deploy again. |
| Onboarding won't complete | Use "Force Complete" to skip. Or ensure guardrail policies are configured in the Guardrails tab. |
| Live status not updating | The SSE connection may have dropped. Refresh the page. Check browser console for connection errors. |
| Cost unexpectedly high | Check token usage. Review the heartbeat interval in Configuration tab — frequent heartbeats with expensive models add up. |
Configuration — LLM model, heartbeat, voice, and model routing settings.
Manager — Assign a supervisor and configure daily catch-ups.
Permissions — Risk levels, rate limits, approval workflows.
Tools — Enable/disable tool categories.
Personal Details — Name, role, personality, and identity.