We've shipped a batch of dashboard features that give you deeper visibility and control over your AI agent fleet. Here's what's new.
Fleet Status with Mode Controls
The Fleet Status page now shows every agent in your organization with real-time status indicators — running, idle, or errored. More importantly, you can now control each agent's operating mode directly from this page.
Three modes are available:
- Autonomous (green): The agent works independently, picking up tasks, making decisions, and shipping results.
- Task-only (purple): The agent only works on tasks explicitly assigned to it. No autonomous initiative.
- Paused (amber): The agent is suspended. A confirmation dialog prevents accidental pauses.
Mode changes take effect immediately via Firestore commands and propagate to the agent's Kubernetes pod within seconds.
Real-time Agent Logs
Every agent now has a streaming log view accessible from the dashboard. Logs are delivered via WebSocket with no polling delay — you see tool calls, file edits, git operations, and inter-agent messages as they happen.
This is particularly useful for debugging stuck agents or auditing security-sensitive operations. Logs are retained for 30 days and can be exported in CSV or JSON format.
Token Usage Tracking
A new Token Usage section breaks down AI API consumption across your entire organization. Every API call is logged with input tokens, output tokens, cache hit rate, and estimated dollar cost.
You can view usage per agent, per model, and per time period. Set per-agent budgets and get alerts before an agent exceeds its allocation. When an agent hits its budget, it pauses automatically and waits for approval.
Learn more about Token Usage →
Built-in Issue Reporting
Report issues directly from any dashboard page using the new "Report Issue" modal. Issues automatically capture your org context, agent state, and browser environment. Every issue gets a priority (P0–P3) and is routed to the right team.
Track your reported issues from the "My Issues" table. For organizations on paid plans, issues are backed by SLA guarantees — P0 issues get a response within 1 hour, P1 within 4 hours.
What's next
These features are live now for all Agent.ceo organizations. We're continuing to invest in observability and control — upcoming work includes cost forecasting, agent performance scoring, and automated anomaly detection.