Skip to main content
Version: v0.2 (Next)

Operations & Live Monitoring

The Operations Center provides real-time visibility into active agent executions, raw log streams via an embedded terminal, sandboxed live web previews, and granular token accounting.

1. Real-Time Telemetry Stream (/api/realtime)โ€‹

The Operations page establishes a secure WebSocket connection to /api/realtime. Every 2 seconds, the server broadcasts an aggregated telemetry snapshot:

  • Host Telemetry: Real-time CPU, RAM, disk, and sensor statistics.
  • Durable Runs: Status, iteration count, and execution source of all active agent runs.
  • Pending Approvals: List of paused mutations requiring operator decision.
  • Token Metrics: Cumulative daily and monthly token counts segmented by model provider.
Server-Side Snapshot Caching

The server-side realtime hub caches telemetry snapshots for 1.5 seconds. Multiple concurrent browser tabs or operators do not multiply host sensor polling or database load.

2. Live Terminal Stream (xterm.js)โ€‹

The embedded terminal window renders structured execution logs as they happen:

  • Color-Coded Output:
    • [INFO] (Green): Normal state transitions and DAG node advancements.
    • [TOOL] (Cyan): Tool invocations, inputs, and execution durations.
    • [WARN] (Yellow): Retries, rate limits, and fallback activations.
    • [ERROR] (Red): Failed executions or sandbox violations.
  • Read-Only Safety: The Operations terminal is purely an observation boundary. Direct mutations cannot bypass the tool authorization layer.

3. Live Canvas Previewโ€‹

When an agent generates HTML, React components, interactive charts, or web applications, it can publish a sandboxed preview URL via the ACTONOS_CANVAS_URL environment variable:

  • Embedded Sandboxed Iframe: Displays live rendered web interfaces generated by agents.
  • Zero-Trust Security: The iframe is restricted (sandbox="allow-scripts allow-same-origin"), sends no referrer headers, and only accepts relative paths or HTTPS/loopback URLs.

4. Token Accounting & Cost Breakdownโ€‹

Track LLM operational costs across all agents and models:

  • Daily & Monthly Spend: Graphical breakdown of token consumption and estimated USD cost.
  • Model Distribution: Pie chart showing usage across OpenAI, Anthropic Claude, Google Gemini, and Local Ollama models.
  • Agent Attribution: Identifies which agents or missions are consuming the largest share of token budgets.