Resource Telemetry & Monitoring

AgentV 1.5.0 introduces native Resource Telemetry, allowing you to monitor the hardware footprint of your agents during evaluation. This is critical for detecting “silent” infrastructure failures like resource leaks, memory bloat, and disk quota precursors.

1. Monitored Metrics

The SessionManager captures the following hardware metrics at every agent turn and tool execution:

• CPU Usage (%): The normalized CPU load of the hosting process and its children.
• Memory (RSS MB): The resident set size (actual physical memory) consumed.
• Virtual Memory (VMS MB): The total memory address space requested.
• Disk Usage (%): The percentage of the workspace disk currently consumed.

---

2. Resource Gradients (Enterprise Feature)

While Core AgentV captures raw metrics, the high-fidelity Gradient tracking (analyzing the rate of change) is available via Enterprise-tier plugins.

Identifying a Memory Leak

• Normal: Memory fluctuates as tools are loaded and unloaded.
• Anomalous: A sustained, positive linear gradient (e.g., +5MB per turn) across more than 3 turns, even during idle periods.

When a leak is detected, the Enterprise analyzer flags the session with a LOGIC_STATE_STALL or INFRA_OOM trigger in the Causal Chain.

---

3. Configuration

Telemetry is enabled by default in the Industrial Edition. You can configure the capture interval and threshold in your config.yaml:

forensics:
  telemetry:
    enabled: true
    interval_ms: 500
    capture_on_tool_call: true
    leak_threshold_mb_per_turn: 2.0

---

4. Viewing Telemetry Data

Hardware metrics are stored in the resource_telemetry field of the Forensic Ledger.

In the Visual Console, you can view these metrics as a sparkline graph overlaying the agent’s trajectory, allowing you to correlate logic spikes with hardware pressure in real-time.