Skip to main content

​
Insaion Copilot Incident Analysis

Insaion Copilot provides guided incident triage directly from the incident drawer. It combines alarm context, reduced telemetry summaries, and packet-level analysis (when needed) into one continuous investigation flow.

​
What Copilot does during incident analysis

When you launch analysis from an incident, Copilot can:
  • Build a focused investigation context from incident metadata and alarm rule details.
  • Correlate relevant logs, metrics, and signal changes around trigger time.
  • Request an MCAP extraction window when baseline evidence is not conclusive.
  • Continue automatically with deep packet-level analysis after MCAP artifacts are ready.
  • Stream findings into the conversation and persist a structured report for later review.
Copilot output accelerates troubleshooting, but teams should always validate conclusions against raw data and domain knowledge.

​
Start analysis from an incident

  1. Open the Incidents page.
  2. Select the incident to investigate.
  3. Open the action menu.
  4. Select Analyze with Copilot.
Copilot opens the incident analysis view with context preloaded and direct access to the embedded Lichtblick visualizer.

​
End-to-end analysis flow

  1. Copilot starts baseline triage on incident context and reduced telemetry.
  2. If baseline evidence is conclusive, Copilot returns a root-cause summary immediately.
  3. If baseline evidence is inconclusive, Copilot prepares an MCAP extraction request with a scoped time window and relevant pipelines.
  4. After approval, the device uploads the window and the artifacts are registered. Incident uploads go before the device’s other uploads, and a window is at most 15 minutes long.
  5. Copilot resumes deep analysis automatically and streams report updates in real time.
  6. The final report is stored in the thread and available on reconnect.
  1. Confirm incident timing, status, and affected entities.
  2. Review Copilot summary and suggested related signals.
  3. Validate key hypotheses in Lichtblick plots and related logs.
  4. Approve MCAP extraction when requested and follow it in the analysis view or the Recording uploads panel.
  5. Review or export the generated incident report.
  6. Acknowledge or resolve the incident based on validated findings.

​
When the device is offline or the data is gone

If the device is offline when you approve an extraction, the request waits in the cloud and the upload starts automatically when the device reconnects. Until then, the analysis view shows that the device is offline and the upload is queued. The extraction can only upload what is still in the device’s rolling buffer. With a camera recording set and the default 1 GB buffer, that can be only a few minutes of history: analyze incidents promptly, or raise the buffer size in Device configuration. If the data was already overwritten, the upload fails with the reason “recorded data was overwritten before it could be uploaded”.

​
AI credits and limits

Copilot analysis consumes AI credits. If your organization reaches its limit:
  • New AI-assisted analysis requests are paused.
  • Manual analysis remains available with the same incident context.
For usage and limits, see the Billing page.

​
Best practices

  • Keep alarm names and descriptions specific to improve Copilot context quality.
  • Add notes and labels during triage to preserve investigation history.
  • Resolve incidents only after confirming the underlying condition is cleared.