AI diagnostics

Skip the guesswork during an incident.

When something goes wrong, PrimesFlow puts together a plain-language summary of the incident from your infrastructure signals. It's built to cut down investigation time, not to replace your judgement.

AI-generated insights are assistive. PrimesFlow drafts a hypothesis based on the signals it can see. Always verify with your own logs and dashboards before acting.

Metrics
Logs
Deploys
Diagnosis

What's inside a summary

Every diagnosis follows the same structure

So whoever is on call always knows where to look first.

What happened

A short natural-language description of the observed behavior — e.g., 'API host web-03 experienced a sustained 8-minute CPU saturation.'

When it started

The first metric sample that crossed threshold or diverged from the baseline, with a precise UTC timestamp.

What was affected

The specific hosts, services, and containers whose telemetry moved with the incident.

Likely cause

Correlated signals: memory pressure, disk I/O spike, restart loop, a recent deploy marker, or an unusual traffic pattern.

Suggested next steps

Concrete checks to run — commands, logs to review, dashboards to open. Always assistive, never automated remediation.

What it looks at

Only infrastructure signals, nothing more

Metric signals

CPU, memory, disk, network, load, and per-service resource usage across the affected hosts.

Log signals

Error-rate and restart signals from systemd units and container runtimes — line counts and severity, not payloads.

Recent deploy markers

Optional deploy events you send via the PrimesFlow API, correlated to the incident window.

Example

What a diagnosis looks like

A representative draft summary, shown as it appears inside the PrimesFlow dashboard.

incident-4821.diagnosis

// Draft — verify before acting

What: Sustained CPU saturation on web-03.

Started: 2026-07-28 14:22:10 UTC (~8m).

Affected: web-03, nginx.service.

Likely cause: Memory pressure preceded CPU spike; a deploy marker arrived 90s prior.

Next steps: Check the deploy diff, review journalctl -u nginx -n 200, and inspect worker restart counts.

Which plans include AI diagnostics

  • Pro — AI diagnoses on active incidents.
  • Team — AI diagnoses, weekly reports, and auto-drafted postmortems.
  • Hobby — not included.
Compare plans →

Data boundary

PrimesFlow reasons only about infrastructure signals the agent already collects: host and service telemetry, restart counts, error-log line rates, and any deploy markers you choose to send via the API.

It never reads your application source, request bodies, database rows, or end-user data.

Put AI-assisted diagnostics to work

Included on Pro and Team plans. 14-day free trial, no credit card required.