Docs
Get Started

How InsightX works

Detecting an incident

A small health agent runs on each server that hosts your app. It watches your app's containers, memory, CPU, disk, and SSL certificates. When something is wrong, it sends an event to Kuberns, and Kuberns records the incident.

Explaining a crash

For a crash, an out-of-memory kill, or a failing health check, InsightX gathers evidence from your server and an AI model investigates it. The model can look at:

  • The last lines of your app's logs from the minutes around the incident.
  • The processes running in the container.
  • The container's memory limits and settings.

Every lookup is listed under What InsightX checked on the incident, so you can see what the explanation is based on.

Explaining a measured condition

A full disk, an expiring certificate, high CPU, or high memory is measured directly, so the cause is already known. These incidents get a fixed explanation with the measurement as evidence. No AI model is involved, so there is no confidence figure.

Secrets in your logs

Before logs are stored or sent to an AI model, Kuberns masks credentials in them. This covers:

  • Values whose names say they are secret, such as PASSWORD, TOKEN, SECRET, and API_KEY.
  • Authorization headers, cookies, and private keys.
  • Passwords inside connection URLs.
  • Well-known token formats, such as GitHub, GitLab, Stripe, AWS, Slack, and Google keys.

Note: Masking is best effort. No list of patterns can know every secret format, so avoid printing secrets to your logs.

When InsightX stays quiet

InsightX does not record an incident:

  • While a deployment is in progress, or in the 10 minutes after a deployment succeeds, because your app is expected to restart then.
  • When the same kind of incident was already recorded in its time window. See How often an incident repeats.