Skip to content
TriageAI

AI incident root cause analysis

From the page to the line of code

When production breaks, TriageAI works the outage like a senior SRE. It finds what changed, where the failure started and the exact code behind it, then hands you a fix and a regression test.

Read-only. Never changes production. You approve every fix.

  1. The pagePagerDuty, 14:06

    CRITICAL checkout 5xx rate 14%

  2. What changedArgo CD, 14:02

    payments-api v3.8.0 deployed 2 minutes before errors started. Strong match.

  3. The line of codepayments-api

    src/db/pool.ts:12

    - const pool = new Pool({ max: 40 });+ const pool = new Pool({ max: poolSizeFor(replicas) });
  4. The regression testpool.test.ts

    it("stays under the database connection limit", () => {
      expect(poolSizeFor(12) * 12).toBeLessThanOrEqual(300);
    });
Example analysis. Every step links to the evidence behind it.

Outages are slow to explain, and slower to actually fix

  • The first half hour goes on looking things up

    Is it real? Who's affected? What changed? Every outage starts with the same questions, answered by hand across a handful of tools.

  • Knowing where to look lives in a few heads

    Senior engineers know which graph to open first. Everyone else on the rotation learns it the hard way, at 3 a.m.

  • Finding the deploy isn't the end

    Rolling back stops the bleeding. Someone still has to find the line that broke, write the fix and the test, usually the next morning, from memory.

One investigation, from the alert to the fix

Most tools stop at “this deploy did it”. TriageAI keeps going until there's a change you can merge and a test that proves it.

  1. Step 1

    Scope the impact

    Checks the alert against real user impact, finds the first bad data point and who is affected.

  2. Step 2

    Find what changed

    Ranks every deploy, config change, flag flip and infrastructure event by how well it fits the start.

  3. Step 3

    Find where it starts

    Reads metrics, logs and traces, and walks the dependency graph to the component where the failure begins.

  4. Step 4

    Go to the line of code

    When the cause is in code, it follows the change to the file and line, then hands you a fix and a regression test.

Bug triage

Not every bug is an outage

The same investigation works from a Jira ticket, Sentry error, GitHub issue or pasted stack trace. You get the root cause, a fix and a regression test, usually in about a minute.

  • Root cause

    What broke, why, and the exact file and line, naming the commit if one introduced it.

  • Suggested fix

    The smallest change that resolves it, as a before/after diff.

  • Regression test

    A test that fails on the buggy code and passes with the fix.

Built for Kubernetes teams, working with the tools you already have

Incidents

  • AlertsAlertmanager, PagerDuty
  • MetricsPrometheus, Datadog
  • Logs and tracesOpenSearch, Loki
  • DeploysArgo CD, Kubernetes

Bug reports

  • JiraIssue webhooks
  • SentryError events
  • GitHubIssues and code
  • Raw log pasteAny stack trace

Safe to point at production

Read the security practices
  • Read-only by design

    Read-only access to your code and observability tools. It can look at everything and change nothing.

  • You make the calls

    Mitigations are suggestions. An engineer approves and runs every rollback, restart and fix.

  • Evidence you can check

    Every claim links to the graph, log line, deploy or commit behind it. When it isn't sure, it says so.

Questions

Does it replace the on-call engineer?

No. It does the evidence gathering an experienced engineer does first, in parallel and without switching tools. Severity, escalation, mitigation and the final call on the cause stay with you.

Does TriageAI roll back or restart anything?

No. TriageAI has read-only access and never changes production. Remediation steps are suggestions an engineer approves and runs, and there is no auto-merge for code fixes either.

What does it need access to?

Read-only access to your alerting, metrics, logging and deploy tools for outage analysis, and to a clone of your repository for the step into code. Secrets are redacted before anything reaches the model.

Which stack does it work with?

It's built first for Kubernetes teams using Prometheus, Loki or OpenSearch, Argo CD and GitHub, with PagerDuty or Alertmanager for paging. Datadog, CloudWatch and Sentry are supported inputs, and more integrations are on the roadmap.

Can I use it for bugs that aren't outages?

Yes. Give it a Jira ticket, Sentry error, GitHub issue or raw stack trace and it returns the root cause with file and line, a fix and a regression test. That's the same last step every outage analysis ends with.

What if it can't find the cause?

It says so. You get the timeline and the strongest candidates with their evidence for and against, plus what it would check next, instead of a confident guess.

Watch TriageAI work a sample outage

Follow it from the alert to the change behind it and the exact line of code, then get the fix and a regression test. You review every step.