Incident intelligence

Stage in the incident loop: DECIDE · LEARN

Explainable incident intelligence.

CallHeim suggests a severity from a keyword ruleset and your own resolved incidents, shows the terms that matched, and a person decides whether to apply it.

Rule-basedDeterministicEvery suggestion shows its reasons.

How it works

Five steps, each with a rule you can read.

  1. 01

    Fingerprint and collapse

    Every alert gets a fingerprint from its source and its identity. A repeat attaches to the incident that is already open for it: at any age when the sender supplies its own dedup key, and within 24 hours when only the title identifies it.

    How grouping works →

  2. 02

    Severity suggestion

    A fixed keyword ruleset, weighted by severity, reads the alert and your own resolved-incident history and suggests a level with the matched terms shown. A person applies it.

  3. 03

    Related-incident insights

    On Pro and above, a related-incidents panel ranks other open incidents by how closely their alert text and service match. It is a starting point for someone triaging, not an automatic merge.

  4. 04

    Fatigue score

    A published formula scores on-call load from your own paging history, so an uneven rotation is visible before someone burns out.

  5. 05

    MTTA and MTTR

    Response-time metrics come from each incident’s own timestamps, shown as mean, p50 and p90.

    How incident response measures it →

Why rules

Every suggestion comes with its reason.

CallHeim’s alert handling is rule-based and deterministic: given the same alert, settings and state, it makes the same decision.

A ruleset returns the terms it matched and their weights, so a wrong call comes with its reason.

See every noise rule and its default →

Published thresholds

Repeat, sender-supplied dedup key
attaches, any age
Repeat, title or subject only
attaches, incident under 24 h old
Flap threshold
4 in 600s
Title-similarity threshold
0.6

These thresholds are fixed. Title grouping can be switched off for your organization.

Severity suggestion

Shown with its reasons, applied by a person.

CallHeim suggests a severity from a fixed keyword ruleset plus your own resolved incidents, shows the matched terms, and a person applies it. Suggestions never change severity or page anyone.

Example

Alert text: “checkout-api: payment authorization timeout.”

“p1”
P1 signal
“latency”
P2 signal
“timeout”
P2 signal

Suggested: P1. A person opens the incident and applies, changes or ignores it.

The suggestion shows the terms that matched and their weights, so you can see why it chose a severity. The keyword list is fixed by CallHeim. You can apply or change the result yourself, or set severity with your own orchestration rules.

The ruleset scores the alert from P1 to P4 and defaults to P3 when nothing matches. The severity your source reported and the outcomes of your own similar resolved incidents nudge the result, and ties go to the more severe level.

For the same alert, the same rules and the same incident history, the result is identical every time. The historical signal comes from your own resolved incidents, inside your own organization.

Your data

Where suggestions run.

No incident data is sent to an external language model, and CallHeim does not train any model on your incidents. Severity suggestions and related-incident insights are rule-based code running inside CallHeim’s own AWS environment.

CallHeim ships no generative assistant and no autonomous agents.

Page e-mail carries incident details to your responders. See where data goes on the security page →

Alert correlation

Related incidents, ranked for a person to check.

The panel ranks open incidents by how closely their alert text matches. Candidates within a 30-minute window are scored 0.7 on cosine similarity of the text plus 0.3 on Jaccard word overlap, with a boost for the same service.

The result is a ranked list on the incident page for a person to read. CallHeim never combines incidents on its own.

Plan

Related-incident insights

Pro and above

Core analytics are not gated by plan. Related-incident and responder insights start at Pro. Compare plans →

On-call fatigue

Who is carrying the paging load.

A published on-call fatigue score: 0.40 sleep-window interruptions, 0.30 volume, 0.20 off-hours, 0.10 severity, over a rolling 14 days. It is a heuristic over your own paging history, not a wellbeing measure.

An uneven rotation shows up on a page instead of staying invisible until someone complains. Rotation assist can use the score to propose a fairer rotation, and nothing changes until a person approves it.

See on-call scheduling →

CallHeim

Put a rule you can read between your alerts and your on-call.

Start a 14-day trial, connect one alert source and send yourself a test page.