How to Use Slack Notification Sounds to Improve Team Resilience in Remote Work
If you carry a pager, your alerting stack already decides what fires and who it wakes. PagerDuty or Opsgenie handles routing, acknowledgement, and escalation. None of that is what this post is about.
This is about the last twenty centimetres of the chain: the sound your laptop actually makes. A correctly routed page that arrives as the same generic Slack knock as a design review comment is a page you will miss, or worse, a page you will hear and not react to because your ears have learned that the knock means nothing. Alert fatigue is usually described as too many alerts. In practice it is often too many alerts that sound identical.
What you are designing
Three properties, in this order.
Distinguishability. You need to know the tier of an alert before you look. This is the whole point. If you have to unlock a screen to find out whether you are getting up, you get up every time and burn out in a quarter.
Cut-through. Tier 1 has to be audible over a running dishwasher, from another room, or through light sleep. This mostly means a sound with energy in the 1 to 3 kHz range and a sharp attack, at a volume you deliberately set higher than everything else. Ambient chimes with slow attacks disappear under background noise.
Silence discipline. Everything not in a tier is silent. Not quiet, silent. A tiered system where tier 4 still makes a small polite noise is a system where you are still being interrupted, and the tiers are decorative.
The tiers
Tier 1: wake me
Pages routed to you as the on-call, plus anything in the active incident channel. Match on the channel prefix your alerting integration posts into, typically something like #prod-* or #alerts-page, with a wildcard so channels created mid-incident are covered without anyone editing config.
Loud, short, repetitive. Repetition matters more than length. One two-second sound is easier to sleep through than three sharp beeps over the same interval.
This tier must contain nothing else. The moment a non-paging alert lands in tier 1, the tier stops meaning "get up" and starts meaning "check whether you need to get up", which is a different and much worse signal.
Tier 2: look now, but stay seated
Direct messages from the secondary on-call, the incident commander, or the escalation channel. Warning-level monitoring output that has not tripped a page. During working hours these need attention within a few minutes; overnight they should be routed out of Slack entirely, not softened.
Distinct tone from tier 1, moderate volume. The key is that it should be impossible to confuse with tier 1 when heard from another room. Change the interval or the number of notes, not just the pitch. Pitch differences collapse over distance.
Tier 3: context
Deploy notifications, status page updates, the noise that tells you what changed just before things broke. This is genuinely useful during an incident and genuinely useless the rest of the time.
Low volume, soft attack. Or better: no sound at all, and you open the channel when an incident starts. Most on-call engineers who think they want a deploy sound actually want a deploy timeline.
Tier 4 and below: silent
Everything else. Team channels, threads, product discussion, the standup bot. Silent for the whole rotation.
What must never make a sound at 3am
Write this list down before you configure anything, because it is easier to reason about in daylight.
Anything with a runbook step of "check in the morning". Anything from a channel where the fix requires a person who is asleep. Non-production environments, no matter how alarming the message. Automated reports. Anything that has fired more than five times this month without anyone acting on it, which is a broken alert wearing a page's clothing and should be fixed upstream rather than muted downstream.
Testing before you need it
Send yourself a test message in each tier channel and stand in the next room. If you cannot name the tier without looking, the sounds are too similar. Do this at the start of a rotation, not during one.
Also verify handoff. The most common failure is an engineer whose rules were built for a previous rotation and no longer match the channels their alerts land in now.
Making the tiers audible
Chirpy is a native macOS menu bar app that plays a sound based on rules you write. It reads incoming macOS notifications through the Accessibility API, locally, so it does not connect to your Slack account and never sees message content on a server. Install, grant Accessibility permission, set Slack's own notification sound to None, then build your rules.
It does not page, escalate, or replace anything in your on-call stack. It makes the pages you already get sound different from each other, which is the part your alerting tool cannot do. Rules match on app, channel, person, or keyword with AND/OR and wildcards, run in priority order, and carry per-rule volume. Rule syntax is in the guides.