How OnCallAlert Works: What Runs Where and What Leaves Your Server

OnCallAlert is a program you run yourself. Our source code is private, so you cannot read it to check what it does. This paper is the next best thing: a plain account of how the product is built, every call it makes to our services, what each call carries, and what happens when we are unreachable. It is written for engineers and security reviewers.

The short version

What runs on your server, and what runs on ours?

You run one container and one PostgreSQL database. That container is the whole product: web UI, API, scheduling, escalation, alert delivery and the engine that decides who gets paged. We host three small services on Cloudflare, and each one is optional or narrow in scope.

ComponentWhere it runsJob
OnCallAlert server and databaseYouEverything: alerts, schedules, escalation, users, notifications, UI, API
License serviceUsConfirms your key and hands back a signed plan token
Push relayUsWakes the mobile app by passing two random IDs to Apple and Google
Dead-man switchUsPushes your phones if your server stops checking in (optional)
Status page hostUsServes the public status page you choose to publish (optional)

The server is a fork of the open-source GoAlert project, with licensing, the mobile app and the extras added on top. The scheduling and escalation core is the same proven design.

What exactly does the server send to OnCallAlert?

Four things, listed here with their fields. Each of these is also stated in the privacy policy, and the two documents are kept in step.

CallWhenWhat it carries
License checkAt startup, then about once a day. Always on.Your license key, the public URL of your instance, a random install ID, the server version, and the number of user accounts (a count only)
Mobile pushOnly when the mobile app is enabled, and only when a phone needs wakingA random device ID and a random alert ID. On pairing: the phone's push token and platform
Dead-man switchOnly if you switch it on, at the interval you choose (no more often than every 5 minutes)Your instance key, your chosen interval and miss count, and the random device IDs to alert
Status pageOnly if you publish one, when it changes and every 15 minutes while a component is downThe page content you wrote, which is public by design

What never leaves your server?

One exception to name plainly: the web interface has an optional setting where an admin can enter a Google Analytics ID for their own users. It is off unless an admin sets it, and it sends nothing to us.

How does mobile push work without seeing my alerts?

The relay is a doorbell, not a mailbox. When an alert needs to reach a phone, your server tells the relay a device ID and an alert ID. Both are random identifiers. The relay forwards them through Google Firebase Cloud Messaging (and Apple's push service on iPhones). The app wakes up and then fetches the alert directly from your server. On iPhones the lock-screen banner is a fixed generic line telling the user to open the app, so no alert text passes through Apple's or our systems.

For each paired phone the relay keeps the device ID, the push token, the platform, the instance it belongs to and timestamps. A phone's record is removed when it is unpaired, or when the push provider reports the token dead. The relay stores no alert IDs and no alert content.

What happens if your servers are down or my network is blocked?

Alerting does not depend on us being reachable. The license is a signed token that your server checks locally, so a failed license call changes nothing by itself. Network errors and temporary failures on our side are simply retried later.

SituationWhat happens
Our license service is unreachableAlerting, the UI and the API keep working. The server retries and keeps using its last valid token
Your license expiresA 30-day grace period starts. Everything keeps working
Grace period ends with no valid licenseThe server locks. The alert engine pauses and the API returns a payment-required error. Login and the license status page stay reachable
All outbound traffic blocked on a running installCore alerting works. Mobile push, the dead-man switch, hosted status pages and any cloud AI provider stop. Email, SMS and voice work only if your own providers are reachable
Brand-new install with no route to our license serviceIt cannot activate, so it stays locked until it can reach the license service once

We would rather be honest about the last two rows: a license is the one thing that can stop alerting, and only after a lapse of more than 30 days or an install that never activated. The server also guards against a rolled-back system clock when it judges expiry.

How is the AI feature kept private?

Ask AI uses your own key with the provider you pick: OpenAI, Anthropic, a local Ollama, or any OpenAI-compatible address. The key is encrypted in your own database. Requests go straight from your server to that provider. They never pass through us. Each request carries a trimmed alert summary, its details and the service name, plus a few recent summaries for context. Point it at a local model and no alert text leaves your network at all.

How is access to the server protected?

What do we not promise?

How can you check these claims yourself?

You do not have to take our word for it. Three checks take about an hour:

  1. Watch the traffic. Run the container behind an egress firewall or a logging proxy for a day. You should see only the license check, plus any optional feature you turned on, and your own providers.
  2. Block us and see. Cut all outbound traffic to our domain on a running install. Alerts, schedules and escalation should carry on, as described above.
  3. Read the policies. The privacy policy and terms list the same flows. If code and policy ever disagree, that is a bug we want to hear about.

Why publish this without the source code?

The code is private, and this paper is not a substitute for reading it. What it gives you is a set of specific, testable statements. If you find traffic or behaviour this paper does not describe, tell us at the contact address on the site and we will fix the paper or the product. See also how OnCallAlert compares with a hosted service, the documentation, and how the optional status page works.

Share this post:

Want to check it yourself? Read the privacy policy, then try OnCallAlert behind your own firewall.

Privacy policy → Pricing & plans

Sources and notes

  • Statements in this paper were checked against our own server and hosted services on 2026-11-14 and match the privacy policy of the same date.
  • Behaviour can change between releases. We update this page and the privacy policy together.