Nine words carry most of the product. This page defines them and, where two sound alike, says what separates them.
The four things that can fail#
A monitor answers "is this reachable and healthy right now". ObserveOne runs it from its own infrastructure on your schedule. There are five types: URL, SSL certificate, TCP port, UDP port, and database.
An API check is a saved HTTP request plus assertions about the response. It goes further than a monitor: you control the method, headers, body and auth, and you decide what counts as a pass. Status code, latency ceiling, a value inside the JSON body. A monitor tells you the endpoint responded. A check tells you it responded correctly.
A heartbeat inverts the direction. Nothing is probed. Your job calls an ObserveOne ping URL when it succeeds, and if that call does not arrive inside the period plus grace you set, the missing ping is the failure. Use it for cron jobs, backups, and anything else that runs where ObserveOne cannot reach in.
An Autopilot suite is a set of Playwright tests generated from a URL. You give it a site, it plans scenarios, writes the specs, and runs them on a schedule. When your markup changes and a selector stops matching, healing rewrites the selector and saves a new version of the spec. Edits are restricted to the locator, so a heal cannot quietly change what the test asserts. The dashboard calls this Autopilot; the CLI calls the same objects suites (obs suite).
The things that organise them#
A project is a container for monitors, checks and environments. Most accounts need one. Reach for a second when you want a hard boundary, like separating a client's resources from your own.
An environment holds variables and a base URL that resources reference instead of hardcoding. Store API_TOKEN and a staging base URL in one environment, the production pair in another, and point the same check at either. Values are referenced as {{KEY}} and resolve server side when the request runs, so a secret never sits in the saved request or in an export.
A schedule is how often something runs, written as a cron expression or picked from the dashboard. Your plan sets the floor: on Free, nothing runs more often than every two hours.
The things that happen when something breaks#
A run (the API and CLI say execution) is one attempt. It has a result, a duration, and for a browser test, a trace you can replay.
An alert channel is a destination. Seven types: email, Slack, Discord, Microsoft Teams, Telegram, SMS, and webhook. Channels attach to monitors and checks, and one channel can serve many.
An incident is a record you work through. It opens, takes an assignee and comments, then resolves or closes. An alert fires once and is gone; an incident stays until someone closes it. Failures can open one automatically, or you can create one by hand during an outage.
A status page publishes chosen monitors and incidents at a public URL, so customers can check for themselves.
Two pairs people mix up#
Environment variables in an environment are not the OBS_* variables in your shell. The first are values your checks reference at runtime. The second configure the CLI itself, like OBS_API_KEY. See Environment variables for the CLI ones.
AI verifications and AI credits are separate allowances. A verification is one Autopilot suite run. Credits pay for the AI work behind planning a suite, generating a test, and healing a broken one. Free accounts get 10 verifications a month and no credits, which is enough to run a small suite but not to keep regenerating it. Credits can be bought as a pack that never expires.
Next steps#
- Cloud quickstart: a monitor and an alert in five minutes.
- API Checks overview: build your first check.
- Config-as-code: keep all of the above in a file in your repo.