Product uptime
Opens the agent's own website and sign-in page every few minutes and records whether they answered and how fast. An agent has no test site of mine to host, so the vendor's product is what is watched.
- Status
- Collecting data
- Last checked
- not yet
- Agents on it
- 1
What the product uptime is
For website and app builders I build test sites and watch those, because a builder's homepage can stay up while its customers' sites go down. An AI agent hosts nothing of mine, so the closest thing to a customer's experience is whether the product itself can be reached and signed in to. That is what this monitor checks.
It runs on the same uptime monitor as the builders, with the agent's pages marked as the vendor's product rather than a test site. The figures appear in the agent's review once the first export has run.
Runs on: A small program at the edge of the internet, with the figures published every 3 hours. Checks of the Manus website and sign-in page started on 28 October 2026; figures publish every 3 hours.
How it works
- Every minute the monitor wakes up and checks a fifth of all the pages it watches, so each page is opened once every 5 minutes with a normal browser identity and a 20-second limit.
- A page that answers with a normal response counts as up. No answer in time, a server error or a missing page counts as down, after a second try 15 seconds later, so one dropped connection is not an outage.
- A page that refuses an automated visitor, with a blocked or rate-limited response, is left out of every figure rather than counted either way.
- If both of the monitor's own control sites fail in the same round, the round is thrown out: the fault was mine, not the vendor's.
- Every check is stored with its time, result, response code and response time. Every 3 hours the checks are exported and summed into daily and 30-day figures, and the site is rebuilt with them.
What it records
- Each check
- Time, page, result (up, down, blocked or no data), response code, response time and how many tries it took.
- Each day
- Share of checks answered, number of rounds, minutes down and the median response time.
- Windows
- The last 24 hours, 7 days, 30 days and 90 days, each with its uptime share and typical response time.
- Outages
- Start, end, length and the response that caused each one.
What it holds today
Read from the monitor’s data file when this page was built. Every reviewed agent is added when its review is made.
Manus
| Figure | Value |
|---|---|
| Pages checked | manus.im/, manus.im/login |
| Figures | the first export has not run yet |
Figure
- Pages checked
- Value
- manus.im/, manus.im/login
- Figures
- Value
- the first export has not run yet
What it cannot catch
- It checks the vendor's website and sign-in page, not the engine that runs tasks. The product can accept a sign-in while tasks queue or fail.
- It runs from one place. A problem between that place and the vendor would look like downtime, which is why the control sites exist.
- A vendor outage on these pages does not page me: alerts are reserved for my own test sites, which is what the builders' monitor is for.
Where it appears on this site
The other AI agent monitors
- Price and limits tracker
- Terms and privacy change watcher
- Trustpilot tracker
- Support response timer
- Quarterly retest schedule
- All monitors
Written by Miles Foster, Owner and tester, RankDude. How RankDude tests and grades AI agents is on the AI agents methodology page.