Uptime & SLA

Availability commitments

| Plan | Commitment | |---|---| | Launch | Best effort, with email support | | Scale | 99.9% monthly uptime SLA, priority support | | Enterprise | 99.95% monthly uptime SLA, named technical account manager |

Uptime is measured on your instance’s web app and API being reachable and healthy. Planned maintenance announced in advance and problems caused by third parties you connect (model providers, messaging platforms, your own systems) are excluded. Service credits and the exact terms are defined in your subscription terms or Enterprise contract.

Maintenance and updates

Your instance is pinned to an ExoHuman release. We move it to newer releases in planned windows:

1. Waves. A release reaches instances in waves, starting with our own, so problems are found before they reach you.
2. Announcement. Administrators are notified before the window with what changes and whether anything needs their attention.
3. Quiet window. Windows are chosen for low usage in your region.
4. Drain. Running executions are allowed to finish; schedules due inside the window are held and run afterwards; queued work waits.
5. Backup. A fresh backup is taken before anything restarts.
6. Update and verify. Platform services restart on the new release and are checked: health, version, sign-in, a real agent reply, and the behaviour the release changed.
7. Watch period. We monitor for 30–60 minutes. Any failed verification, new error class or user-visible regression triggers an automatic rollback to the pre-update backup.

What you notice: a brief reload of the web app and a few minutes of unavailability at restart. Agent containers keep running through the update and adopt the new agent base image the next time they are recreated.

Database migrations only move forward, so a rollback is a restore of the pre-update backup — which is why that backup is always taken.

Incidents

• We monitor every instance and respond to incidents 24×7 for Scale and Enterprise.
• Incidents affecting your instance are communicated to your administrators by email with updates until resolved.
• After a significant incident we share a summary of cause and remediation.

To report a problem, email [email protected] with your instance URL and, if possible, the time and an execution ID. See Support.