Operations
Service Levels
What we target for availability, how downtime is measured, what's excluded, and what applies during the public beta.
During the public beta
There is no contractual uptime commitment and no service credits are payable. This page describes what we target and how we measure it, so you know what to expect and can hold us to it in conversation — but during the beta it is a statement of intent, not a warranty. See Beta Terms.
When the beta ends, the terms below become contractual for paid plans, and we will say so here on that date.
Availability target
99.9% monthly availability for the CogniWork web application and API.
That allows roughly 43 minutes of unavailability in a 30-day month.
How availability is measured
Availability is measured per calendar month as:
availability = (total minutes − unavailable minutes) / total minutes × 100A minute counts as unavailable when the service returns errors for, or fails to respond to, more than 5% of valid requests during that minute, as measured by our external monitoring.
What’s covered
| Component | Covered |
|---|---|
| Web application | Yes |
| REST API | Yes |
| Workspace MCP endpoint | Yes |
| Sign-in and authorisation | Yes |
| Scheduled jobs firing within 15 minutes of their time | Yes |
| AI response latency | No — see below |
| Third-party systems you connect | No |
| Messaging channels you connect | No |
AI latency is not covered. Model response times vary with provider load and with the size of the job. A worker taking longer than usual is not downtime. A worker that cannot start because the platform is down is.
Exclusions
Time does not count as unavailable when it is caused by:
- Scheduled maintenance announced at least 48 hours in advance, capped at 4 hours per month and scheduled outside 09:00–18:00 UTC where possible.
- Emergency maintenance to address a security issue, announced as early as we can.
- Your side of the line — your network, your browser, your credentials, or a third-party system you connected being down.
- Suspension under the Terms of Service or Acceptable Use Policy.
- Force majeure — events beyond our reasonable control.
- Model provider outages that we route around as fast as we can but do not control. We will always tell you when this is the cause.
Service credits (post-beta)
Once contractual, the following credits apply against the next invoice:
| Monthly availability | Credit |
|---|---|
| 99.0% – 99.9% | 10% |
| 95.0% – 99.0% | 25% |
| Below 95.0% | 50% |
Credits are the sole and exclusive remedy for failure to meet the availability target. To claim, email [email protected] within 30 days of the end of the affected month with the dates and times you observed.
Status and communication
- Incidents affecting availability are posted as they happen, not after they resolve.
- Major incidents get a written post-mortem within 5 business days: what happened, why, what we changed.
- Planned maintenance is announced by email to workspace owners and in the product.
Support response
Response-time targets are on the Support page.