What we commit to.
Per plan: how much of each month your deployed tools answer, how fast a person answers you, and what you get back when we miss. Measured, not estimated.
Last updated: October 4, 2026
Targets by plan
| Plan | Monthly uptime | Downtime it allows | First response | Severity 1 response | Support |
|---|---|---|---|---|---|
| Free | No commitment | — | Best effort | Best effort | Best effort, by email |
| Starter | 99.5% | about 3.6 hours | 16 business hours | 8 hours, any day | Email, business days |
| Team | 99.9% | about 43 minutes | 8 business hours | 2 hours, any day | Priority email, business days |
| Enterprise | 99.9% | about 43 minutes | 4 business hours | 1 hours, any day | Named contact |
Downtime is shown for a 30-day month. Business hours are 09:00 to 18:00 India Standard Time, Monday to Friday, outside public holidays in India. A response is a person replying about your problem, not an automatic acknowledgement; it is not a promise of a fix in that time.
What the uptime target covers
The serving API: the REST and MCP calls your agent makes to a deployed tool with a tool key. That is what your product depends on, so that is what we commit to. The dashboard, imports, runs and the workspace MCP are not covered by the uptime target; problems with them are still support requests at your plan's response times.
How it is measured. An external monitor checks the serving API every 3 minutes. A check fails when the API answers with a server error (5xx) or doesn't answer within 15 seconds. Downtime runs from the first failed check to the next check that succeeds. Monthly uptime is the share of the month that was not downtime. What the monitor sees is public, with its history, at status.cloudcrane.ai.
A monitor only sees what it checks. If your tool calls failed with server errors while the monitor saw the API up, that time counts too: send us the request ids from those failed calls (every response carries one in X-Request-Id) and we count it from our own logs.
Not counted as downtime:
- Scheduled maintenance announced by email at least 48 hours ahead, up to 4 hours a month in total
- Your own model provider, your network, or anything else outside CloudCrane
- Requests refused because they are over a key's or a plan's rate limit or quota (429), or are otherwise invalid (other 4xx answers)
- Suspension for a breach of the terms or for abuse
- Features labelled beta or planned
Severity
- Severity 1
- A deployed tool is down, or is returning results that break a safety rule: a record that should have been excluded, or a safety value missing that the release had.
- Severity 2
- Degraded or partial: a tool answering slowly or failing for some calls, or a run, sync or release that can't complete.
- Severity 3
- A question, a minor bug, or a request for something new.
Raise any of them by email to support@cloudcrane.ai from an address in the workspace, with “Severity 1” in the subject when it is one. Include the request id: every API response carries it in its X-Request-Id header, and the dashboard shows it beside any server error.
Service credits
When a paid plan's monthly uptime falls below its target, the workspace gets a credit against a later month's fee:
| Monthly uptime | Credit, of that month's fee |
|---|---|
| Below the target, 99% or more | 10% |
| 95% to below 99% | 25% |
| Below 95% | 50% |
To claim one, email support@cloudcrane.ai within 30 days of the end of the month, with the workspace and the times or request ids affected. Credits apply to future fees, are not paid out as cash, and are the whole remedy for a missed uptime target. A month's credits never exceed half of that month's fee. The Free plan has no uptime target, so no credits.
Changes
We will email workspace owners at least 30 days before lowering any target here. A paying workspace keeps the terms it signed up under until the end of the month it has paid for. Enterprise agreements can set their own targets, and where they do, the agreement wins.