Legal
Service levels and support
What we commit to when we host Neural Inverse for your organisation, and how our support team responds, whether we host it or you do.
[Draft: to be approved by legal counsel before publication]
This page is for information. Your order form and agreement with us govern; where they differ from this page, they prevail.
What this covers
The availability targets and service credits apply to the service we host: the shared service, and dedicated deployments we run for one organisation. They do not apply to trials and evaluations.
When you run Neural Inverse yourself, on your servers, in your cloud account or air-gapped, we support the software and you run the service. See self-hosted deployments.
Availability targets
Each calendar month, we aim to keep the service available for at least:
- Shared service: [shared availability target: to be confirmed]
- Dedicated deployment: [dedicated availability target: to be confirmed]
How availability is measured
We check the service from outside the network it runs in, every [check interval: to be confirmed]. A check passes when the service answers its health check correctly and in time. The service counts as unavailable from the start of a run of failed checks lasting at least [downtime window: to be confirmed], until checks pass again.
We measure the region of the shared service your organisation uses, or your dedicated deployment. Monthly availability is the minutes in the month, less the minutes unavailable (not counting the exclusions below), as a percentage of the minutes in the month.
Exclusions
Time unavailable for these reasons does not count against the target:
- planned maintenance announced as described below;
- causes on your side: your configuration, your identity provider, your network or firewall, or domains and certificates you manage;
- outages of services you connect that we do not run, such as an AI model provider or a source code host, while the service itself stays available;
- features marked as preview or beta;
- suspension under your agreement or the acceptable use policy;
- events beyond our reasonable control, as your agreement defines them.
Maintenance
We announce planned maintenance that may interrupt the service at least [maintenance notice: to be confirmed] ahead, on the status page and by email to the contacts your organisation names. Planned maintenance is limited to [maintenance per month: to be confirmed] a month.
An urgent security fix may need shorter notice. Time unavailable during it counts against the target.
Service credits
If monthly availability falls below the target, you can claim a credit against your fees for the service we host:
- credit: [credit schedule: to be confirmed];
- the most in any month: [credit cap: to be confirmed];
- claims within [credit claim window: to be confirmed], through support, with the dates and times affected.
Credits are applied to your next invoice and are not paid in cash. Your agreement sets out whether they are your only remedy for missing the target.
Support severity levels
When you report a problem, tell us how severe you think it is; we agree the level with you and may change it as we learn more. The first response is a reply from an engineer working on your problem, not an automatic acknowledgement.
| Level | What it means | First response |
|---|---|---|
| P1 | The service is down, or a function people depend on (signing in, the model gateway, policy checks) does not work for most of your users, and there is no workaround. | about 45 minutes from the time of your request |
| P2 | An important function is failing or badly degraded for many users, or a P1 problem has a workaround that is hard to keep up. | [P2 response: to be confirmed] |
| P3 | A function is not working as documented, but most work carries on, or there is a reasonable workaround. | [P3 response: to be confirmed] |
| P4 | A question, a documentation problem, a cosmetic fault or a request for a new feature. | [P4 response: to be confirmed] |
We keep working on P1 problems until the service is restored or a workaround is in place, and keep you updated as we go.
Support hours
[support hours: to be confirmed]
Contacting support
Use the contact form and choose “Support for an existing customer”, or the support channel named in your agreement. Include your organisation, whether you use the shared service, a dedicated deployment or your own, the severity, what happened and when.
To report a security vulnerability, do not contact support: email security@neuralinverse.com and see our vulnerability disclosure policy.
Service status
The status page shows the current state of the service we host, incidents while they are open, and planned maintenance.
Self-hosted deployments
When you run Neural Inverse on your own infrastructure, its availability is yours: you choose the hardware, network, database, backups and monitoring, and the availability targets and credits above do not apply.
We support the software. The severity levels, response targets and support hours above apply to problems in it, for the releases your agreement covers. We help you diagnose problems, fix defects in new releases, and publish fixed vulnerabilities as security advisories.