Service Level Agreement

1. Commitment

GridPort will make the serverless inference API available at least 99.9% of each calendar month (the "Monthly Uptime Percentage"). If we miss it, you receive the service credit below.

The SLA applies to organisations with a paid top-up or subscription in the month concerned. Usage paid for only with trial or promotional credit is not covered.

2. What is covered

Covered models are those listed in the public catalogue as generally available — not beta, preview or deprecated. Covered requests are well-formed requests to a covered model through the OpenAI-, Anthropic- and Gemini-compatible endpoints.

Not covered: dedicated endpoints, fine-tuning and training jobs, batch jobs and asynchronous jobs (which have their own completion windows), and anything marked beta or preview, unless an order form says otherwise.

3. How uptime is measured

Error rate: for your organisation, the share of covered requests in a five-minute interval that fail with a server-side error (an HTTP 5xx status). Errors caused by the request itself, rate limits, spending limits or balance, and refusals by the model are not server-side errors.

A five-minute interval counts as downtime when your error rate in it is above 5%, or when our external probes could not reach the API during it — so an outage in which requests never reach us still counts. Any other interval counts as available.

Monthly Uptime Percentage = 100% minus the share of the month's five-minute intervals that were downtime. We compute it from the gateway's request records, which you can inspect in the console's request log, and from our external probes, whose results feed the status page.

4. Service credits

The credit is a share of what you spent on covered requests in the month the SLA was missed:

• Below 99.9% but at least 99.0%: 10%
• Below 99.0% but at least 95.0%: 25%
• Below 95.0%: 50%

Credits are added to your balance as non-refundable credit, capped at 50% of the month's covered spend, and are the sole remedy for unavailability.

5. How to claim

Write to [Support address — to be stated before these terms take effect] within thirty days after the end of the month, from the account owner's address, with your organisation ID and the times or request IDs affected. We confirm or decline the claim within ten business days, with the measurements behind the decision.

6. Exclusions

Downtime does not include time caused by:

• maintenance announced in the console at least 48 hours ahead, up to four hours a month;
• your own systems, network or code, or credentials you exposed;
• suspension under the Terms of Service or the Acceptable Use Policy;
• events beyond our reasonable control, such as a failure of the public internet or a natural disaster.