Effective date: October 4, 2026
This Service Level Agreement (SLA) describes the availability commitment that Gradiated, a division of PostHog Inc., makes for paid inference and the credits you can claim if we miss it. It forms part of our Terms of Service. A separate agreement signed with Gradiated takes precedence where its SLA terms differ.
Our uptime target and commitment
Our general operating target is 99.99% monthly uptime for production inference. The contractual commitment under this SLA is 99.9%, with usage credits determined by the thresholds below.
We commit to 99.9% monthly uptime for each production model served through the Gradiated inference API. This SLA covers direct paying customers, including prepaid and invoiced accounts. Free services, models expressly marked beta or experimental, training, the website and the console are outside its scope. If you buy access through a reseller, claims go through that reseller under your agreement with them.
How we measure uptime
We measure each model separately for your organization over a calendar month in UTC. For each five-minute interval, availability is the percentage of eligible requests that did not fail because of Gradiated. Monthly uptime is the average of those interval percentages. An interval with no eligible requests counts as 100% available.
An eligible request uses a supported inference endpoint and model, valid credentials and input, and stays within your account’s published or agreed limits. Failures include server errors (HTTP 5xx), inability to connect to our API, and responses that time out or stop before completing because of a fault in our service. For timeouts, the applicable limit is the documented API timeout or a timeout agreed with us in writing.
We use our monitoring and request records together with evidence you provide to verify a claim. Deployment failures, routine maintenance and faults in infrastructure we use to deliver inference count when they cause covered requests to fail. This SLA covers availability; it does not promise a particular response speed or the accuracy of model outputs.
Usage credits
| Monthly uptime | Usage credit |
|---|---|
| 99.9% or higher | None |
| 99% to below 99.9% | 5% |
| 95% to below 99% | 25% |
| Below 95% | 100% |
The credit is that percentage of your net inference charges for the affected model during the 30 days immediately before the first covered outage in the affected month. A covered outage is an incident that causes eligible requests to fail. We use the same window for all incidents for that model in that month, so usage after the first outage cannot increase the credit. The window is 30 consecutive 24-hour periods, measured in UTC, ending at and excluding the outage’s start time.
Net charges exclude taxes, refunds and promotional usage. Prepaid deposits are counted when spent on inference, not when added to your balance. Spending on other models or services is excluded.
For example, if monthly uptime is 99.3% and your charges for the model in that 30-day window were $1,000, you receive a $50 usage credit. If you have used the model for fewer than 30 days, we use your actual charges in the window without extrapolating them. If you had no paid usage in the window, the credit is zero.
We issue SLA compensation as usage credits added to the same organization’s Gradiated balance for future inference. They cannot be transferred or redeemed for cash. One credit band applies per model per month; bands and incidents do not stack. Total SLA credits for a model cannot exceed its net charges in that same 30-day window.
How to claim
Email team@gradiated.com with the subject “SLA credit claim” within 30 days after the affected month ends. Include your organization ID, model ID, month, incident times in UTC and any request IDs, error codes or connection logs you have. Do not send API keys, prompts, outputs or other private customer content.
We will review the claim, explain our calculation and apply any confirmed credit within 30 days of receiving the information needed to verify it.
What is excluded
- Invalid requests, authentication errors, exhausted balances, client cancellations and requests above your published or agreed limits.
- Failures caused by your application, equipment, configuration or internet connection outside Gradiated’s service boundary.
- Access suspended in accordance with our Terms, including for non-payment or misuse.
- Extraordinary events beyond our reasonable control, such as natural disasters, war or government orders. Ordinary failures of our hosting providers are not excluded on this basis.
A capacity rejection within your published or agreed limits is not excluded just because it uses HTTP 429. If a Gradiated capacity fault causes it, we count it as a failure.
Relationship to our Terms
These credits are the remedy under this SLA for missing the availability commitment. Other provisions of our Terms continue to apply. Nothing here removes rights or remedies that the law does not allow us to limit.