Service Level Agreement.

"Service" means the Atlas Cloud product accessed via app.useatlas.dev, including the query API, semantic-layer service, and dashboard. The Service does not include the open-source distribution, customer-hosted deployments, or third-party data warehouses connected by Customer.

"Downtime" means any minute during which the Service returns a 5xx response or fails to respond within 30 seconds for >50% of requests originating from Customer’s account, as measured by Atlas’s external monitoring (OpenStatus probe of api.useatlas.dev plus internal regional health checks).

"Monthly Uptime Percentage" is calculated as (Total Minutes − Downtime Minutes) / Total Minutes for a calendar month, rounded to two decimal places.

"Service Credit" means the percentage of monthly subscription fees applied to Customer’s next invoice when an Uptime Target is missed.

Atlas commits to a 99.9% Monthly Uptime Percentage on the Business plan (≈ 43 minutes/month allowance). Custom enterprise contracts may negotiate higher targets — up to 99.99% (≈ 4 minutes/month) — in an order form.

If a target is missed, Service Credits are applied automatically to Customer’s next invoice. Credits do not require a claim to be filed, though Customer may request retroactive review for up to 30 days after the month in question.

Service Credits are Customer’s sole and exclusive remedy for any unavailability or non-performance of the Service. The maximum credit in any single month is 50% of that month’s subscription fees.

Atlas targets the following 95th-percentile latencies for end-to-end query execution on Atlas Cloud:

SQL generation: p95 < 5s — measured from prompt submission to generated SQL ready for execution. API response (non-query): p95 < 500ms — admin, auth, settings, and other non-agent endpoints.

Query execution latency against Customer’s warehouse is excluded from these targets — performance is determined by the warehouse, not by Atlas. Atlas does not store warehouse contents; queries pass through in-memory only.

Downtime does not include unavailability caused by: (a) scheduled maintenance announced ≥7 days in advance via atlas.openstatus.dev; (b) Customer-caused issues, including misconfiguration, exceeding rate limits, or use of unsupported model endpoints; (c) failures of Customer’s data warehouse, identity provider, or model provider; (d) force majeure events; (e) suspension of the account for non-payment or violation of the Acceptable Use Policy.

Maintenance windows are limited to 2 hours per month and are scheduled outside 09:00–18:00 in Customer’s primary business timezone where reasonably possible.

Atlas provides email support at support@useatlas.dev with the following first-response targets, measured 24×7 from the time a ticket is opened:

Severity 1 (Service unavailable, no workaround): Business — 4 hours. Custom enterprise — 1 hour. Severity 2 (Major feature broken, workaround exists): Business — 8 business hours. Custom enterprise — 4 hours. Severity 3 (Minor issue, request for information): Business — 1 business day. Custom enterprise — 8 business hours.

Custom enterprise customers receive a named Customer Success Manager and a dedicated Slack Connect channel. Sev-1 issues on custom contracts trigger immediate paging of the on-call engineer.

Atlas maintains a public status page at atlas.openstatus.dev showing current Service health, ongoing incidents, and historical uptime.

For any Severity 1 incident, Atlas will publish a written postmortem within 5 business days of resolution. Postmortems include timeline, root cause, remediation steps, and a list of preventive measures.

Customers may subscribe to status updates via email, RSS, Slack, or webhook.

Customer data stored in the Service (semantic-layer definitions, audit logs, user accounts) is replicated across at least three availability zones in the Customer’s selected region. Atlas targets a Recovery Point Objective (RPO) of 5 minutes and a Recovery Time Objective (RTO) of 1 hour for catastrophic failure scenarios.

Atlas does not store the contents of Customer’s data warehouse. Query execution is read-only and pass-through; result sets are cached in-memory only for the duration of the user session.

If Atlas misses its Uptime Target in any three consecutive calendar months, or in any four months within a rolling twelve-month period, Customer may terminate the agreement for cause with 30 days’ written notice and receive a pro-rata refund of any prepaid fees for the remaining term.