Service Level Agreement

This Service Level Agreement ("SLA") sets out RadMah AI's availability commitments for the RadMah AI platform and defines the service credits available in the event those commitments are not met. This SLA applies to: — RadMah AI Sovereign plan subscribers (billed directly through RadMah AI) — Enterprise plan customers (subject to negotiated SLA terms that may differ from or supplement this SLA) This SLA does NOT apply to: — Free tier users (see Section 2) — Beta features, preview environments, or experimental engine modes explicitly designated as such This SLA is incorporated into and forms part of the Terms of Service (www.radmah.ai/legal/terms). In the event of any conflict, the Terms of Service govern.

The Free tier is provided on a best-efforts basis. RadMah AI makes no uptime commitment, response time guarantee, or service credit commitment for Free tier users. Free tier generation and training jobs run on AWS Fargate Spot capacity. Spot instances are subject to interruption by AWS capacity reclamation without advance notice. Interrupted Spot jobs will be automatically retried once if queue capacity allows, but successful completion is not guaranteed within any specific timeframe. Free tier users who require guaranteed availability, predictable job completion times, and priority queue access should upgrade to the RadMah AI Sovereign or Enterprise plan.

3.1 Definitions: "Monthly Uptime Percentage" means (Total Minutes in Month − Downtime Minutes) ÷ Total Minutes in Month × 100%. "Downtime" means any period (of 5 or more consecutive minutes) during which the applicable service endpoint returns error responses at a rate of 95% or greater of all requests from a given customer, excluding errors caused by Excluded Events (Section 6). "Scheduled Maintenance" means planned downtime announced at least 48 hours in advance via email to registered account contact addresses and posted on the RadMah AI status page at status.radmah.ai. "Emergency Maintenance" means unplanned maintenance required to prevent immediate harm to platform security or availability. Emergency maintenance may occur without advance notice. 3.2 Monthly Uptime Targets: | Plan | API and Platform Availability | Measurement | |---|---|---| | RadMah AI Sovereign | 99.9% per calendar month | Excluding Scheduled Maintenance and Excluded Events | | Enterprise | Negotiated individually (default 99.95%, higher SLAs available) | As specified in Enterprise contract | 3.3 Services in Scope: The following services are in scope for SLA measurement: — REST API (api.radmah.ai) — authentication, job submission, job status, data retrieval — WebSocket API for real-time job status — Customer dashboard (app.radmah.ai) — login, job management, evidence bundle access — Generation engine execution (for queued jobs that have commenced execution) 3.4 Services Out of Scope: — Third-party AI provider APIs (OpenAI, Anthropic, Google) used by AI Orchestrator and Agentic AI features — these are governed by the respective AI provider's own SLA — Data connector imports from customer-controlled systems (Snowflake, BigQuery, Postgres, etc.) — The RadMah AI marketing website (www.radmah.ai) — CLI and SDK tooling on the customer's local machines

4.1 If RadMah AI fails to meet the Monthly Uptime Percentage for a given calendar month, the Customer is entitled to a service credit against the next billing cycle as follows: | Monthly Uptime Achieved | Service Credit (% of Monthly Fee) | |---|---| | 99.0% – 99.9% | 10% | | 95.0% – 98.99% | 20% | | Below 95.0% | 30% | 4.2 Maximum Credit. Service credits will not exceed 30% of the Customer's monthly subscription fee (or monthly equivalent for annual plans) in any given calendar month. Service credits do not accumulate across months. 4.3 Credits Are Non-Cash. Service credits are applied as a credit to the Customer's next billing cycle and are not redeemable for cash under any circumstances. Credits expire if not used within 12 months of issuance. 4.4 Credits Are Sole Remedy. Service credits are the Customer's sole and exclusive remedy for RadMah AI's failure to meet the Monthly Uptime Percentage. They do not constitute an admission of fault or liability. No other damages (direct, indirect, consequential, or otherwise) are payable for SLA failures. 4.5 Credit Stacking. Service credits from multiple months do not stack against the Monthly Uptime Percentage calculation.

5.1 To claim a service credit, the Customer must: (a) Submit a written credit request to billing@radmah.ai within 30 days of the end of the calendar month in which the SLA failure occurred; (b) Include in the request: the Customer's Account ID; the specific dates and times of the alleged downtime; a description of the impact on the Customer; and any supporting evidence (such as monitoring data or error logs from the Customer's own systems). 5.2 RadMah AI will review all credit requests within 10 business days of receipt. RadMah AI's determination of whether a qualifying SLA failure occurred, and the quantum of any service credit, is final, provided RadMah AI acts in good faith and with reference to its own monitoring data. 5.3 Failure to submit a credit request within the 30-day window constitutes a waiver of the credit for that month.

The following events are excluded from Downtime calculations and do not constitute SLA failures: (a) Scheduled Maintenance — Planned maintenance announced at least 48 hours in advance. (b) Emergency Maintenance — Urgent maintenance to prevent an imminent security breach, data loss, or critical system failure. RadMah AI will endeavour to minimise the duration and announce emergency maintenance as soon as safely possible. (c) Force Majeure — Events outside RadMah AI's reasonable control, including: natural disasters; war; government orders or regulatory actions; large-scale internet infrastructure failures; DNS failures at upstream providers; DDoS attacks that overwhelm scrubbing infrastructure; or AWS service failures classified as region-wide outages by AWS. (d) Customer-Caused Events — Downtime resulting from actions or omissions by the Customer, including: invalid API requests; the Customer's use of the Service in a manner inconsistent with the Terms of Service or documentation; the Customer's network or connectivity issues; Customer-side resource exhaustion; or Customer-side misconfigurations. (e) Third-Party Service Failures — Failures of third-party services not operated by RadMah AI, including AI provider APIs (OpenAI, Anthropic, Google), payment processors (Stripe), or customer-operated data connectors. (f) Free Tier and Beta Features — As described in Section 2. (g) Service Suspended or Terminated — Any period during which the Customer's Account has been suspended or terminated in accordance with the Terms of Service.

7.1 RadMah AI measures availability using internal synthetic monitoring from multiple geographic locations, checking the primary API endpoints at 1-minute intervals. 7.2 The RadMah AI status page at status.radmah.ai provides real-time and historical availability information and is updated automatically by monitoring infrastructure. 7.3 RadMah AI retains monitoring data for at least 12 months and will make relevant monitoring data available to customers who submit a credit request for the relevant period. 7.4 If a Customer's own monitoring data materially conflicts with RadMah AI's monitoring data, RadMah AI will investigate in good faith and may commission a joint review. RadMah AI's monitoring data governs unless a material error is demonstrated.

8.1 RadMah AI targets low-traffic periods (typically Saturday or Sunday between 02:00 and 06:00 UTC) for planned maintenance windows. 8.2 Planned maintenance affecting RadMah AI Sovereign or Enterprise customers will be communicated at least 48 hours in advance by email to the registered account contact address and posted on status.radmah.ai. 8.3 Planned maintenance affecting only non-core services (such as the marketing website, documentation, or the admin dashboard) may occur without advance notice. 8.4 Enterprise customers may negotiate specific maintenance window exclusion periods (e.g., quarter-end, regulatory reporting periods) in their Enterprise contract.

The following support response time targets apply by plan and priority level. These are targets, not guarantees, and are separate from the availability SLA: | Priority | Definition | RadMah AI Sovereign Target | Enterprise Target | |---|---|---|---| | P1 — Critical | Service fully unavailable; all jobs failing | 2 hours (24/7) | 1 hour (24/7) | | P2 — High | Major feature unavailable; significant job failure rate | 4 business hours | 2 business hours | | P3 — Medium | Feature degraded; workaround available | 1 business day | 4 business hours | | P4 — Low | Questions, feature requests, documentation | 3 business days | 2 business days | Business hours: 09:00–18:00 Monday–Friday, US Eastern Time, excluding US federal holidays. Support channels: support@radmah.ai (all tiers); dedicated Slack channel (Enterprise); 24/7 pager (Enterprise P1/P2 with On-Call SLA addendum). Note: Support response time SLAs are not available on the Free tier.

10.1 RadMah AI performs automated backups of all customer data stored in RDS PostgreSQL instances and S3 object storage as follows: — Daily automated snapshots (retained for 7 days) — Weekly snapshots (retained for 4 weeks) — Point-in-time recovery (PITR) to within 5 minutes for the previous 35 days (RDS) 10.2 Recovery Time Objective (RTO): RadMah AI targets a Recovery Time Objective of 4 hours for restoration of service following a complete platform failure requiring a recovery from backup. 10.3 Recovery Point Objective (RPO): RadMah AI targets a Recovery Point Objective of 5 minutes (within the PITR window) for database recovery. 10.4 Backup Testing: Backup restoration is tested at least quarterly. Results are available to Enterprise customers on request. 10.5 Customer Responsibility: RadMah AI's backup and recovery commitments protect against platform-level failures. Customers are responsible for exporting and maintaining their own copies of Generated Data, Evidence Bundles, and Contract K specifications that they wish to retain beyond the platform's retention periods.

RadMah AI may modify this SLA with 30 days' written notice to RadMah AI Sovereign customers. Material reductions in SLA commitments will be communicated to affected customers by email to the registered account contact address. Enterprise customers' SLAs are governed by their individual Enterprise contracts. Changes to this standard SLA do not automatically modify negotiated Enterprise SLA terms. The current version of this SLA is always available at www.radmah.ai/legal/sla.

SLA credit requests: billing@radmah.ai Technical support: support@radmah.ai Enterprise SLA enquiries: legal@radmah.ai Status and incidents: status.radmah.ai