The same DataRobot agreement covers hosted and installed use, but the obligations that attach to each delivery model differ. DataRobot’s enterprise architecture documentation lists the supported environments as multi-tenant SaaS (MTSaaS) on AWS within three regions, single-tenant SaaS (STSaaS) on AWS, GCP and Azure in many regions, self-managed VPC installations on AWS, GCP and Azure, and self-managed installations in a wide variety of data centers.[1] The Master Subscription Agreement (MSA) defines SaaS as software hosted in and delivered from DataRobot’s managed cloud environment, and on-premise as a version deployed in the customer’s data center or in the customer’s virtual private cloud.[2] A software asset manager needs the delivery model first, because it decides which policies apply, who applies upgrades, how usage is observed and what happens when a subscription ends.
Editions
Multi-tenant SaaS, single-tenant SaaS and self-managed
The documentation treats the three as deployment types of one platform. Some capabilities are explicitly tied to the type: the LLM-DR and LLM-BYOK usage reports are only available for single-tenant SaaS,[3] the Usage Report is only for system administrators on self-managed installations,[4] and seat licence assignment is available to multi-tenant SaaS organizations with a seat allocation and to single-tenant SaaS or self-managed installations whose licence key has a cluster-level seat allocation.[5] For DataRobot-managed deployments the DataRobot team allocates seats to organizations; for self-managed deployments the customer’s system administrator does.[5] See Self-Managed AI Platform, Single-tenant SaaS and Multi-tenant SaaS.
Evaluation and trial access
Evaluation Software may be delivered on-premise or as software-as-a-service and is a no-fee, trial, demo, alpha, beta or early access offering with a 30-day Evaluation Term, a licence solely to evaluate suitability for purchase, and a right for DataRobot to terminate, downgrade, limit or modify the evaluation at any time without notice.[6] Models generated in the evaluation may only be used to evaluate features and must be destroyed unless the customer buys DataRobot’s software within three months of access ending.[6] See Evaluation Software.
Metrics
Metrics differ by deployment. On self-managed installations the licence page shows the concurrent workers limit, maximum active users, prepaid deployment limit and maximum deployment limit, and the system administrator’s Usage Report charts CPU and GPU usage against the organization’s licence limit for core usage.[7][4] On SaaS, the Usage explorer in Account settings reports GPU, CPU, LLM API, agentic span and storage usage by service.[3] For MLOps in either model, the Max deployments setting cannot exceed Prepaid deployments and inactive deployments do not count.[8][9] For organizations on the serverless platform the maximum compute instances default to 8.[8] See DataRobot seats, workers and deployment limits for these constructs.
Counting and floors
The commercial counting rules come from the Order. Customers who buy through a partner count entitlements as stated in the partner order communicated to DataRobot.[2] The MSA cover note states that the standard terms are meant to work for both on-premise and SaaS deployments if the customer switches at any time, which means the contract does not need to be replaced when the delivery model changes, although the Order must still describe what was bought.[10] Maintenance and Support fees are included in the fees for the Solution.[2]
Virtualization and partitioning
For on-premise installations the unit that links the licence to infrastructure is the licence limit for core usage in the Usage Report, together with the limits on workers and active users in the licence key.[4][7] The MSA treats a deployment in the customer’s virtual private cloud as on-premise.[2] Software copying is limited: the customer may not copy the Solution other than for backup purposes where the agreement is not for SaaS.[2] Organizations can cap shared modeling workers, but this allocation is independent of the cluster configuration and the documentation advises not to oversubscribe the physical capacity of the cluster.[8]
Cloud and BYOL
Data and usage monitoring
The data restrictions apply where the Solution is used as SaaS: the customer must not import PCI DSS data, financial credentials, government ID numbers, sensitive personal data, personal data of individuals under 16 and similar data, and HIPAA-regulated information is only permitted if the Order expressly allows it.[2] DataRobot may delete offending Customer Data or suspend access where the customer breaches those rules, where removal is needed to protect security or integrity, or to respond to a governmental authority, giving notice as soon as possible.[2] For SaaS, DataRobot may automatically collect and analyze usage data through the User Activity Monitor, including logs, login frequency, number of Models deployed and feature usage, plus clickstream and navigation data, and may collect Metadata about datasets, Models and Projects that is always anonymized; for on-premise use it can only analyse such data if the customer or a user provides it.[2]
Third-party and generative AI products
Where the Order includes a generative AI product, the customer may be granted limited access to Third Party Products such as large language models and vector databases for evaluation; they are not part of the Solution, may change during the term and carry no DataRobot warranty.[2] DataRobot does not provide Maintenance or Support for any Third Party Product or for customisations, scripts or extensions created for the customer unless agreed in writing.[11] Users are also bound by the provider’s own terms for those products.[12]
Hosted self-service
The AI Platform Agreement covers hosted use on an Order Form or registration page. Under it the customer may use the service subject to unexpired Credits, charges and Credit purchases are payable in advance and non-refundable, the service is not intended for personal information about identifiable natural persons beyond limited user data, and the customer grants DataRobot a non-exclusive licence to use Customer Data to improve and enhance the services.[13] DataRobot’s liability under that agreement is limited to the greater of the amounts paid in the preceding twelve months or $200.[13] This differs from the MSA, which provides that DataRobot only processes Customer Data as necessary to perform its obligations and caps liability at 100% of annual fees.[2]
Programs
Support
Customers open cases through the Support Portal and report errors there. Severity 1 means all or a critical portion of the Solution is not operating, with DataRobot resources available 24x7 and an initial response target of within 1 hour; Severity 2 targets a response within 2 Business Hours, Severity 3 within 8 Business Hours and Severity 4 within 2 Business Days.[11] Support for upgrades is designated Severity 3.[11] DataRobot may access error and application logs held by the customer with written permission for proactive support. The policy does not apply to software not purchased from DataRobot or where the customer is in material breach or has an overdue undisputed invoice.[11]
Maintenance
DataRobot makes upgrades available without additional charge. For the on-premise (self-managed) version the customer downloads and installs upgrades once released, DataRobot provides Maintenance Releases for each LTS Release until the next LTS Release (it usually provides one LTS Release per year), and provides Critical Patches for non-LTS releases until the next non-LTS release.[11] For SaaS, upgrades are applied automatically, DataRobot gives no less than 14 days’ notice of scheduled maintenance at its status page, and access may be impeded during maintenance.[11] DataRobot may give self-managed customers an option to pin the runtime for existing and legacy Models, in which case it is not responsible for fixes and security updates of a pinned runtime that is not the most recent.[11]
Availability
The Availability Policy applies only to the SaaS version, both single-tenant and multi-tenant, as indicated in the Order. It sets a Quarterly Uptime of at least 99.95%, and where uptime falls to 97.0% to 99.94% the service credit is 5 percent of the fees for the affected Solution for the quarter; the other bands are 10 percent for 95.0% to 96.9% and 20 percent below 95%.[14] To receive credits the customer must submit a written request within 30 days after the end of the quarter, and the credits are the sole and exclusive remedy for service interruption.[14] The MSA cover note mentions a 99.9% availability guarantee with tiered credits, which differs from the 99.95% in the policy.[10]
Audits and compliance
For self-managed installations, the compliance point is licence validity: users can still make predictions with existing models if the licence expires, but model building, new insights and Compliance Documentation stop until a new key is applied.[7] The licence expiry banner appears when the licence is within 30 days of expiring by default.[7] For SaaS, suspension for unpaid invoices follows 14 days’ notice once undisputed payments are 60 days overdue, and the MSA allows deletion of data or suspension for breach of the data restrictions.[2] The MSA has no audit clause; evidence comes from the product’s license page, seat pages and usage reports.[7][5][4]
Out of scope
This article does not cover DataRobot’s Information Security Policy, Data Processing Policy or Japan-specific versions of the legal documents, which are listed on its legal page, or the infrastructure sizing requirements in the self-managed installation guide.[15]