Legal - Service Terms

·Service Terms

·Last Updated on October 8, 2026

·The Service Terms govern your use of and access to the Services. Capitalized terms used in these Service Terms but not defined below are defined in the Terms of Service Agreement published at https://www.digitalocean.com/legal/terms-of-service-agreement or other written agreement in effect between you (referred to from time to time as a "Customer" for purposes of these Service Terms) and us (referred to from time to time as DigitalOcean) governing use of and access to the Services (the "Agreement").

11. Universal Service Terms (Applicable to All Services).

1.11.1 General. Service Terms include those set forth herein and any additional terms DigitalOcean may provide in writing, including terms presented at the time of purchase, in-product notices, or terms made available through DigitalOcean's Documentation. Access to or use of the Services constitutes acceptance of the applicable Service Terms by Customer. If Customer does not agree to the applicable Service Terms, it may not access or use the Services, and DigitalOcean grants no license or right to use the Services.

1.21.2 Service Level Commitments. Certain Service Terms include applicable service level agreements available at https://www.digitalocean.com/sla (each, an "SLA"). Each SLA applies solely to the Services identified therein, is subject to the terms, conditions, limitations, and restrictions specified in such SLA, and is incorporated into and forms part of the Agreement. Service Credits provided for under applicable SLAs are Customer's sole and exclusive remedy with respect to DigitalOcean's failure to meet the Service Commitment under the applicable SLA.

1.31.3 Miscellaneous.

1.3.11.3.1 You may not transfer outside of the Services any software (or related Documentation) you obtain from us or a third-party licensor in connection with the Services without specific, written permission to do so.

1.3.21.3.2 You must comply with current technical product documentation applicable to the Services available at https://docs.digitalocean.com/ (and any successor or related locations designated by us).

1.3.31.3.3 You will ensure all information you provide to us is accurate, complete, and not misleading.

1.3.41.3.4 From time to time, we may apply upgrades, patches, bug fixes, or other maintenance of the Services. We may provide you notice of such scheduled maintenance (except for emergency maintenance), and you agree to use reasonable efforts to comply with any maintenance requirements we notify you about.

1.3.51.3.5 If you process the personal data of any third parties, including your End Users or other identifiable individuals in your use of a Service, you are responsible for providing legally adequate privacy notices and obtaining necessary consents for processing, storage, use, and transfer of such data, including providing any required notices and obtaining any required parental consent under the Children's Online Privacy Protection Act ("COPPA") or any other similar or different laws. You represent to us that you have provided all necessary privacy notices and obtained all necessary consents. You are responsible for processing such data in accordance with applicable law.

1.3.61.3.6 Certain Services may incorporate generative AI features, some of which may be powered by the Gradient AI Platform Service, that enable you to use prompts or inputs to generate outputs. The output you generate is considered your Services Content. Due to the nature of machine learning, output may not be unique across customers and the Services may generate the same or similar results across customers. Outputs constitute draft text or images, created based on probabilistic methods, which may not be accurate and should not be relied upon without confirmation that they meet your specific needs. Services incorporating generative AI features are not intended for use in, or in association with, the operation of any hazardous environments or critical systems that may lead to serious bodily injury or death or cause environmental or property damage, automated decision-making, or to make legal, tax, or financial decisions. Services used in support of healthcare services are not medical devices and are not intended to be used by themselves for any clinical decision-making or other clinical use. We have no obligation or liability to you with respect to any claim arising from your use of generative AI features of select Services that: (i) arises from output you generate through use of the generative AI features of Services that either alone or as combined by you infringe or misappropriate any third-party's intellectual property rights; (ii) if you interfere with or fail to enable available filters, guardrails, or other tools, or disregard instructions on how to properly use the Services provided in the TOS and product documentation or in other materials we may make available to you; (iii) if your use of the Services breaches this TOS; (iv) if you have fine-tuned, refined, customized or modified the output or the generative AI feature of select Services and the alleged infringement or misappropriation would not have occurred but for this fine-tuning, refinement, customization, or modification; (v) arising after you receive notice to stop using the Service or feature of the Service; (vi) arising from output that you know or reasonably should know may infringe or misappropriate any third-party's intellectual property rights; and (vii) alleging that your use of the output infringes on a third-party's trademark or related rights. To the fullest extent possible under applicable law, you are responsible for your use of generative AI outputs generated by the Services.

22. DigitalOcean GradientTM AI Platform Service and Playground Feature

2.12.1 As part of providing the Gradient AI Platform Service (or successor name to the platform) and Playground, we may use manual and automated abuse detection mechanisms designed to detect harmful content, including related to potential violations of our or Third-Party AI Model providers' terms of service or acceptable use policies. Further, if these mechanisms detect apparent child sexual abuse material, you agree and instruct that we may report the incident to the National Center for Missing and Exploited Children or other authority. In addition, we may share information, that does not include your Services Content, about your use of a Third-Party AI Model on the Gradient AI Platform Service with the provider of that Third-Party AI Model.

2.22.2 The Gradient AI Platform Service and Playground feature may allow you to customize AI models with data you provide (for example, by fine-tuning). By using these customization features, you represent that you have all necessary rights, permissions, and licenses (including but not limited to for the data you use) to fine-tune or customize any model. You will have exclusive use of your customized model. We will not access or use your customized model except as necessary to maintain or provide the Gradient AI Platform Service and the Playground feature, or as necessary to comply with law or a binding order of a governmental body.

2.32.3 Attribution. For information about the Gradient AI Platform Service and Playground open source attribution notice, please click here: Gen AI Platform Service - Attribution Notice File.

2.42.4 The Gradient AI Platform Service's Playground feature is intended for experimentation purposes only, and any outputs or agents that may be created using the feature should not be used with production data or put into production. Customers may be subject to dynamic usage limits or throttling at our sole discretion to maintain service quality and prevent abuse.

2.52.5 Output generated by models accessed through the Gradient AI Platform Service or through the Playground feature may include information such as metadata, digital signatures, or watermarks to identify it is generated using a generative artificial intelligence model ("Provenance Data"). Neither you nor any End User may modify, tamper with, remove, obscure, or otherwise alter such Provenance Data.

2.62.6 Guardrails Provided by Us. We may provide Guardrails as part of the Gradient AI Platform Service. Guardrails are provided as capabilities within the platform to help you implement safeguards on outputs generated from foundation models. You are responsible for determining what guardrails need to be implemented to best meet your needs. DigitalOcean is not responsible if Guardrails do not perform as intended or expected.

33. Cloudways Services

·"Cloudways Services" mean any services offered by DigitalOcean under the brand name Cloudways on its Website including but not limited to cloud application deployment, management and infrastructure management - commonly referred to as the "Cloudways Platform." The following Service Terms apply to your use of Cloudways Services inclusive of the Cloudways Platform:

3.13.1 We charge in arrears for Cloudways Services, meaning that you will receive an invoice during the first week of each month for the Cloudways Services that you used in the previous month. You may cancel your Cloudways Services account at any time through your account settings or by contacting Cloudways support to avoid future charges (see https://support.cloudways.com/en/articles/5119918-how-do-i-cancel-my-account for more information on how to cancel).

3.23.2 Cloudways Services are a "pay as you go" service, meaning you will only pay for the resources that you have actually used and for the time that you have actually used them (minimum time fraction for billing related purposes is one hour increments).

3.33.3 Notwithstanding anything to the contrary in the TOS, we will refund your unused prepaid account funds for Cloudways Services if a valid refund request is submitted within three months of funds addition. However, the refund will not include the amount already deducted for Cloudways Services consumed prior to the date such refund is issued. For the avoidance of doubt, this refund provision only applies to Cloudways Services.

3.43.4 Some of the Cloudways Services may require that you establish your own account with one or more third-party cloud providers ("Cloud Provider"). You will be solely responsible for paying for and establishing an account ("Customer Cloud Provider Account") with each third-party Cloud Provider you select, and for compliance with the third-party Cloud Provider's terms. You must provide us with your Customer Cloud Provider Account credentials (the "CP Key") that are necessary for us to access and manage your use of the third-party Cloud Provider's servers, including, without limitation, accessing servers as necessary to identify or resolve technical problems or to respond to any issues related to or connected with the Cloudways Services. We will use reasonable efforts to maintain the confidentiality and security of your CP Keys.

3.53.5 Solely in connection with the Cloudways Services, we have also established our own accounts ("Cloudways Provider Accounts") with one or more third-party Cloud Providers, which we may use to provide you with the Cloudways Services. We will be responsible for paying for and establishing Cloudways Provider Accounts, provided that you will be responsible for paying us the portion of the Cloud Provider fees allocated to you (of which you will be notified in advance), and for compliance with applicable Cloud Provider terms.

3.63.6 PLEASE KEEP IN MIND THAT SOME OF THE CONTENT THAT WE MAKE AVAILABLE TO YOU THROUGH CLOUDWAYS SERVICES MAY ORIGINATE FROM THIRD-PARTY CLOUD PROVIDERS AND APPLICATIONS. ALL SUCH THIRD-PARTY CONTENT IS PROVIDED TO YOU "AS IS" AND YOUR USE THEREOF MAY BE SUBJECT TO CHANGE AND/OR REMOVAL AT ANY TIME. WE SHALL, UNDER NO CIRCUMSTANCES WHATSOEVER, ACCEPT LIABILITY RESULTING FROM YOUR USE OF ANY SUCH THIRD-PARTY SERVICES AND/OR CONTENT. WE ARE NOT LIABLE OR RESPONSIBLE FOR SERVICE UNAVAILABILITY CAUSED BY THE ACTIONS OF THE OTHER CLOUD PROVIDERS.

3.73.7 You may neither share nor re-sell your Cloudways Services account to any third-parties. Unless you have been specifically permitted to do so in a separate agreement with us, you agree that you will not reproduce, duplicate, copy, sell, trade or resell the Cloudways Services for any purpose.

3.83.8 We may offer (or act as intermediary in the offering of) certain application trials including the ability for you to install the application on a temporary basis for trial purposes. YOU ACKNOWLEDGE THAT SUCH TRIALS ARE INTENDED TO BE TEMPORARY INSTALLATIONS AND THAT WE MAY DELETE APPLICATIONS AND SHUT DOWN RELEVANT SERVERS AT ANY TIME.

3.93.9 We may automatically shut down applications and servers at various times, including without limitation, when you log out of the Cloudways Service, if you do not respond to a prompt after a certain period of time, or if you explicitly request us to do so, or if we detect that your application or server is malfunctioning. The data and software may be backed up during this process and may be restored, but such restoration is not guaranteed to reproduce the exact state of the application or server at the time the backup was created by us, particularly if you have used the "terminal" capability or otherwise directly accessed the server.

3.103.10 We do not promise to retain any preservations or backups on your behalf, and do not guarantee the completeness of any backup. You are solely responsible for the integrity, preservation, and backup of your Cloudways Services and content, regardless of whether your use of Cloudways Services includes a backup feature or functionality. Without limiting the foregoing, to the fullest extent permitted by law, we shall have no liability for any data loss, unavailability, or other consequences related to the foregoing.

3.113.11 Cloudways Velocity means the Cloudways Service that provisions and manages a Node.js runtime environment for hosting your applications, together with related deployment, process management, logging, monitoring, and security features. Sections 3.1-3.10 apply to Cloudways Velocity except as supplemented below.

3.11.13.11.1 Cloudways Velocity is offered on a per-plan basis, and each plan hosts the number of applications specified for that plan on dedicated resources. Fees for each plan and any separately metered resources (including storage charged per GB) are billed in accordance with Sections 3.1 and 3.2 at the rates published at the time of purchase.

3.11.23.11.2 We support only the Node.js LTS versions, runtimes, frameworks, and stack components we designate from time to time, and we may add, deprecate, or discontinue support for any of them, including end-of-life versions. We will use reasonable efforts to notify you before discontinuing support for a version you use, but you are responsible for keeping your application compatible with a supported version.

3.11.33.11.3 You are solely responsible for your source code, dependencies, build and start commands, configuration, environment variables, secrets, deployment scripts, and all content you run through the Service. We do not author, review, or control your application and make no representation that it will build, deploy, or run correctly, including that any deployment or reload will complete without interruption.

3.11.43.11.4 The Service relies on third-party and open-source software, including the Node.js runtime, package registries, package managers, and frameworks, all provided "as is" and subject to their own license terms, which you are responsible for reviewing and complying with. Where you connect a repository or other Third-Party Service, you remain responsible for your credentials and the third party's terms, and Section 3.4 applies to that access. We are not responsible for any third-party component's availability, security, or defects, including vulnerabilities in packages you install.

3.11.53.11.5 Velocity build and runtime processes can consume significant resources, and each plan is subject to its associated limits. If your usage exceeds those limits or threatens the stability, performance, or security of the platform or other customers, we may throttle, suspend, or require you to upgrade or reduce usage, consistent with Section 3.9. We may scan your application, dependencies, and content and may quarantine or remove content, or suspend the Service, where we reasonably detect malware or other threats. These protections are provided on a reasonable-efforts basis and do not guarantee that your application is secure.

3.11.63.11.6 Where we provide guided or managed migration assistance, we do so on a commercially reasonable-efforts basis with no guarantee of completeness, timing, or result. You represent that you have all rights necessary to transfer the code, data, and content you ask us to migrate, and you are responsible for verifying the result.

3.11.73.11.7 The Service is available only on DigitalOcean infrastructure and in the regions we designate, and Sections 3.4-3.6 govern the underlying cloud provider. On expiration or termination, you must migrate your application, data, and content off the Service before termination takes effect, and remaining content may be permanently deleted without further notice. Your use of the Service remains subject to the Acceptable Use Policy and to Section 3.7 (no sharing or resale).

44. AI Model Terms

4.14.1 Third-Party AI Models. The Services provide access to AI models under either (i) a "Bring Your Own Model" ("BYOM") option, or (ii) as elected (including any AI model deployed as part of a 1-Click deployment), configured, or integrated by you to operate with the Service platform ("Third Party AI Model").

4.24.2 Third Party AI Models are deemed Third Party Services and are subject to the following terms (unless you have other terms in place with the Third-Party AI Model provider for use of such Third Party AI Model). Customer acknowledges that the Third Party AI Model terms below are provided for convenience only and the list may be non-exhaustive. Customer should refer to the model provider's official documentation to ensure compliance with such applicable third party terms: https://www.digitalocean.com/legal/third-party-ai-models.

4.34.3 Service Models. In addition to Third Party AI Models, the Services may provide access to certain AI models embedded as part of the Services ("Service Model"). The following additional terms and/or disclosures apply to Service Models:

4.3.14.3.1 Black-forest-labs FLUX.1-schnell (provided by Fal.ai): This AI Model is provided by Fal.ai, as a subprocessor to the Service. (Last Updated: March 2026)

4.3.24.3.2 Elevenlabs TTS Multilingual v2: This AI Model is provided by Fal.ai, as a subprocessor to the Service. (Last Updated: March 2026)

4.3.34.3.3 Stability AI SD-XL 1.0 (Fast SDXL) and Stable Audio 2.5 (Test-to-Audio): This AI Model is provided by Fal.ai, as a subprocessor to the Service. (Last Updated: March 2026)

4.3.44.3.4. Plano-Orchestrator-30B-A3B: This Model is subject to the Katanemo Community License Agreement.

4.44.4 AI Model Changes. We may update, modify, replace, suspend, or remove any hosted AI Models made available through the Services, including but not limited to, changes driven by provider upgrades, model diversification initiatives, performance optimization, safety, security, or compliance issues, legal or regulatory requirements, or other operational concerns. These changes are necessary to ensure the continuous evolution, competitiveness, and integrity of the Services. Such updates may affect the functionality, availability, performance, or outputs of AI features provided through the Services, and you acknowledge that outputs or model behavior may vary over time as models are updated or improved, which is inherent to the evolving nature of AI technologies and will not be deemed a defect in the Services.

4.54.5 AI Safeguards.

4.5.14.5.1 In connection with your use of any AI model or other AI features of the Services, you must implement appropriate safeguards, controls, and review mechanisms when incorporating model outputs into applications or services provided to End Users. You are responsible for ensuring that outputs generated through the Services are appropriately evaluated and used responsibly and in compliance with applicable laws. Where your applications generate or distribute AI-generated content to End Users, you must implement reasonable measures, where technically feasible, to disclose or identify such content as AI-generated, including through watermarking, labeling, or other transparency mechanisms as required under applicable law.

4.5.24.5.2 We may, using automated and manual processes, monitor, scan, and analyze inputs and outputs to detect, prevent, or investigate violations of applicable law, this Agreement, or security risks, and to maintain the integrity and safety of the Services. We may retain relevant content and associated metadata to the extent reasonably necessary for such purposes, subject to its confidentiality and security obligations. Where we reasonably substantiate a violation based on corroborated evidence and internal review ("Confirmed Violation"), DigitalOcean may take appropriate action, including suspension, removal of content, or reporting to appropriate authorities or impacted parties, as required by law or where reasonably necessary to prevent harm.

55. Batch Inferencing

·The Batch Inferencing Service is designed to process large request collections asynchronously, and supports Third Party AI Models including OpenAI and Anthropic text models. Your output is subject to the Third Party AI Model provider's data retention settings (which may involve extended time periods and purposes in accordance with the applicable third party provider terms), which is not performed, controlled or managed by us. We do not retain backups of batch files in the Batch Inference Service, and you are responsible for maintaining and managing any backups you may need to meet your requirements. As with any AI output, Batch Inferencing Service outputs may be inaccurate and should not be relied upon without you confirming they meet your specific needs.

66. Launchpad

·Launchpad starter kits are opinionated application templates and automated deployment workflows that bundle multiple DigitalOcean Services into a pre-configured stack. Launchpad starter kits are convenience-only, non-production-ready resources provided "as is" without warranties. Customer is solely responsible for determining suitability, ensuring compliance (including compliance with all third-party licenses), configuring and implementing all necessary security controls, human oversight, and output validation. DigitalOcean has no obligation to maintain, support, or secure any resulting deployment.

77. Evaluations

·The following Service Terms apply to your use of the DigitalOcean Evaluation Service, which provides tools to test, score, and compare the performance of AI models using your own datasets, and an LLM-based "Judge Model." Capitalized terms not defined here are defined in the Agreement or the Service Terms.

7.17.1 Definitions

·Evaluation Dataset means the data, prompts, ground-truth, or other materials you submit to or select within the Service to evaluate a model.

·Evaluation Output means the scores, metrics, rankings, and results generated by the Service.

·Judge Model means the AI model you select to score or assess the outputs of another model.

·Candidate means models Customer selects for evaluation and models made available under the Bring-Your-Own-Model option and inference routers Customer has configured.

7.27.2 Your Data, Ownership, and Confidentiality

7.2.17.2.1 Ownership. As between you and DigitalOcean, you retain all ownership rights in your Evaluation Datasets and Evaluation Outputs. These are your Services Content and will be treated as your Confidential Information in accordance with the TOS.

7.2.27.2.2 No Training Commitment. DigitalOcean will not use your Evaluation Datasets or Evaluation Outputs to train, fine-tune, or improve any DigitalOcean hosted model, including the Judge Model.

7.2.37.2.3 Telemetry and Performance. To maintain and improve the performance and reliability of the Service, we may process anonymized, aggregated telemetry data (e.g., latency, throughput). This data will not include your Evaluation Datasets or Evaluation Outputs.

7.2.47.2.4 Isolation. DigitalOcean maintains technical and organizational measures designed to protect your data against cross-tenant access, as described in our Documentation.

7.37.3 Your Obligations

7.3.17.3.1 Rights and Permissions. You represent that you have all necessary rights and licenses to provide your Evaluation Datasets and to permit DigitalOcean to process them, without violating third-party rights or applicable law.

7.3.27.3.2 Data Sensitivity. You are responsible for ensuring your Evaluation Datasets are de-identified and do not contain sensitive data or personally identifiable information, unless your use is expressly governed by the Data Processing Agreement and the applicable Service tier supports such data.

7.3.37.3.3 Third-Party Models. Your use of third-party models is subject to https://www.digitalocean.com/legal/third-party-ai-models and any applicable third-party license terms, which you are responsible for following. Where applicable, DigitalOcean may share usage metadata (excluding your Services Content) with the relevant model provider for billing or abuse-detection purposes.

7.47.4 Benchmarking and Publication

7.4.17.4.1 Usage. You may use the Service to benchmark, test, and compare Candidates available through the Services.

7.4.27.4.2 Publication. You may publish your Evaluation Outputs, provided that you (a) do not present the results as a DigitalOcean endorsement or certification; (b) accurately describe your methodology, including the use of a probabilistic Judge Model; and © verify that your publication complies with any benchmarking, comparison, or publication restrictions imposed by the relevant third-party model providers. You assume all risk and liability related to your compliance with such restrictions.

7.57.5 Nature of Results; Human Oversight

7.5.17.5.1 Estimates Only. Evaluation Outputs are estimates generated by probabilistic methods and may be inaccurate, incomplete, or biased. They are not a guarantee of production performance, suitability, or legal compliance. You must independently verify all results before relying on them.

7.5.27.5.2 Human Oversight. Evaluation Outputs are intended to inform - not replace - human judgment. You are responsible for ensuring appropriate qualified personnel review all results before making deployment or other business decisions.

7.5.37.5.3 No Professional Advice. Safety and bias metrics are directional tools only and do not constitute legal, medical, financial, or other professional advice, nor do they guarantee that a model is "safe" or "compliant" for any regulatory purpose.

7.67.6 Pricing, Retention, and Deletion

7.6.17.6.1 Pricing. Fees for the Evaluation Service are set out on the DigitalOcean pricing page. While storage of Evaluation Datasets and Evaluation Outputs is currently provided at no additional charge, DigitalOcean reserves the right to introduce or modify storage fees in the future.

7.6.27.6.2 Retention. We retain your data to enable comparative analysis, subject to the retention periods and storage limits described in our Documentation. We do not provide backups; you are responsible for maintaining your own copies.

7.6.37.6.3 Deletion. Upon termination of your access, we will delete your Evaluation Datasets and Evaluation Outputs in accordance with our standard data-retention practices. You must export any data you wish to retain prior to termination.

88. Managed Databases

·The following Service Terms apply to your access to and use of DigitalOcean Managed Databases. These terms apply in addition to the Universal Service Terms and the Agreement. Capitalized terms used but not defined in this Section are defined elsewhere in these Service Terms or in the Agreement.

8.18.1 Service Level Agreement. The DigitalOcean Databases SLA is available at https://www.digitalocean.com/sla/databases (the "Databases SLA"). As provided in Section 1.2, Service Credits under the Databases SLA are Customer's sole and exclusive remedy for any failure to meet the applicable Service Commitment.

8.28.2 Beta Features. DigitalOcean may make certain features, engine versions, or cluster topologies available on a preview, beta, early access, or other pre-release basis (each, a "Beta Feature"). Beta Features are Beta Services under the Agreement, are provided "AS IS" and without any SLA or support commitment, and may be changed, suspended, or discontinued at any time. The Databases SLA does not apply to Beta Features.

8.38.3 Fees and Billing. Fees are based on the compute, memory, and storage resources you provision, at the rates published on the DigitalOcean pricing page. Storage is billed on a per-node basis. DigitalOcean does not separately charge for database data transfer (egress).

8.48.4 Backups; Customer Responsibility for Data. Regardless of any backup functionality provided as part of the Services, you remain responsible for maintaining backups and exports of your data sufficient to meet your business continuity, recovery, and compliance requirements. To the fullest extent permitted by law, and except for the Service Credits expressly provided under the Databases SLA, DigitalOcean will have no liability for any loss, corruption, or unavailability of data.

8.58.5 Maintenance and Failover. DigitalOcean may perform maintenance, version upgrades, and planned or unplanned failover operations, which may result in brief interruptions to connectivity. You are responsible for configuring your applications to reconnect following such events.

99. Prompt Caching

·The Prompt Caching feature reuses previously processed input token prefixes across requests to reduce time-to-first-token and the billed cost of eligible input tokens for supported models.

9.19.1 Cached Content. Cache entries are derived from your Services Content and are treated as your Services Content under the TOS. We process cache entries solely to provide the Services to you and do not use them to train, fine-tune, or improve any DigitalOcean model. Cache metadata, including cached and uncached token counts, hit-rate, and estimated-savings metrics, constitutes Usage Data.

9.29.2 Pricing. Cached-token pricing, discount rates, and model eligibility are set out on the DigitalOcean pricing page and Documentation and may change in accordance with Section 6 of the TOS. Any discount applies only to eligible cached input tokens; standard rates apply to all other input and output tokens. Realized savings depend on your prompts, traffic patterns, and cache hit rate, which are determined by your workload. We do not guarantee any cache hit rate, retention outcome, or level of savings, and any savings figures shown in the control panel or Documentation are estimates only.

9.39.3 Retention; Ephemerality; No Backup. Cache entries are ephemeral and subject to eviction and retention windows described in the Documentation, including any in-memory and persistent (tiered-storage) retention behavior, which we may change. Prompt Caching is not a storage or backup feature, and Section 4.6 of the TOS applies. We do not guarantee that any cache entry will persist or be available for any period.

9.49.4 Isolation and Customer Controls. We maintain per-account cache isolation by default. You are responsible for your use of cache-control mechanisms, including any prompt cache key and cache isolation key, to group or separate traffic within your account. You are responsible for configuring these controls appropriately for your data-separation requirements, and we are not responsible for cache reuse or sharing within your account resulting from your configuration. Beyond our isolation controls described in the Documentation, we do not warrant that response-timing or similar characteristics cannot be observed.

9.59.5 Opt-Out.You may disable Prompt Caching as described in the Documentation and request that applicable cache entries be flushed. Disabling Prompt Caching means eligible tokens will be billed at standard rates.

9.69.6 Routing and Processing. To provide Prompt Caching, we may route requests to, and store or process cache entries on, infrastructure and storage tiers we select, consistent with Section 4.7 of the TOS. Where Prompt Caching is used with a Third Party AI Model, caching and retention may also be subject to that provider's terms and data-retention settings, which we do not control (see Section 9.1).