# Legal document changelog

Source: https://docs.serenityedge.ai/terms/changelog/

> **Document control**
> 
> | Field | Value |
> | --- | --- |
> | Version | 1.3 |
> | Effective date | 18 November 2026 |
> | Last revised | 18 November 2026 |
> | Approved by | Substrate Artificial Inteligence, S.A. |
> | Contracting entity | Substrate Artificial Inteligence, S.A. |

This page records the version history of the [Legal notice](https://docs.serenityedge.ai/terms/legal-notice/), the [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/), the [Privacy Policy](https://docs.serenityedge.ai/terms/privacy-policy/), the [Data Processing Agreement](https://docs.serenityedge.ai/terms/data-processing-agreement/), the [Sub-processors](https://docs.serenityedge.ai/terms/sub-processors/) list, the [Acceptable Use Policy](https://docs.serenityedge.ai/terms/acceptable-use/) and the [Cookie Policy](https://docs.serenityedge.ai/terms/cookie-policy/), and states from which date each version applies.

## Version 1.3

**Published and effective 18 November 2026.** Version 1.3 of every legal document takes effect on publication and replaces version 1.2. It adds the terms for media models, the models that generate or edit images, generate video or generate speech, and states fees in units other than tokens. A media model is available to you only once the Documentation lists it.

Version 1.3 takes effect without the thirty days of notice that section 16 of the Terms of Service and clause 13.1 of the Data Processing Agreement provide for substantial changes, because no customer uses a media model yet: the new terms apply only to media models, and nothing changes for the models in use today.

### What changes for you

-   **Media models have their own terms.** The Terms of Service gain Annex 2 and the Data Processing Agreement gains Exhibit 5. Nothing changes for the other models: their prompts and completions are still processed in memory only, never written to disk or logs, and never kept beyond the prompt cache.
-   **Generated images and videos are kept for up to 60 minutes so that you can download them.** The answer gives a temporary address instead of the file. The file is held in memory only, never written to disk or logs, can be downloaded any number of times for 60 minutes and is then deleted, in any case within 61 minutes. Anyone who holds the address of an image can download it until then, so keep it confidential. Ask for an image in the response itself and nothing is kept. Speech is returned in the response and nothing is kept.
-   **Automated safety checks on images and video.** The prompts, input images and outputs of the models that generate images or video are checked automatically before and after generation. The checks keep no copy, and no person reviews them. A request or an output that fails a check is refused, and repeated refusals pause an end user's or an API key's media requests for a while.
-   **Every generated image, video and audio is marked as made by AI.** It carries an invisible watermark, and an image may also carry a signed content credential. The marks do not identify you or your users. Do not remove or alter them on purpose.
-   **Voices.** The voices of the speech models are synthetic preset voices that come with the model. Do not present them as the voice of a real person. Voice cloning is not offered.
-   **Real people, characters and trade marks.** Do not generate or edit images, video or audio of a real, identifiable person, or imitate their voice, without that person's consent, and do not use protected characters or marks without the right to do so. You are responsible for your prompts, your inputs and what you do with the outputs; there is no filter for characters or marks.
-   **Model use policies.** Where the author of a model publishes a policy on how its model may be used, and the Documentation identifies it for that model, you must follow it, as you already follow the model's licence.
-   **Models served with a published adapter or as a quantisation.** An open model may be served with an adapter published for it by a third party, for example to generate in fewer steps, or as a quantisation of its published weights (stored at a lower precision), published by a third party or made by Substrate AI. It keeps its own model identifier and its licence, and the Documentation says how it is served.
-   **Fees in other units.** Besides tokens, a model may be priced per second of audio, per character, per image or per second of video, as the Documentation publishes for it.

### Changes by document

-   [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/): the Preamble and §5.1 describe the models beyond large language models and the compatibility with the OpenAI API in general; new definitions of "Generated Media", "Media Model", "Media URL" and "Model Use Policy"; the definitions of "Fees", "Input" and "Request Metadata" cover other billing units, images and audio; §5.1 adds "Published adapters and quantisations"; §6 refers to Annex 2 for the retention of Generated Media, the content checks and the marks, and makes Output ownership subject to Model Use Policies; §7.1 states fees in other billing units; §8 and §10.3 add Model Use Policies; §9, §13.3 and §14.2 refer to Annex 2; new Annex 2 "Media Models" (definitions, availability, delivery and retention, automated safety measures, marking, voices, real people, characters and trade marks, fees).
-   [Acceptable Use Policy](https://docs.serenityedge.ai/terms/acceptable-use/): the introduction states the automated safety measures of Media Models; §1 prohibits depicting real people without their consent, including sexual or intimate content of a real person, and impersonation with generated media or voices, and names protected characters and trade marks; §3 prohibits circumventing the safety measures and removing or altering the marks; §4 adds Model Use Policies; §5 adds the marks and the disclosure of deep fakes; §7 adds the pause of media requests and the suspension of a key on which child sexual abuse material is detected.
-   [Data Processing Agreement](https://docs.serenityedge.ai/terms/data-processing-agreement/): the Preamble and §2.1 cover Media Models; new definitions of "Generated Media", "Media Model" and "Media URL"; the definitions of "Inputs" and "Request Metadata" cover images, audio and other billing units; §2.2, §4.2, §5.1, §5.2, §9.1, §9.2, Exhibit 1 (C, D, F and I) and Exhibit 2 (1) refer to the holding of Generated Media; new Exhibit 5 "Media Models" (scope and precedence, holding and deletion, nature and purpose, automated safety measures, marking, Sub-processors).
-   [Sub-processors](https://docs.serenityedge.ai/terms/sub-processors/): no new Sub-processor. The introduction and the row of Subgen AI Spain, S.L. state that Generated Media is held in volatile memory for up to 60 minutes, and at most 61, and the inputs of a video job for at most 30 minutes.
-   [Privacy Policy](https://docs.serenityedge.ai/terms/privacy-policy/): the Overview, §1, §3.2, §3.5, §4, §5 and §9 cover Media Models: the holding of Generated Media, the automated safety measures and the marks, and the billing units in Request Metadata.
-   [Legal notice](https://docs.serenityedge.ai/terms/legal-notice/) §2: the service gives access to large language models and other artificial intelligence models, and prices may be expressed in units other than tokens.
-   [Cookie Policy](https://docs.serenityedge.ai/terms/cookie-policy/): no change other than the version.

## Version 1.2

**Published and effective 14 November 2026.** Version 1.2 of every legal document takes effect on publication and replaces version 1.1. It covers Sandboxes, describes the processing locations in general terms and corrects some facts.

### What changes for you

-   **Sandboxes are covered.** The Terms of Service gain Annex 1 and the Data Processing Agreement gains Exhibit 4 for Sandboxes, in private preview. A sandbox stores your files while it exists, and a paused sandbox for at most 30 days. Do not put production secrets or personal data in a sandbox during the preview. We keep audit records of sandbox requests for 12 months, and records of each sandbox's outbound connections, never their content, for 90 days.
-   **Where your data is processed, in general terms.** Inference runs on Substrate AI's own GPUs in the European Union, in facilities operated by Substrate AI or, on its behalf, by Subgen AI, and request handling runs in Microsoft Azure data centres in the European Union. The documents no longer name individual sites or cloud regions; the commitment to process your data in the European Union does not change.
-   **Serenity Star accounts.** With a Serenity Star account, your requests and their answers pass through Serenity Star's API, which authenticates and meters them. Version 1.1 said that Serenity Star did not receive them.
-   **Request metadata.** Its definition now describes all the technical information we log about a request (for example the client's user agent, the request size and the cost), instead of a closed list. It still never includes your prompts or completions, and it is still kept for 90 days.
-   **Our websites.** The Legal notice, the Privacy Policy and the Cookie Policy now cover `serenityedge.ai` as well as `docs.serenityedge.ai`, including the playground: its security check, its limits per IP address, kept for at most two hours, and what it stores in your browser.
-   **Capacity limits.** A request over a rate or capacity limit is answered with HTTP 429, never 503.

### Changes by document

-   [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/): the API is described by its address, `https://api.serenityedge.ai`, and the endpoints in the Documentation; Sandboxes join the Services (Preamble, definitions of "API", "Fees", "Services", "Sandbox", "Sandbox Data" and "Sandbox Records"); the definition of "Request Metadata" describes categories instead of a closed list; §5.1 states that rate and capacity limits answer HTTP 429; §6, §7.1, §13.3 and §14.2 refer to Annex 1; new Annex 1 "Sandboxes" (scope, private preview, access, location, storage and deletion of Sandbox Data, responsibilities, Sandbox Records, fees and suspension).
-   [Data Processing Agreement](https://docs.serenityedge.ai/terms/data-processing-agreement/): the Preamble, §2.1 and §2.2 cover Sandbox Data and Sandbox Records; the definitions of "Request Metadata" and "Services" change as in the Terms of Service; §2.3, §6.4 and Exhibit 1 (J) describe the Serenity Star route as it works, with Inputs and Outputs passing through Serenity Star's API; §7.1, §8.2 and Exhibit 2 (5) state the locations in general terms; §4.2, §5.1, §5.2, §9.1, §9.2, Exhibit 1 (B) and Exhibit 2 (8) refer to Sandboxes or to the new definition; new Exhibit 4 "Sandboxes".
-   [Sub-processors](https://docs.serenityedge.ai/terms/sub-processors/): no new Sub-processor. Cloudflare covers `serenityedge.ai` and the DNS of the sandbox domains; Microsoft Azure covers the routing to Sandboxes and the relay of their encrypted traffic, in data centres in the European Union; Subgen AI's facility also hosts Sandboxes; the Serenity Star row states that Inputs and Outputs pass through Serenity Star's API on that route.
-   [Privacy Policy](https://docs.serenityedge.ai/terms/privacy-policy/): scope extended to `serenityedge.ai` and its playground; definitions of "Request Metadata", "Services" and "Website", and of the Sandbox terms; §3.2 adds Sandbox Records, §3.3 the playground, §3.4 the Serenity Star route, §3.5 Sandbox Data; §4 adds the playground as a purpose; §5 adds the retention of Sandbox Records, Sandbox Data and the playground's limits; §6 and §7 state the locations in general terms.
-   [Acceptable Use Policy](https://docs.serenityedge.ai/terms/acceptable-use/) §3 and §7: uses of Sandboxes that are prohibited, and the measures Substrate AI may take on a Sandbox.
-   [Legal notice](https://docs.serenityedge.ai/terms/legal-notice/): covers `serenityedge.ai` and its playground, and names Sandboxes and their prices page.
-   [Cookie Policy](https://docs.serenityedge.ai/terms/cookie-policy/): covers `serenityedge.ai`: Cloudflare Turnstile on the playground, and the local and session storage it uses; Cloudflare Web Analytics is described as used where enabled.

## Version 1.1

**Published and effective 17 October 2026.** Version 1.1 of the [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/), the [Acceptable Use Policy](https://docs.serenityedge.ai/terms/acceptable-use/), the [Data Processing Agreement](https://docs.serenityedge.ai/terms/data-processing-agreement/), the [Sub-processors](https://docs.serenityedge.ai/terms/sub-processors/) list, the [Privacy Policy](https://docs.serenityedge.ai/terms/privacy-policy/), the [Legal notice](https://docs.serenityedge.ai/terms/legal-notice/), the [Cookie Policy](https://docs.serenityedge.ai/terms/cookie-policy/) and this changelog clarifies how the routes through routers and marketplaces are treated, and takes effect on publication.

The access route through OpenRouter becomes the route through a Platform: a third-party router, gateway or marketplace, such as OpenRouter, cortecs or EUrouter, through which a Customer may access the Services. The Platforms on which the Services are available, and those on which their onboarding is in progress, are listed on the [Routers and marketplaces](https://docs.serenityedge.ai/docs/access/channels/) page; the legal documents no longer state which Platform is available. The Serenity Star Account and Enterprise Agreement routes, the Platform Sub-processors and the locations of the Processing do not change.

-   [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/): the definitions of "Platform" and "Platform Partner Addendum" replace the definition of "OpenRouter". The Parties paragraph and §3 state that a person who uses the Services only through a Platform that resells them does not become a Customer, or accept the Terms of Service, by doing so: that person's relationship is with the Platform. §4.2.1 "Through a Platform" (formerly "Through OpenRouter") states that the agreement Substrate AI has signed with a Platform governs the relationship with that Platform, and that the Platform's own terms govern its relationship with its customers. On a marketplace, the Customer contracts with the Platform under the Platform's terms, the Platform meters and bills the Customer, and the Terms of Service, the Privacy Policy and the Data Processing Agreement apply from the moment the request reaches the API; the Platform is not a sub-processor engaged by Substrate AI, and a Commercial Customer is responsible for the Platform leg and any transfer it entails. A Platform may resell the Services in its own name only under a reseller agreement and a Platform Partner Addendum signed with Substrate AI; it is then Substrate AI's Customer and the Customer under the Data Processing Agreement, and it must bind its customers to restrictions at least as protective as the Acceptable Use Policy and the Open Model Licences, and its API Keys may be suspended in case of misuse. The Preamble, §2.2, §4.1, §5.2, §5.3, §7, §10.2, §12.1 and §13.2 refer to Platforms instead of OpenRouter.
-   [Data Processing Agreement](https://docs.serenityedge.ai/terms/data-processing-agreement/): new definitions of "Platform" and "Platform Partner Addendum"; §2.3, §3.2, §6.1, §6.5 "Platforms" (formerly "OpenRouter") and Exhibit 1 (C and J) refer to Platforms. §2.3 and §6.5 describe the Platform Partner Addendum, which a Platform that resells the Services must sign with Substrate AI and any other Platform that processes its customers' Personal Data as their Data Processor may sign. Under it, the Platform is the Customer under the Data Processing Agreement and Substrate AI acts as its sub-processor under Article 28(4) GDPR; the Platform accesses the Services only through the API, with partner keys issued by Substrate AI; Inputs and Outputs are processed in the European Economic Area, with inference only on GPU infrastructure owned and operated by Substrate AI or its group; and Personal Data Breaches are notified within 48 hours. For that Platform, the addendum prevails over the Data Processing Agreement.
-   [Sub-processors](https://docs.serenityedge.ai/terms/sub-processors/): the introduction and §3 "Platforms" (formerly "OpenRouter") explain why a Platform is not a Sub-processor and refer to the Platform Partner Addendum.
-   [Privacy Policy](https://docs.serenityedge.ai/terms/privacy-policy/): the Scope, a new definition of "Platform", §4, §6, §7 and §9 refer to Platforms instead of OpenRouter.
-   [Acceptable Use Policy](https://docs.serenityedge.ai/terms/acceptable-use/) §7: requests sent through a Platform may be declined.
-   [Legal notice](https://docs.serenityedge.ai/terms/legal-notice/) §2: Serenity Edge can be contracted through third-party routers, gateways and marketplaces, listed on the Routers and marketplaces page, instead of naming OpenRouter as a route in onboarding. §4 and §5 refer to third-party routers and marketplaces among the marks of third parties and the linked websites, instead of naming OpenRouter.
-   [Cookie Policy](https://docs.serenityedge.ai/terms/cookie-policy/): the websites of third-party routers and marketplaces have their own cookie policies.

## Revision of 14 October 2026

Non-substantial changes under section 16 of the [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/); the version stays 1.0 and no document takes effect on a new date.

-   [Legal notice](https://docs.serenityedge.ai/terms/legal-notice/) and [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/) §4.2.1: OpenRouter is described as a route in onboarding, not available until OpenRouter enables it. Both said the Services were already available on the OpenRouter marketplace, which was not yet the case. The rest of §4.2.1 applies from the day the route is available.

## Revision of 23 September 2026

Non-substantial changes under section 16 of the [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/); the version stays 1.0 and no document takes effect on a new date.

-   [Privacy Policy](https://docs.serenityedge.ai/terms/privacy-policy/) §2.2: the sentence left open on whether a Data Protection Officer must be designated is removed. The section states the fact: Substrate AI has not designated one and requests go to the contact details in §2.1.
-   [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/) and [Legal notice](https://docs.serenityedge.ai/terms/legal-notice/): the Commercial Registry entry is given as Commercial Registry of Toledo, Section 8, Sheet TO-48960, IRUS 1000310248756. The volume and folio placeholders are removed.
-   [Sub-processors](https://docs.serenityedge.ai/terms/sub-processors/): the status page provider is Better Stack, Inc., with its region of processing and transfer basis.

## Version 1.0

**Published and effective 22 September 2026.** First version of the eight legal documents of Serenity Edge:

1.  [Legal notice](https://docs.serenityedge.ai/terms/legal-notice/)
2.  [Terms of Service](https://docs.serenityedge.ai/terms/terms-of-service/)
3.  [Privacy Policy](https://docs.serenityedge.ai/terms/privacy-policy/)
4.  [Data Processing Agreement](https://docs.serenityedge.ai/terms/data-processing-agreement/)
5.  [Sub-processors](https://docs.serenityedge.ai/terms/sub-processors/)
6.  [Acceptable Use Policy](https://docs.serenityedge.ai/terms/acceptable-use/)
7.  [Cookie Policy](https://docs.serenityedge.ai/terms/cookie-policy/)
8.  This changelog

Each document carries a document control block stating "Version 1.0", "Effective date: 22 September 2026", "Last revised: 22 September 2026" and "Approved by: Substrate Artificial Inteligence, S.A.". Version 1.0 is the first published version of each document and supersedes nothing.

### Origin of the documents

The Terms of Service, the Privacy Policy, the Data Processing Agreement and the Sub-processors list are adapted from version 2.0 of the corresponding Serenity Star documents, published by Subgen AI Spain, S.L. on 31 August 2026 with effect from 1 October 2026. Their structure, definitions, tone and level of detail have been kept; every provision that only made sense for the Serenity Star platform (agents, workspaces, subscription plans, the credits interface, Google user data, moderation filters and the chat widget) has been removed, and every provision has been rewritten for an inference API sold by a different entity. The Legal notice, the Acceptable Use Policy and the Cookie Policy are new: the Serenity Star site does not publish equivalents, and the acceptable use rules that sit inside section 8 of the Serenity Star Terms of Use have been moved into a separate policy that the Terms of Service incorporate by reference.

### Contracting entity

The contracting entity, data controller and data processor under these documents is **Substrate Artificial Inteligence, S.A.**, a company incorporated under the laws of Spain and listed on BME Growth, with Spanish tax identification number (NIF) A98306228, registered with the Commercial Registry of Toledo under Section 8, Sheet TO-48960, with its registered office at Avenida Real Fábrica de Sedas 28, 45600 Talavera de la Reina (Toledo), Spain. The registry entry is identified by its sheet and unique registry identifier (IRUS). The single contact address for notices, privacy requests and legal matters is [contact@substrate.ai](mailto:contact@substrate.ai). No Data Protection Officer has been designated.

Subgen AI Spain, S.L., the contracting entity of the Serenity Star documents, appears in the Serenity Edge documents as a group company of Substrate AI: it operates the facility in Valencia where inference runs and operates the Serenity Star accounts through which customers may obtain API keys and credits. The intra-group data processing agreement between the two companies is in preparation.

### Governing law and jurisdiction

Spanish law governs every document. The courts of Madrid, Spain, have jurisdiction, replacing the courts of Valencia named in the Serenity Star documents. The consumer carve-out of the Serenity Star documents has been kept: a consumer does not lose the protection of the mandatory provisions of the law of their country of residence, may bring proceedings either before the courts of their place of residence or before the courts of Madrid, and may only be sued before the courts of their place of residence.

### Access routes

The Terms of Service describe the three routes through which Serenity Edge can be used and what each one entails: through OpenRouter, where the customer's contract is with OpenRouter and Substrate AI is a provider on that marketplace; with a Serenity Star account, where account creation, API keys, credits and billing are operated by Subgen AI Spain, S.L. under the Serenity Star terms as a group company acting for Substrate AI; and under an enterprise contract negotiated individually, which is the only route with a service level agreement. The Data Processing Agreement and the Sub-processors list treat OpenRouter as an independent marketplace rather than a sub-processor, and Serenity Star as an access route provider engaged only when the customer chooses that route.

### Data policy

The documents record the data policy of the service as a binding commitment: prompts and completions (Inputs and Outputs) are processed in memory and are never written to disk or logs anywhere in the pipeline; to speed up repeated requests, prompt prefixes may be kept in volatile memory (a cache) for at most 24 hours, after which they are automatically evicted, and cached content is never used for training and is not accessible to staff; no customer content is used to train any model; only request metadata (timestamps, token counts, model, latency, status codes and API key identifier) is kept, for 90 days, for billing, security and operations. The cache clause appears in Section 6 of the Terms of Service, in sections 3.5, 5 and 9 of the Privacy Policy, and in clauses 4.2, 5, 9 and Exhibits 1 and 2 of the Data Processing Agreement. Inference runs on Substrate AI's own GPU infrastructure in Valencia, Spain; request handling runs in Microsoft Azure Spain Central; authentication and billing for Serenity Star account traffic run in Microsoft Azure France Central. No personal data is transferred outside the European Union. The Data Processing Agreement adapts the Article 28 GDPR clauses to that design: data subject requests, breach notification, return and deletion at the end of the services and audits are all expressed for a processor that persists nothing.

### Fees and service levels

Fees are published per million tokens, in United States dollars, and are metered on the `usage` object of each response. Price increases are announced with at least thirty days of notice. Prepaid credits are non-refundable except where the law requires otherwise, including the consumer right of withdrawal. Outside an enterprise contract the service is provided on commercially reasonable efforts, without a service level agreement, and rate limits are enforced with HTTP 429 responses. Section 5.2 of the Terms of Service states that the current and historical availability of the Services is published at [https://status.serenityedge.ai](https://status.serenityedge.ai/). Models may be added, updated or retired; retirement carries at least thirty days of notice except for security or legal reasons.

### Sub-processors

The Sub-processors list names Cloudflare, Inc. (edge network), Microsoft Ireland Operations Limited (Microsoft Azure Spain Central for request handling and, for the Serenity Star route, France Central) and Subgen AI Spain, S.L. (operation of the Valencia facility and of Serenity Star accounts). The status page provider was added on 23 September 2026 (see below).

### Document control

Every document carries a version number, an effective date, a last-revised date, the approver and the contracting entity. This changelog is the version history they refer to.
