> For the complete documentation index, see [llms.txt](https://doc.dastra.eu/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://doc.dastra.eu/en/features/ai-systems/establishing-a-record-of-ai-systems.md). # Establishing a record of AI systems To set up a record of your AI systems, you have two options: * Either you already have a registry that you can import directly into Dastra (go to the next section: [Import your AI systems](/en/features/ai-systems/import-your-ai-systems.md)) * Or you don't have one. In this case, you'll have to create one yourself ## Create a record of AI Systems To add an AI system, click **"Create a new AI system"**. A window offers three creation modes:
Create a new AI model dialog with three options: Generate with AI, Built-in template, Personalised

Three creation modes: AI generation, built-in template, or blank form

* **Generate with AI** — The AI assistant automatically generates a complete record from a short description (name, publisher, URL…). It uses web browsing to enrich the information. * **Built-in template** — Choose from the Dastra AI system library or from another workspace. * **Personalised** — Create a blank record and fill in each field manually. ### Generate an AI system with the AI assistant Select **"Generate with AI"**, then describe the system to document (name, publisher…). You can attach a file or provide a URL to help the generation.
AI system generation form with description field and URL option

Describe the system to generate — the assistant uses web browsing to complete the information

The assistant produces a pre-filled record (name, description, learning type, systemic risk, publisher…) which you can review and validate before saving.
AI-generated record for the Gemini 3.1 Pro system

The generated record should be reviewed before validation — AI can make mistakes

{% hint style="warning" %} Data generated automatically by AI may contain errors. Review and correct the proposed record before saving it to your registry. {% endhint %} Once you've entered the required information, you'll be redirected to a 10-step form. This form will enable you to give as much detail as possible about the AI system.
## The 11 Steps of the AI System Form Below are the 11 steps to complete when documenting an AI system. *** ### 1. General Enter **basic information** about the AI system: * **Name** of the system * **Brief description** of its purpose and functionality *** ### 2. Responsibilities Define your **role and responsibilities** under the European AI Act. The same organisation may hold several roles simultaneously: * **Provider** — develops an AI system and places it on the market or puts it into service under its own name or trademark. * **Deployer** — uses an AI system under its authority in the course of professional activities. * **Importer** — imports an AI system from a third country for placing on the EU market. * **Distributor** — makes an AI system available on the EU market without being a provider or importer. * **Authorised representative** — a natural or legal person in the EU mandated in writing by a provider established outside the EU. * **Product manufacturer** — if an AI system is integrated into a product subject to sector-specific legislation (e.g. medical devices), the product manufacturer is treated as a provider for that system. {% hint style="info" %} **Guided responsibility assessment** If no responsibility has been defined yet for this system, Dastra offers a **10-question questionnaire** to help determine which role(s) apply to your organisation. Once responsibilities are identified, the AI assistant can **automatically generate a justification**. {% endhint %} #### Contextual alerts based on declared role When declaring the **Deployer** role, Dastra displays contextual alerts based on the system's risk level: * **Unacceptable risk** — If you declare yourself as the deployer of an unacceptable-risk system, an alert reminds you that deploying such a system is **prohibited under the AI Act**. * **High risk** — If you declare yourself as the deployer of a high-risk system, an alert informs you of the situations where a [**Fundamental Rights Impact Assessment (FRIA)**](/en/features/ai-systems/fundamental-rights-impact-assessment-for-high-risk-ai-systems.md) is mandatory, and redirects you to the relevant questionnaire. *** ### 3. AI Models Specify the **AI model(s)** used to process data within this system. > ℹ️ For more details, refer to the AI Models Repository. *** ### 4. Stakeholders Identify **stakeholders involved** in implementing and managing this AI system, including their roles (e.g. Data Scientist, DPO, Product Owner). *** ### 5. Assets Add the **assets supporting this AI system**, such as: * Infrastructure components * Software tools * APIs * Documentation resources *** ### 6. Datasets List the **datasets associated** with this AI system. Indicate their usage among the following phases: * **Training**: the dataset used to **train the AI model**, enabling it to learn patterns, relationships, or classifications based on historical data. * **Validation**: a separate dataset used to **tune model parameters and prevent overfitting**. It helps assess model performance during training and guides adjustments for optimal results. * **Testing**: another distinct dataset used to **evaluate the final performance** of the trained and validated model before deployment. It provides an unbiased measure of how the model will perform on new, unseen data. * **Production inference**: data processed by the AI system **during actual operation**, where the trained model generates predictions, classifications, or decisions in real-world scenarios. *** Ensure that each dataset’s **purpose, composition, and linkage to this AI system** are clearly documented for transparency and compliance purposes. *** ### 7. Data Subjects Specify the **categories of data subjects** whose personal data is processed by the AI system (e.g. customers, employees, users). *** ### 8. Risk Analysis Assess the **level of risk** based on: * Types of data processed * Processing activities * Potential impacts on individuals’ rights and freedoms *** ### 9. Business Value Determine a **business value score** reflecting the system’s contribution to your organization to: * **Prioritize** high-value projects * Align AI initiatives with strategic objectives *** ### 10. Documentation Attach relevant **documents and information leaflets**, such as: * User notices * Technical guides * Compliance assessments (e.g. DPIAs) *** ### 11. Summary Review a **comprehensive summary** of all information entered for this AI system before final validation and registration. *** ## Linked processing activities and data synchronisation The **Data processings** tab of an AI system record lets you associate one or more processing activities from your GDPR record with that system. This link keeps your processing register and your AI systems registry consist ## Restore a previous version The **Activity history** panel on an AI system's record keeps every creation and modification. You can restore a past version directly from this panel.
A version's details in the Activity history with the Restore button

Each version in the Activity history offers a restore button

Each history entry shows a restore button, visible to users with write permission. After confirmation, Dastra creates a **new AI system** from the data of the chosen version; its name receives the suffix "- restored version - \[date]" and you are redirected to this new record. The original AI system remains unchanged.
Restored AI system suffixed 'restored version' in the list

The restored version appears as a new AI system suffixed "- restored version - [date]"

--- # Agent Instructions This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com. ## Querying This Documentation If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question. Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter: ``` GET https://doc.dastra.eu/en/features/ai-systems/establishing-a-record-of-ai-systems.md?ask=&goal= ``` `ask` is the immediate question: it should be specific, self-contained, and written in natural language. `goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal. The response will contain a direct answer to the question and relevant excerpts and sources from the documentation. Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.