Draft – legal review pending

Passages marked [TO BE VERIFIED: …] need a business decision or legal review. The draft describes the Platform's data flows as of October 2026 (development environment). It must be reconciled with the actual production environment before launch. [TO BE VERIFIED: only a development environment exists so far; the production domain, region and provider configuration have not been defined.]

1. Controller and contact

The controller for the processing of personal data on the appvanced AI Studio platform ("Platform", "Studio") is:

appvanced Jan Kronenberg Am Fuchsberg 2b
41468 Neuss
Germany
Phone: +49 170 955 3 669
E-mail: kontakt@appvanced.de

Data protection contact: kontakt@appvanced.de [TO BE VERIFIED: set up a dedicated privacy address (e.g. datenschutz@appvanced.de)?]
[TO BE VERIFIED: no data protection officer is named on appvanced.de. Check whether one must be designated (Art. 37 GDPR, Section 38 BDSG – among other things for large-scale processing as a core activity); if so, add contact details here.]

2. Overview: in which role we process your data

• As controller we process the data we need for your account, sign-in, management of the AI budget, Platform security and communication with you.
• Personal workspaces: if you use your personal workspace as a private individual, we are also the controller for the processing of your project content. If you use it for professional or business purposes and bring in personal data of third parties, we process that data on your behalf (next point). [TO BE VERIFIED: controller/processor delimitation for personal workspaces.]
• As processor we process personal data that Customers bring into their projects in team workspaces (e.g. names or third-party data in requirements, code or test data) and the data of the end users of our Customers' apps stored in the backends we provide. The respective Customer is the controller for this data; the data processing agreement concluded with the Customer applies. [TO BE VERIFIED: draft the DPA.]
• Partner organizations providing expert help act in their relationship with the Customer; we are not part of that relationship (section 6.8).

3. What data we process, for which purposes and on which legal basis

3.1 Accessing the website and technical logs

When you access the Platform, our servers and our hosting provider process technically necessary data: IP address, date and time, requested address, status code, amount of data transferred, browser and operating system details and error messages.

• Purpose: delivering the Platform, operational security, error analysis, defence against attacks.
• Legal basis: Art. 6(1)(f) GDPR (legitimate interest in secure and stable operation); when you are signed in, also Art. 6(1)(b) GDPR.
• Retention: application logs generally 30 days. [TO BE VERIFIED: retention of cloud logging is not explicitly configured; 30 days is the provider default.]

3.2 Registration, account and sign-in

• Data: e-mail address, name or display name, invitation code (which code was redeemed by whom and when), confirmation of the minimum age, time and version of the accepted Terms and of the acknowledged Privacy Policy, memberships and roles in workspaces and projects, language, profile and notification settings, session data. If you use two-factor authentication: the TOTP secret and recovery codes (stored with the authentication provider).
• Process: you sign in via a link you receive by e-mail (magic link). Sign-in, session management and two-factor authentication are handled by a specialized authentication provider (section 6.2), which processes your e-mail address, memberships and session data for this purpose and sends the sign-in e-mails.
• Personal workspace: on registration we automatically create your personal workspace; it carries your display name.
• Purpose: providing the account, access control, proof of the accepted contract terms.
• Legal basis: Art. 6(1)(b) GDPR (contract); for the proof of the accepted version additionally Art. 6(1)(f) GDPR (interest in evidence and defence of claims).
• Requirement: the e-mail address is required for use; without it no account can be created.
• Minimum age: the Platform is not aimed at children or adolescents. Users must be at least 18 years old. [TO BE VERIFIED: minimum age (Terms section 3(5)).]
• Protection against registration abuse (Friendly Captcha): on the registration page we use the bot protection Friendly Captcha of Friendly Captcha GmbH, Germany (section 6.6). Your browser solves a computational puzzle (proof of work); our server verifies the result with the provider. In particular your IP address and technical details of your browser and request are transmitted to Friendly Captcha. According to the provider, no cookies are set and no tracking takes place; processing takes place in the EU. If verification is not possible, registration is refused. We also rate-limit registration attempts and may reject disposable e-mail addresses. Legal basis: Art. 6(1)(f) GDPR (protection against automated mass registrations); for access to your device Section 25(2) no. 2 TDDDG. We store registration attempts for [30] days. [TO BE VERIFIED: The registration including Friendly Captcha is built but not yet switched on; confirm the provider's address, the EU endpoint, retention at the provider, DPA and classification under Section 25 TDDDG.]

3.3 Workspaces, invitations and collaboration

• Data: workspace name, kind (personal/team), members with role, invitations (invitee's e-mail address, inviting person, role, status), times of joining and end of access.
• Purpose: collaboration in the workspace, rights management.
• Legal basis: Art. 6(1)(b) GDPR; for invitees who do not yet have an account, Art. 6(1)(f) GDPR (the Customer's interest in inviting people to collaborate).
• Note: other members of a team workspace see your name, e-mail address and role. A personal workspace has no further members.

3.4 Projects, chat with Maira and AI runs

• Data: all content that you or your team members bring into a project or have generated there: chat histories with the AI assistant Maira, requirements and items, concepts, plans, decisions, code, files, images, run logs and reports, project settings (including credentials for the Customer's own services, stored encrypted in a secret store), and the record of who triggered or decided what and when.
• Processing by AI: to generate answers and carry out runs, we send the content needed in each case (e.g. your chat message with the conversation history, relevant project content, source code) via our AI gateway to providers of language models (section 6.3). Some of them are located outside the EU (section 7). Agents execute code in isolated execution environments on our infrastructure.
• Data use by model providers: we cannot rule out that model providers use the content transmitted to them to train their models or for other purposes of their own. Whether and for how long a provider stores content, whether it uses it for training or other purposes, and when it deletes it, is governed by the terms of the respective provider. We name the providers used in the list of sub-processors (Annex A). [TO BE VERIFIED: collect each provider's terms and link them in the list; check compatibility with Art. 28(3)(a) GDPR where we act as processor; possibly a transparency notice before the first chat.]
• Purpose: providing the contractually agreed services.
• Legal basis: Art. 6(1)(b) GDPR; where third-party data in team workspaces is concerned, we process it on behalf of the Customer (Art. 28 GDPR).
• Note: please do not enter special categories of personal data (e.g. health data) or third-party data that is not needed for app development in chats and projects. Use fictitious data for tests wherever possible.

3.5 Content checks, security and abuse detection

• Content checks: when runs start, when repository content is read in and for input, we automatically check texts for attacks (e.g. prompt injection, jailbreak attempts), malicious links and accidentally included secrets (passwords, keys). For this we use our own rules, a checking service of our cloud provider in the EU (section 6.1) and a language model as a judge (section 6.3; without EU-only a provider outside the EU). AI answers are filtered for secrets and internal Platform details before they are displayed.
• Abuse detection: in the execution environments we monitor compute load and network traffic (target host, port, data volume, decision – no content) to detect unrelated use such as cryptocurrency mining, attacks or data exfiltration. Anomalies are attributed to the workspace, User and run.
• Purpose: protecting the Platform, Customers and third parties; enforcing the Terms.
• Legal basis: Art. 6(1)(f) GDPR (security, abuse prevention); Art. 6(1)(b) GDPR.
• Retention: network logs and abuse indicators 90 days; the status of a suspension for as long as it lasts.

3.6 AI budget and billing

• Data: start credit per workspace; in team workspaces the budget and the monthly spending limit per User; in personal workspaces, in future, the booked price plan, its quota and its use as well as confirmations regarding the right of withdrawal; costs per AI call (model, token volumes, cost, time, attributed User, project and run), reservations, grants.
• Purpose: allocating and limiting AI use, cost transparency, later billing.
• Legal basis: Art. 6(1)(b) GDPR; for billing records Art. 6(1)(c) GDPR in conjunction with Section 257 HGB, Section 147 AO.
• [TO BE VERIFIED: once paid services are introduced, add payment provider, invoice data and any credit checks.]

3.7 E-mail notifications

• We send sign-in e-mails (via the authentication provider) and notifications about invitations, expert help and account events (via an e-mail delivery provider, section 6.5). Notifications contain no project content, only a fixed text and a link into the Platform. The recipient address is determined only at the time of sending.
• Legal basis: Art. 6(1)(b) GDPR; you can unsubscribe from optional notifications in your profile settings.
• We do not send advertising e-mails or newsletters. [TO BE VERIFIED: if newsletters are planned, add consent and double opt-in.]

3.8 Code repositories (GitHub) and access to your code

• Every project receives a private Studio Repository in a GitHub organization managed by us. Check and build steps partly run on GitHub servers.
• Customer Repository: you can connect a repository of your own to the project via our GitHub App. We then take changes from the chosen branch of your repository into Studio and transfer changes from Studio into your repository as a pull request. This is how you receive your code (section 9). You can end the connection at any time by uninstalling the GitHub App.
• Data: code and commit data (including name and e-mail address in commit metadata, where present), installation and repository identifiers, GitHub account name when connecting.
• Legal basis: Art. 6(1)(b) GDPR.

3.9 App preview

• For the in-browser preview of mobile apps we upload a build of the app to a device emulation provider (section 6.6). When you open the preview, your browser loads scripts and an embedded frame directly from this provider; your IP address and browser data are transmitted to it.
• Web previews and backends of Customer apps run on an instance that we operate ourselves in the EU.
• Legal basis: Art. 6(1)(b) GDPR.

3.10 Expert help (team workspaces only)

• For Customers: when an administrator requests expert help, the Platform prepares a case brief from the project data and shows it to suitable partner organizations admitted by us (in an open tender, a summary without the name of the Customer, the project or repository content). Only if the administrator selects this does the chosen partner receive the name and e-mail address of the requesting person. At engagement, the Customer invites a person of the partner into the project as an Expert; that person then sees the project content within the scope of the Expert role (no secrets, no billing or member data, no Maira chats of the Customer). Every access is recorded in a log visible to the Customer. We store requests, consents (including the version of the notices and a price snapshot), offers, ratings and decisions.
• For partner organizations: we process the application details (organization, legal entity, contact persons, competences, evidence, prices, website, languages), the admission decision and the acceptance of the Partner Platform Terms. The published profile (including website and ratings) is visible to Customers.
• Legal basis: Art. 6(1)(b) GDPR (contract with Customer or partner); passing on the requesting person's name and e-mail: Art. 6(1)(f) GDPR or the Customer's decision. [TO BE VERIFIED: legal basis for optionally passing on the contact data (consent of the person themselves rather than the administrator's decision?).]

3.11 Support and platform administration

• appvanced staff with administrative rights see workspaces (personal and team) and members only in masked form (e.g. "j•••@company.com"). Showing plain text requires a reason and is logged. We access project content only within a support access that the Customer approves in the individual case; it is time-limited and logged visibly for the Customer. [TO BE VERIFIED: admin area and support access have been decided but not yet built.]
• We analyse usage data only per workspace, and per project without names and content.
• Legal basis: Art. 6(1)(b) and (f) GDPR.
• Retention: log of administrative actions 3 years from the end of the year of the entry (contains only identifiers and codes).

3.12 Operations, quality assurance and improvement

• We log calls to language models (input, output, model, duration, cost, errors) in a self-hosted analytics tool (Langfuse) and in our AI gateway, both on our infrastructure in the EU (Google Cloud, GKE). Langfuse's transmission of telemetry data to its vendor is switched off. Both logs are deleted after 90 days by automatic deletion jobs. We use these logs to find errors, control costs and improve the quality of the agents. They may contain complete inputs and outputs, including project content.
• Legal basis: Art. 6(1)(f) GDPR (operational security, error analysis); Art. 6(1)(b) GDPR.
• Retention: 90 days, after which the logs are deleted.
• No training by us: we do not train AI models with your content. Analysing the logs for error analysis and to improve the agents is not training of models. For data use by model providers see section 3.4.

4. Cookies and local storage

We use only technically necessary cookies and entries in your browser's local storage. We do not use analytics, advertising or tracking tools, do not embed social media plugins and serve fonts from our own servers.

Legal basis: Section 25(2) no. 2 TDDDG (strictly necessary), Art. 6(1)(b) and (f) GDPR. No consent is required for this.

When you sign in, register (bot protection) or open the app preview, your browser connects to the services in sections 6.2 and 6.6; these may set their own technically necessary cookies or storage entries. [TO BE VERIFIED: check the cookies and device data set or collected by the third-party sign-in SDK and preview script themselves (device fingerprint?); classification under Section 25 TDDDG.]

5. Retention overview

We store personal data only as long as necessary for the respective purpose or as required by law.

Model providers apply their own retention periods (section 3.4).

[TO BE VERIFIED: automatic deletion after these periods has been partly decided but is not yet active technically (the database deletion routine currently runs in count-only mode; final deletion of organizations and users is not yet implemented, AIS-1760). Activate before launch or adapt the text.]

6. Recipients and processors

We use carefully selected service providers who process data on our behalf under Art. 28 GDPR. We name all of them – including all providers of language models – with purpose, seat, place of processing and transfer safeguard in our list of sub-processors in Annex A. The following sections summarize them. [TO BE VERIFIED: publish and maintain the list; conclude or document data processing agreements with all providers.]

6.1 Hosting and infrastructure – Google Cloud

The Platform (web application, database, agent execution environments, AI gateway, secret store, logs, analytics tool) runs on Google Cloud in the europe-west1 (Belgium) region. The content and secret checking service (Model Armor, Sensitive Data Protection) runs in europe-west4 (Netherlands). Provider: Google Cloud EMEA Limited, Ireland; access by Google LLC, USA, cannot be ruled out. [TO BE VERIFIED: confirm contracting entity and Cloud Data Processing Addendum.]

6.2 Authentication – Stytch

Sign-in, sessions and two-factor authentication: Stytch, Inc., USA. Processed: e-mail address, name, memberships, session and device data, TOTP data. [TO BE VERIFIED: place of processing, DPF certification or standard contractual clauses, possible EU data residency at Stytch.]

6.3 Providers of AI language models

For chat, runs and content checks we transmit the content needed in each case (input, conversation history, project content, source code) via our self-operated AI gateway to providers of language models:

• Default (without EU-only): by default we currently use models of MiniMax, Z.ai (Zhipu AI) and Moonshot AI (seated in the People's Republic of China, processing in China or Singapore). If a provider fails, the Platform may switch to a fallback model of one of these providers or of Alibaba Cloud (Qwen, Singapore).
• Selectable: per project, models of Anthropic (USA), OpenAI (USA), Google (Vertex AI, global processing), xAI (USA), Mistral AI (France, EU) and Alibaba Cloud (Singapore) can also be selected.
• EU-only: if you activate "Route only in the EU" in the project settings, that project uses only models that Mistral AI provides via its EU endpoint (processing in the EU according to the provider). In EU-only mode there are no fallback models outside the EU. EU-only is not the default.

For the use of content by the providers (training, retention, deletion) see section 3.4; for third-country transfers see section 7. The complete, current overview with models is contained in the list of sub-processors. [TO BE VERIFIED: per provider, confirm the DPA or data processing terms, contracting entity (EU or non-EU company) and place of processing.]

6.4 GitHub

Code hosting of Studio Repositories, synchronization with Customer Repositories and execution of check and build steps: GitHub, Inc., USA. [TO BE VERIFIED: DPF certification / standard contractual clauses; for Customer Repositories the Customer itself is GitHub's contractual partner.]

6.5 E-mail delivery – Brevo

Sending notifications: Sendinblue SAS (Brevo), France. Processed: e-mail address and delivery data; the e-mails contain no project content. [TO BE VERIFIED: confirm place of processing (EU) and Brevo's sub-processors; DPA.]

6.6 App preview – Appetize; bot protection – Friendly Captcha

• Mobile device emulation in the browser: Appetize.io, Inc., USA. Transmitted: the project's app build and, when the preview is opened, the Users' IP address and browser data. The app is deleted with the project. [TO BE VERIFIED: company name and address, place of processing, DPF / standard contractual clauses, DPA; note that test data in the preview app is processed by the provider.]
• Bot protection at registration: Friendly Captcha GmbH, Germany (section 3.2).

6.7 Self-operated services

We operate the following services ourselves on our infrastructure in the EU (section 6.1); no transfer to further third parties takes place: AI gateway (LiteLLM), analytics tool for AI calls (Langfuse), backends of Customer apps, agent execution environments, network proxy.

6.8 Partner organizations (expert help)

Within expert help, admitted partner organizations receive the data described in section 3.10. Customer and partner conclude their contract directly; we are not party to it. The partner processes the data under its own responsibility or on the basis of an agreement with the Customer. Partners are not sub-processors of appvanced. [TO BE VERIFIED: role of the partner (separate controller or processor of the Customer) and role of the Platform in the transfer; may partners be located outside the EU?]

6.9 Services connected by the Customer

If the Customer connects its own services (e.g. its GitHub account or Customer Repository or – in future – its own MCP servers), the Platform transmits data to these services at the Customer's instruction. The Customer is responsible for this transfer and the place of processing.

6.10 Other recipients

Authorities, courts and advisers (e.g. tax advisers, lawyers) receive data only where we are legally obliged to disclose it or where it is necessary to establish or defend legal claims (Art. 6(1)(c) or (f) GDPR).

7. Transfers to third countries

The Platform does not process data exclusively in the EU. Without EU-only mode, project content (chat messages, project content, source code) is transferred to model providers in third countries, in particular the USA, the People's Republic of China and Singapore. Further service providers (authentication, code hosting, app preview) are located in the USA.

• Safeguards: we safeguard transfers to third countries with the European Commission's standard contractual clauses (Art. 46(2)(c) GDPR) and assess them in a transfer impact assessment; where necessary we add supplementary measures. For recipients in the USA certified under the EU-US Data Privacy Framework we rely on the European Commission's adequacy decision (Art. 45 GDPR). There is no adequacy decision for China and Singapore.
• Risk: in third countries a level of data protection equivalent to EU law cannot be ensured in every case; in particular, authorities there may be able to access data without you having equivalent legal remedies.
• Your choice: if you want to avoid transfers of project content to model providers outside the EU, activate "Route only in the EU" (EU-only) in the project settings. This applies per project; service providers other than model providers (e.g. authentication, GitHub, app preview) are not affected.

The safeguard for each recipient is stated in the list of sub-processors. You can request a copy of the safeguards using the contact details in section 1.

[TO BE VERIFIED: standard contractual clauses with every recipient in third countries and the transfer impact assessments (in particular for China and Singapore) must be in place before production launch. Check DPF certification of each US recipient. For Customer content in team workspaces, regulate the transfer in the DPA (module 3 of the standard contractual clauses).]

8. Automated decision-making

We do not take decisions based solely on automated processing that produce legal effects concerning you or similarly significantly affect you (Art. 22 GDPR). Automated content checks and abuse detection may reject individual inputs or abort individual runs; suspensions of accounts or workspaces are decided by a human. The Platform proposes partners for expert help based on an automated fit; the Customer makes the choice. The bot protection at registration may refuse a registration if verification fails. [TO BE VERIFIED: classification of automatic run aborts, flagging of workspaces on suspected abuse and refusal by the bot protection.]

9. Your rights

Under the GDPR you have the following rights:

• access to the data stored about you (Art. 15),
• rectification of inaccurate data (Art. 16),
• erasure (Art. 17) – you can delete your account yourself in the settings (section 5),
• restriction of processing (Art. 18),
• data portability (Art. 20): you receive your code in a commonly used, machine-readable format by connecting a Customer Repository of your own (section 3.8). Other data you have provided to us (e.g. account data, chat histories) we provide on request to the contact details in section 1. [TO BE VERIFIED: define format and path for data other than code.]
• objection to processing based on legitimate interests on grounds relating to your particular situation (Art. 21),
• withdrawal of consent given, with effect for the future (Art. 7(3)).

Please contact kontakt@appvanced.de. If your request concerns data we process on behalf of a Customer (e.g. content in a team workspace), we forward it to the Customer or assist the Customer in responding.

Right to lodge a complaint: you can lodge a complaint with a data protection supervisory authority, in particular in the Member State of your habitual residence, place of work or place of the alleged infringement. The authority competent for us is: [competent supervisory authority] [TO BE VERIFIED: not named on appvanced.de. With the registered office in Neuss, presumably the State Commissioner for Data Protection and Freedom of Information of North Rhine-Westphalia (LDI NRW); confirm and add the address.]

10. Data security

We protect your data with technical and organizational measures, including encrypted transmission (TLS), storage of Platform data in the EU, role-based access control, encryption of credentials in a secret store, isolated execution environments for code, short-lived access keys limited to individual repositories, logging of administrative access and optional two-factor authentication. [TO BE VERIFIED: encryption of traffic between the nodes of the agent cluster is still open (AIS-1964).]

11. Changes to this Privacy Policy

We update this Privacy Policy when the Platform, the service providers used or the legal situation change. The current version with its version number is available at /legal/privacy, the current list of sub-processors in Annex A. We inform you of material changes in the Platform or by e-mail.

Annex A: List of Sub-processors

A.1 Purpose of this list

This list names all service providers that process personal data for appvanced Jan Kronenberg, Am Fuchsberg 2b, 41468 Neuss, Germany ("appvanced"), in operating appvanced AI Studio. It serves

• as the list of sub-processors where appvanced processes data on behalf of Customers (data processing agreement under Art. 28 GDPR), and
• as the list of processors and recipients where appvanced itself is the controller (Privacy Policy section 6).

We inform Customers with a data processing agreement of changes to this list in advance; they may object to a change in accordance with the data processing agreement. [TO BE VERIFIED: define period and procedure of advance notice, e.g. 30 days by e-mail to the owners of team workspaces.]

Abbreviations: SCC = standard contractual clauses of the European Commission (Art. 46(2)(c) GDPR); TIA = transfer impact assessment; DPF = EU-US Data Privacy Framework (adequacy decision, Art. 45 GDPR).

Important note on model providers: we cannot rule out that model providers use the content transmitted to them to train their models or for other purposes of their own. Training, further data use, retention and deletion are governed by the terms of the respective provider (column "Terms"). appvanced itself does not train models with Customer content.

[TO BE VERIFIED: SCC and TIA must be in place for every recipient in a third country before production launch. Check DPF certification of each US recipient and note the date. Confirm the contracting entity for each provider (several providers have an EU or Irish contracting company).]

A.2 Infrastructure and platform services

A.3 Providers of language models (via the AI gateway)

All model providers receive the content needed for the respective call: input and conversation history, project content, source code, possibly images; for content checks, the text to be checked. Which providers a project uses depends on the setting:

• Default = models preset for the agent roles without EU-only;
• Fallback = fallback model if a provider fails (not in EU-only mode);
• Selectable = selectable by the Customer per project;
• EU-only = the only providers when "Route only in the EU" is activated for the project.

[TO BE VERIFIED: (1) The model selection changes often; define how this list is updated on every change of the gateway catalogue (infrastructure/modules/agent-runtime-k8s/litellm.tf). (2) For every provider, collect and link the data processing terms, training opt-out, zero data retention and retention periods. (3) Whether model providers can be processors (Art. 28 GDPR) if they may use data for their own purposes; if not, classify them as separate controllers and clarify the legal basis of the transfer.]

A.4 Not sub-processors

The following are not in this list:

• Self-operated services on appvanced's Google Cloud infrastructure in the EU: AI gateway (LiteLLM), analytics tool for AI calls (Langfuse; logs 90 days), backends and web previews of Customer apps, agent execution environments, network proxy. They are covered by Google Cloud (section A.2).
• Partner organizations of expert help: they act under their own contract with the Customer, not on behalf of appvanced.
• Services connected by the Customer (own GitHub account or Customer Repository, in future own MCP servers): the Customer initiates the transfer.