Privacy

Zero-Data-Retention (ZDR)

• For AI inference requests, we send your input via OpenRouter (OpenRouter, Inc., USA) to the provider of the selected model.
• We have configured these requests so that OpenRouter only forwards them to provider endpoints that, according to their policy, neither store content nor use it for training (zero data retention). If no such endpoint is available, the request is rejected.
• Zero data retention applies to these AI inference requests, not to voice input, voice output, web search, or email sending (see below).

AI Processing & Service Providers

To provide the features, we transmit certain content to carefully selected third-party providers (processors):

• What data: your messages and the prior chat history, attached files and images, voice recordings; for research, search queries and retrieved web addresses; for email sending, the result you send or, for account emails, your email address.
• AI processing: OpenRouter (OpenRouter, Inc., USA) forwards AI inference requests (chat, research and agent features, file and image analysis, automatic title generation) to the provider of the selected model, with zero data retention as described above. Processing may take place outside the EU. The transfer to OpenRouter is based on EU Standard Contractual Clauses.
• Voice input: Spoken input is transcribed by Mistral AI via OpenRouter. These requests only go to endpoints that do not use content for training. Zero data retention is currently not available for this route.
• Voice output: On the web, we send the text to be read aloud directly to Mistral AI (mistral.ai). Use for training is disabled in our Mistral account. In the iOS app, text is read aloud on the device.
• Web search: For research, our research server sends search queries to the Brave Search API (Brave Software Inc., USA). The AI model formulates the search query from your message. It contains no account or session data and no chat histories, but may contain information you write into your question yourself. Brave may retain search queries for up to 90 days for billing and troubleshooting. When the research server retrieves individual web pages, the page only sees our server's address and the requested URL, no data about you. If our research server is unavailable, web search may fall back to OpenRouter.
• Email sending: Via Resend (USA) we send results that you send by email, as well as account emails, in particular for confirmation, password reset, and email address changes. Resend stores email content and logs for 30 days. The transfer is based on EU Standard Contractual Clauses.
• For what purpose: to answer your requests, power research and agent features, generate titles automatically, transcribe speech, produce speech output, send results by email at your request, and deliver account emails; also for the development and testing of new features, in particular for processing longer chats. Your content is not used to train AI models in the process.

Hosting & GDPR

• We run our application and the permanent storage of your data ourselves on our own servers at Hetzner in the EU. This also applies to our research server.
• Cloudflare protects access and forwards traffic encrypted.
• We delete logs of our database, API, sign-in, storage, and realtime services after 30 days. Server operating logs are rotated automatically based on size. Technical operating logs generally do not contain chat or search content.
• GDPR-compliant.

Legal Bases

• Art. 6(1)(b) GDPR (performance of a contract): account and authentication, chat and AI processing, research, voice features, sending results by email at your request, required account emails, payments and subscriptions.
• Art. 6(1)(c) GDPR (legal obligation): statutory retention of payment and invoice records.
• Art. 6(1)(f) GDPR (legitimate interest): technical operating and error logs, security, abuse prevention, and development and testing of new features.

Account & Security

• Authentication via JWT; sign in with email/password, Google, GitHub, or Apple.
• Row-Level Security (RLS) on all tables, with access limited to your own data.
• No secrets in the client; server-side validation of all inputs.
• Rate limiting on all API routes; origin lockdown via Cloudflare Tunnel.
• Uploaded files and manually saved artifacts are stored in private storage buckets and provided via time-limited signed URLs. Automatically saved documents are stored as text in an access-protected database table.

Automatically Saved Documents

Automatically saved documents remain stored until they are removed through an intended deletion process. When you delete an associated chat, the documents tied to it are deleted as well; deleting your account also removes the saved documents. Editing a response in the chat does not change a document that has already been saved.

Deleting a Chat

When you delete a chat, we remove the chat, its messages and the content automatically saved from it as documents. For agent and research runs from this chat, we delete the request, the answer and the content of the individual steps. The billing and measurement values of these runs (for example credits used, number of web searches and token counts) are kept without any content. Your research documents, uploaded files and saved artifacts are kept as well; you can delete each of them individually. Runs that are no longer clearly linked to the chat (for example because you edited or regenerated the answer) stay in your run history, where you can delete them individually. If an agent is still running in the chat, the chat can only be deleted once it has finished.

Deleting Your Account

When you delete your account, all of your own data (chats, projects, files, artifacts, research documents, agents) is removed; an active subscription is canceled beforehand. If you only use your account for Agentic Chat, your login is deleted as well. If you also use the same account for Imitation AI, the shared login remains so that service keeps working; your Agentic Chat data is deleted regardless.