Sub-processors
The processing agreement governs the list of sub-processors that fall under the data processing of AI-Public. Other vendors or internal administrative systems are not included here, as they fall outside the scope of processing in AI-Public.
Google Cloud
We use Google Cloud for hosting the virtual environment of AI-Public.
Google Cloud processes name, email address, and chat history.
Storage and processing take place in the Netherlands and Belgium. Google Cloud lists the United States as its country of establishment.
AI-Public uses Google Cloud’s Cloud Data Processing Addendum. Google Cloud is ISO 27001 certified.
Google Identity Toolkit
Google Identity Toolkit is used for user authentication.
Google Identity Toolkit processes email address, IP address, user agent, and user ID.
Storage and processing take place in the Netherlands and Belgium. Google Identity Toolkit lists the United States as its country of establishment.
Google Gemini
We use Google Gemini as the AI model that allows users to ask questions, generate texts, and analyze data.
Google Gemini processes input, prompts, and chat history.
Storage takes place in the Netherlands and Belgium. AI processing of prompts may take place worldwide, including outside the EU. Google Gemini lists the United States as its country of establishment.
With Gemini models, Google may temporarily store and reuse the beginning of a prompt (implicit caching), never longer than 24 hours and only within AI-Public's Google Cloud project. This is a project-level setting that is the same for all apps; AI-Public cannot turn it on or off per user or per request.
OpenAI
We use OpenAI for GPT models.
OpenAI processes input, prompts, and chat history to generate responses.
OpenAI's European environment is used: storage and processing of prompts take place entirely within the EU. Zero Data Retention applies: OpenAI does not retain input and output after the response has been generated. A signed processing agreement with OpenAI is in place.
In the desktop apps, AI-Public can use OpenAI's prompt caching, but only with your consent. The app asks the first time you use an OpenAI model; your choice is stored with your account and can be changed at any time in Settings. OpenAI then temporarily stores an encrypted, derived computational representation of the beginning of a prompt (not readable text) in the memory of its processing servers, within the same processing region, for at least 30 minutes and never longer than 24 hours. Caches are not shared with other organizations; each user has their own pseudonymized cache key. Without consent, and in the web and mobile apps, AI-Public does not use prompt caching. AI-Public does not let OpenAI store responses for later use.
Anthropic
We use Anthropic for Claude models.
Anthropic processes input, prompts, and chat history to generate responses.
Processing takes place in the United States. Anthropic lists the United States, San Francisco as its country of establishment. A Data Processing Addendum with Anthropic is in place.
In the desktop apps, AI-Public uses Anthropic's prompt caching. Anthropic then stores a derived computational representation and checksums of the beginning of a prompt (not readable text) in the memory of its processing servers, for up to 5 minutes after the last use. Caches are not shared with other organizations. In the web and mobile apps, AI-Public does not request prompt caching from Anthropic.
Mistral AI
We use Mistral AI for Mistral models.
Mistral AI processes input, prompts, and chat history to generate responses.
Processing takes place in France. Mistral AI lists France as its country of establishment.
Mistral automatically stores the beginning of a prompt temporarily and reuses it (prompt caching); it does not disclose in what form or for how long. This applies to all apps; AI-Public cannot turn it on or off per user or per request.
No sub-processors: BFL and Stability AI
Black Forest Labs (BFL/FLUX) and Stability AI are not sub-processors of AI-Public. The image links are technically constrained so that these providers cannot receive personal data from users or images from the application.