Processing Details

Version: 1.6
This English version is provided for convenience. The German version controls.

1. Subject matter and purposes

Doli processes Customer Data for transient real-time interpretation, chat and subtitle translation, image/PDF/document explanation, suggested replies, secure routing, feature control, and deletion. Doli separately processes account, contract, purchase, and its own security purposes as an independent controller.

2. Categories of data subjects

• authorized users and Customer personnel;
• Customer clients, prospects, visitors, and suppliers;
• conversation partners and people appearing in submitted content;
• business-account representatives and administrators.

3. Categories of personal data

Customer-directed content: microphone and speech audio, source and target language, chat, subtitles, response wishes, photos, scans, images, PDFs, document content, and generated translations or explanations.

Required routing data: short-lived technical session, ticket, language, format, and error data strictly necessary for immediate content processing.

Special-category or criminal-offence data may occur in free speech or documents although Doli is not designed for it. The Customer may submit it only with a lawful basis, necessity, and appropriate safeguards.

4. Processing operations

Receipt, transmission, authentication, validation, bounded buffering, analysis, translation, generation, output, blocking, and deletion. Doli does not sell Customer Data or use it for advertising, profiling, or training its own or third-party general AI models.

5. Retention and deletion criteria

Category	Criterion
Content in Doli systems	Bounded active-request/session memory and technically necessary temporary recordings on the device; temporary recordings are deleted after processing or cancellation; discarded on completion, cancellation, or secure termination. No Doli conversation history or content in Doli logs/backups.
Google Vertex AI	Zero data retention for Google Cloud project 674711097607 as defined by Google: approved exception from abuse-monitoring prompt logging, data cache disabled, no storing features. Content is processed only for the respective request.
Google features	No Live Session Resumption, Search/Maps Grounding, File API, stored Interactions, or explicit context caching; the project's data cache is disabled.
Routing data	Only as required for active processing, secure termination, and coarse error handling; then deleted or anonymized.
Mandatory evidence	Restricted and purpose-limited only for the applicable statutory period.

Remaining Google retention is a documented service limitation; Doli cannot cause early deletion through the app.

6. Doli independent-controller processing

The DPA excludes account/authentication, business profile, representative, role selection, acceptance/evidence, web or earlier Store purchase, entitlement, allowance, reconciliation, accounting, legal defense, and Doli's own support, security, fraud, abuse, and reliability data. Doli's privacy notice governs it. Annex 3 describes Google's separate role for account, billing, communications, feedback, and technical usage data.

Since 25 September 2026, Doli processes all content through Google Cloud Vertex AI: spoken translation and the language request at start with Gemini 3.8 Live, chat, captions and letter and image explanations with Google's Gemini Flash models. The model location is set to the Google Cloud EU multi-region for all of these features; in normal operation they all use the EU endpoint of the Vertex API. The same route serves personal and business accounts; there is no longer a commitment to Germany. The Gemini Developer API of Google LLC is no longer used.

For captions, Doli's relay splits the microphone signal by loudness into speech segments and sends each segment as a short audio file inline in the request to a Gemini Flash model; neither the File API nor Google Cloud Storage is used for this. The relay only detects pauses in speech; it identifies no persons and creates no voice profiles. The segments stay only in memory until the translation returns (at most 12 seconds) and are not stored. As a fallback, captions can temporarily run on Gemini 3.5 Live Translate. The relay then streams the microphone signal continuously through the global Vertex endpoint, because Google does not offer this model on the EU endpoint; the model location remains the EU multi-region. In both cases the same subprocessor processes the same data types for the same purpose.