How Whisperit uses AI

Whisperit is a multi-tenant cloud service operated by Whisperit SA. You do not host it yourself. All persistent data (accounts, documents, transcripts, and document embeddings) is stored in Switzerland: compute, document storage, OCR, and vector search run on Microsoft Azure in the Switzerland North region, and the PostgreSQL database and the web application are hosted by Infomaniak in Geneva. Backups and replicas stay in Switzerland. Swiss Only is the default: every AI request stays in Switzerland, and if the Swiss provider is unavailable a request fails rather than being rerouted to another jurisdiction. Two further tiers are available on request, one processing through an EU-based provider and one adding an international fallback for availability.

AI is a core function of Whisperit, not an optional add-on. Dictation and transcription, document drafting and editing, summarisation, semantic search across a firm's own case files, and research assistance are all AI-driven; without them the product would not do its job. AI never acts alone. Every AI output is a proposal shown to the user for review, editing, and explicit acceptance before it becomes part of a document or a case file. Nothing is filed, sent, or decided autonomously, and every access, edit, and export is recorded in an audit log.

Large language models (LLMs) for drafting, editing, summarisation, and conversational assistance. Automatic speech recognition (ASR) for dictation and audio transcription, including speaker diarisation. Optical character recognition (OCR) for scanned documents. Text embeddings combined with vector search, used for retrieval-augmented generation (RAG) so that answers are grounded in the firm's own documents rather than in a model's memory. Text-to-speech for reading documents back.

Whisperit does not build its own foundation models. We integrate third-party models under enterprise agreements. Language models: hosted in Switzerland, on Microsoft Azure in Zurich and by Infomaniak Network SA. Speech: Azure OpenAI transcription models and Azure Speech Services, both running in Switzerland. Model routing is configured per tenant by us rather than exposed in the interface, so it cannot be changed by accident. The complete provider list is published on our sub-processors page.

No. Whisperit does not fine-tune models on client data, and no provider we work with is permitted to either. Every AI provider is bound by an enterprise agreement that contractually prohibits using client data for training, fine-tuning, or model improvement, and processing is transient: nothing is retained by the provider beyond the duration of the API call. Whisperit is adapted to legal work through prompt engineering and retrieval over the firm's own documents (RAG), not by training a model on client content.