Trust Center

Sign-in (email + password or Google) is implemented on our own EU infrastructure. Password hashes and account data live in our Frankfurt database, so no third-party identity provider stores your users. "Sign in with Google" is user-initiated and shares only your Google identity confirmation with us, nothing else.

• Deleting a recording removes the audio file, every transcript word, summaries, chat threads and generated documents, rows and storage objects alike, within seconds, via an auditable deletion job.
• Deleting a workspace runs every recording through the same pipeline, then purges workspace chat, tags, ingest addresses and connected cloud accounts including their access tokens.
• Data exports (Art. 20) are automatically deleted after 7 days.
• Database backups rotate within 7 days; deletions are re-applied on any restore.

Every provider is bound by a Data Processing Agreement under Art. 28 GDPR. This list is complete; new providers are added here before they process any data.

Optional cloud import connectors (Google Drive, Dropbox, OneDrive) transfer data only when you connect them yourself; they act on your instruction, not as our sub-processors.

We only use enterprise API tiers, never consumer or free tiers, because the terms differ sharply between them. What each provider commits to:

• Mistral AI (paid API): inputs and outputs are not used to train models. Content is retained up to 30 days for abuse monitoring, then deleted. The free "Experiment" tier, which does allow training, is not used for customer content.
• Azure OpenAI: Microsoft states that prompts and completions are not used to train OpenAI, Microsoft or third-party models. A subset of content flagged by automated abuse detection may be stored up to 30 days and reviewed by authorised Microsoft staff under EU Data Boundary rules.
• Google Vertex AI: under the Service Specific Terms ("Training Restriction"), Google does not use customer data to train or fine-tune models without instruction. We do not use the Gemini Developer API, whose free tier would allow product improvement from prompts.
• Google Cloud Speech-to-Text: data logging is opt-in and we do not opt in, so audio and transcripts are used only to return the result.

In short: no provider trains on your content, and the short abuse-monitoring windows above are the honest exception to "nothing is retained". Zero-retention options exist at the enterprise tiers of these providers and are available on request for larger deployments.

• In transit: TLS on every connection, browser to app and app to every provider. HSTS (1 year, includeSubDomains, preload) prevents protocol downgrade.
• At rest: database and backups encrypted by the hosting provider; every audio object encrypted with AES-256 in storage.
• Application level: credentials of connected cloud accounts (OAuth tokens) are additionally encrypted inside the database; the key lives outside it, so a database copy alone reveals no third-party access. Passwords are hashed with Argon2id, API tokens stored only as SHA-256 hashes.
• Honest scope: conversation content itself is protected by forced tenant isolation and at-rest encryption, not field-level encryption, because the product computes on transcripts (search, word-level citations), which field encryption would break. Customer-managed keys are on the enterprise roadmap.

• Tenant isolation enforced in the database itself (PostgreSQL row-level security on every business table, forced for all roles).
• Background jobs re-validate tenant and object ownership before touching any data.
• Audio is served through short-lived signed URLs (10 minutes); the storage bucket is private.
• API tokens are stored as SHA-256 hashes; OAuth connector tokens are encrypted at the application level (key held outside the database) and deleted with the workspace.
• TLS everywhere with HSTS; encryption at rest for database and object storage; passwords hashed with Argon2id.
• Complete audit trail for privacy-relevant actions (export, deletion, access changes).

Access and portability (Art. 15/20): every workspace can export its full archive as JSON from the settings page. Erasure (Art. 17): recordings and whole workspaces can be deleted in the product, effective immediately as described above. For anything else, rectification, objection, or questions, write to info@heydiane.ai; we answer within 30 days.

Data Processing Agreement (Art. 28), template · Technical and Organizational Measures (Art. 32) · Data Protection Impact Assessment (Art. 35)
Privacy Policy · Terms of Service
Individually signed DPA copies and security questionnaires: info@heydiane.ai