What your compliance officer needs to know

Dictations pass through a client-side and server-side redactor (src/lib/pii-redaction.ts) that scrubs common identifier shapes — patient / provider names, DOB, address, phone, MRN, SSN, email — before the text is sent to Google Gemini through the Lovable AI Gateway.

We do not use dictations to train models. The AI provider does not retain prompts for model improvement under our gateway configuration.

• Dictations: stored in your browser's local storage only, for up to 10 days. They never reach our database.
• Account: email, display name, and subscription status.
• Billing: Stripe holds payment details. We store the Stripe customer / subscription IDs and invoice metadata, never card numbers.
• Redaction telemetry: counts of identifier categories that were stripped (e.g. "1 DOB, 2 phone"), never the identifiers themselves.

A BAA would require us to accept PHI. We deliberately do not — the product is architected around scrubbed, de-identified clinical descriptions. Offering a BAA without the corresponding infrastructure (segregated storage, audit logging, breach notification workflows, subprocessor BAAs) would be misleading. If your workflow requires PHI-in / PHI-out, this is not the right tool.

• Lovable Cloud (Supabase) — auth, database, and edge functions.
• Google (Gemini) via the Lovable AI Gateway — LLM inference.
• Stripe — billing and payment processing.
• Resend — transactional email (confirmations, receipts, dunning).

• Row-level security on every user-facing table.
• Server-side entitlement checks on Generate / Copy.
• PHI redactor runs before any text reaches the AI gateway.
• Stripe webhook signatures verified with HMAC + timestamp tolerance.
• Report a security issue: skazmi1991@outlook.com.

• Browser dictations: 10 days, then automatically purged.
• Account deletion: email skazmi1991@outlook.com and we remove your profile, subscription record, and any support correspondence.