Subprocessors

To provide the Harmony service, we engage the following third-party entities ("Subprocessors") to process personal data on our behalf. We ensure that these Subprocessors meet strict security and privacy standards.

List of Subprocessors

• Anthropic (Claude): Language models that power live coaching, journal insights, session summaries, and message translation. Data is not used for training.
• Google (Gemini): Voice models for solo voice sessions and audio cleanup for journal dictation. Data is not used for training.
• Supabase: Database, authentication, and edge function infrastructure that backs the app.
• Apple: Sign in with Apple, App Store subscription billing, and APNs push notifications.
• Google (Sign in with Google): Optional sign-in provider for account creation.
• PostHog: Product analytics, error tracking, and session replay. Journal entries, partner chat, and other sensitive views are masked client-side before any recording is uploaded.
• AppsFlyer: Install attribution for paid advertising campaigns. Used only to measure which ad campaign (e.g., Meta, TikTok, Google Ads) brought you to Harmony. Sharing your device's advertising identifier (IDFA) requires your explicit consent via Apple's App Tracking Transparency prompt; if you decline, attribution falls back to Apple's privacy-preserving SKAdNetwork.
• Vercel: Marketing website hosting (talkharmony.app).

Live transcription for journal dictation runs on-device using Apple Neural Engine acceleration; no audio is sent to a third party for that path unless the on-device model is unavailable, in which case the recording is processed by Google (Gemini) as a one-shot transcribe and discarded.