What it collects
Type Studio sees only the data you hand it. No analytics SDKs, no Google Analytics, no Segment, no pixels. The backend runs on Google Cloud Run, which keeps standard infrastructure logs (IP, user-agent, HTTP status) used only to debug outages.
Sign-in runs through Supabase Auth with email and password. Supabase stores the session record. Song search runs server-side against Last.fm, so only the song and artist text you type leaves the backend. Analysis runs on a song's 30-second preview or on a file you upload, never on a full-length catalog track.
Your work is saved per account, each artifact keyed to your user id. The Recommender and Song Report write result JSON. Prompt to Song and the Song to Prompt optimizer write their audio to Cloud Storage and per-iteration metadata to Firestore so runs resume across devices. Typer Vision writes its rendered 360 poster as a PNG, and its Reconstruct Loop keeps every round's audio, DSP scores, and poster. MIDI Render keeps the WAV, the .mid, and the transcribed-notes JSON. Song Structure, Compare Two Songs, and Arabic Fusion store their results. If you upload your own audio, its embedded tags (artist, title, ISRC) are stripped before it is stored or analyzed.
Who gets your data
Every external call comes from the backend, not your phone. Google Vertex AI runs Gemini for the Recommender, Song Report, the Song to Prompt optimizer, Typer Vision, Song Structure, Compare Two Songs, and Arabic Fusion, so your seed song, the Last.fm candidate pool, and your refinement text go there. For Song Structure and Song Report the song's audio is sent so Gemini can hear it; for Typer Vision the rendered 360 poster is sent so Gemini can read it. Vertex also runs Google's Lyria for Prompt to Song, and the deep-research dossier runs Gemini with Google Search, which carries the topic about the song, not your identity or IP.
Modal runs the ACE-Step generator on its GPUs. When you pick ACE-Step for the optimizer or run the Typer Vision Reconstruct Loop, each round's text prompt and settings (seed, length) are sent to Modal, not your identity or sign-in. Last.fm, iTunes, Deezer, Apple Music, and MusicBrainz receive only song and artist names so the backend can resolve a song and fetch its 30-second preview. Apple, Google Play, and RevenueCat handle the Type Pro subscription and receive a pseudonymous account id and purchase events, not prompts or audio. No provider receives your password.
What it generates
Audio, not just analysis. Prompt to Song runs Lyria. The Song to Prompt optimizer runs Lyria or ACE-Step in a Gemini-driven loop, grading each round against a reference with DSP measurements (tempo, harmony, timbre, brightness). Typer Vision's Reconstruct Loop climbs back toward a song's 360 view the same way. MIDI Render plays the exact transcribed notes through a FluidSynth soundfont. Generated output belongs to you, is stored under your account, and is never used to train any model.
Subscriptions
Type Pro subscriptions are sold and processed by the Apple App Store or Google Play, depending on platform. The store handles the entire transaction, so Type Studio never sees your card. RevenueCat sits between the store and the backend to confirm an active subscription, receiving a store purchase token and an app-user id, not the payment instrument.
What it doesn't do
No data sold, no advertising, no training on your data or output. Google Vertex AI and Modal each operate under their own retention policy for API traffic. Google Vertex AI's enterprise terms state that data submitted through Vertex is not used to train Google's foundation models.
Where your data is processed
The backend runs on Google Cloud (Cloud Run and Vertex AI) in the United States, and the ACE-Step generator runs on Modal, also in the United States. The providers it calls (Supabase, RevenueCat, Last.fm, iTunes, Deezer, MusicBrainz, Apple Music) process requests on their own infrastructure, in the US or elsewhere. If you use the app from the European Economic Area, the UK, or another region with cross-border transfer rules, your data is transferred to and processed in the United States under each provider's standard contractual clauses (or equivalent transfer mechanism).
Cookies and storage
No cookies. The session token is kept on-device in the iOS Keychain or Android Keystore. Clearing the app or signing out removes it.
Retention
Recommender and Song Report results and generated audio stay until you delete them; failed partials auto-clean after an hour, and in-memory run records evict two hours after a run finishes. The optimizer keeps every iteration's audio and metadata, Typer Vision keeps each poster and every Reconstruct Loop round, and MIDI Render keeps each render, all until you delete them. Deleting a run removes its audio, JSON, and Firestore documents. A full wipe removes everything keyed to your account.
Your rights
Canada (PIPEDA), EU and UK (GDPR), and California (CCPA) rights apply: access, correction, deletion, portability. No data is sold. To exercise them, email [email protected] from your sign-in address.
Children
Not directed at children. No knowing collection from anyone under 13, or under 16 in the European Economic Area.
Changes
Material changes update the date above and are flagged in the app at least a week before they take effect.
Questions, export, or deletion: [email protected]. See the terms of use, the FAQ, and support.