VoiceJar is offline-first by design. Your recordings, transcripts, notes and API keys live on your phone. We do not proxy your data. We do not set tracking cookies. We do not sell anything to anyone.
The simplest path through the app is also the most private one. You don’t need to flip a switch to keep things on-device.
Every API key you bring — Anthropic, OpenAI, Gemini, Groq, ElevenLabs — is written to the OS Keychain (iOS) or Keystore (Android) via flutter_secure_storage. Never plaintext preferences, never our backend.
Whisper.cpp runs entirely on your device. If you never enter a cloud key, you can record, transcribe and summarize without your phone ever talking to a transcription server. Cloud is opt-in, per call.
Recordings, the Drift SQLite database, projects and tasks all live in the OS app-private sandbox. No other app can read them, and there is no “sync to our cloud” happening behind your back.
This website uses Umami — cookie-less, first-party analytics. No Google Analytics, no Tag Manager, no Meta Pixel, no ad-tech. Nothing follows you across the web because of us.
Stored in the OS app-private sandbox or in the OS secure-storage backend. Other apps cannot read it. We cannot read it.
Written into VoiceJar’s app-private directory. Orphan files are reconciled with the database on every launch so storage doesn’t drift over time.
Notes, transcripts, summaries, projects, tags, tasks. App-private storage, isolated by the OS sandbox.
OS Keychain (iOS) / Keystore (Android) via flutter_secure_storage. Never the Drift DB, never preferences.
The derived AES key for locked projects is cached in Keychain/Keystore so you can biometric-unlock instead of retyping the password every time.
Provider choice, language, trash retention, app-lock toggles — on-device only.
The downloaded transcription model sits in app storage and runs through your CPU. No round-trip required to transcribe.
When you tap a cloud provider, VoiceJar talks directly to that provider’s API from your phone. There is no VoiceJar middleware in between.
If you choose OpenAI Whisper API, ElevenLabs Scribe or Groq, your audio chunk is uploaded directly from your phone to their endpoint, using your API key. The provider’s privacy policy applies. If the call fails and you have a local Whisper model, VoiceJar retries on-device automatically.
Summaries, titles, tags and task extraction can use Claude, GPT, Gemini or Groq — you pick. Your transcript text is sent directly to the chosen provider with your API key. We are not in the loop.
The live conversation translator speaks the translation aloud via ElevenLabs or OpenAI text-to-speech. The translated text is sent directly to that provider with your key to synthesize the audio. Without a TTS key the translation stays on-screen as text — nothing is uploaded for speech.
If you subscribe to Pro, Stripe handles the checkout and card details. After confirmed payment, VoiceJar creates a license key and emails it to you for activation on Android or Windows — we never see your full card number.
To prevent endless trial resets, the app sends a one-way device hash on first launch so the 14-day trial cannot be farmed by reinstalling. The hash is not joined to any identity, never sold, never used for marketing.
Text and audio you send to a third-party AI provider (Anthropic, OpenAI, Google, Groq, ElevenLabs) are transmitted for one purpose only: to produce the transcript, summary, title, tags, task list or spoken translation you asked for. VoiceJar keeps no copy, runs no analytics on that content, and never uses it for training, advertising or resale. Because the call goes straight from your device with your key, we are technically incapable of doing so. Each provider processes the request under its own privacy policy — you choose which provider, and nothing is sent until you do.
Optional Pro sync mirrors your notes, projects, tags and tasks across your devices through a Supabase database — but every record is encrypted on your device with AES-256-GCM before it leaves. The server only ever stores ciphertext. The encryption key is generated on your phone and uploaded only in a form wrapped by a passphrase that only you know (Argon2id). Your passphrase never reaches our servers, so we are cryptographically unable to read your synced content — this is true zero-knowledge, end-to-end encryption.
The Vault is opt-in, per project. Use it for the projects that hold things you wouldn’t want anyone with momentary access to your unlocked phone to read.
AES-256-GCM for the encrypted payload (PointyCastle). Master password → vault key via PBKDF2-SHA256, 100 000 iterations. A stored verifier hash rejects wrong passwords up-front — you never get garbage out of a bad password.
Optional lock screen on every cold launch and resume. Biometric unlock (fingerprint or Face ID) via the OS authenticator, with PIN fallback. Off by default — enable it in Settings if you want it.
Every cloud call is made from your device to the provider you chose, using your key. There is no VoiceJar middleware that sees your plaintext, your audio or your responses. No backend logs — because there is no backend in that path at all.
Project export is in every tier — pull your notes out as needed. Trash retention is configurable; deleted notes linger only as long as you say. Uninstall removes everything. There is no shadow profile to clean up afterwards.
“If you never tap the cloud, your phone never talks to a transcription server. That’s not a setting. That’s the default.”
We’d rather be flat about the edges than ship copy we can’t back up. Here is exactly where VoiceJar is not the strongest claim you might read elsewhere.
The Drift database itself is not at-rest encrypted. Protection on the SQLite file relies on the OS app-private sandbox plus the optional app-lock. Only the contents of locked projects get the AES-256-GCM vault wrapper. If you need every note encrypted at rest, lock the project that holds it.
We do not claim end-to-end encrypted cloud sync. Cross-device sync is not the live product today. When it ships, we’ll publish the threat model and crypto choices on this page first — not in a press release.
When you use a cloud provider, their policy applies. We can’t make OpenAI, Anthropic, Google or Groq treat your data differently than they treat anyone else who uses their API. Pick the provider whose policy you’re comfortable with — or stay on Whisper offline.
iOS is coming soon. VoiceJar is currently distributed for Android and Windows. The iOS build is not publicly available yet.
No dark patterns, no consent walls, no creepy ad-tech. Same doctrine as the app: less surveillance, more clarity.
We use Umami to count page visits and see which pages help people. GDPR-friendly by design: no cookies, no cross-site tracking, only aggregated counts — no IP-based identity.
No Google Analytics. No Google Tag Manager. No Meta Pixel. No advertising SDKs of any kind. The only third parties on this page are Google Fonts (for the typeface) and our own server. That’s it.
Because the app stores your data on your device and the website doesn’t identify you, there is in practice nothing about you that we hold on our infrastructure. If you email us, the thread sits in our inbox — you can ask us to delete it at any time.
Use our contact form or write directly to info@voicejar.net. We’ll answer in plain language — not legalese.
Install the app, record your first note. Nothing leaves your phone unless you set up a cloud key and tap the cloud option. That’s the deal.