Home screen
In-app record button
Single tap on the home screen to start, tap again to stop. Pause and cancel from the foreground notification while recording is in flight.
No vapor, no roadmap items dressed up as today. Everything below ships in the current Android build and is traced against the source tree before it lands on this page.
Four ways to start a recording, all designed to remove friction between thought and saved audio.
Single tap on the home screen to start, tap again to stop. Pause and cancel from the foreground notification while recording is in flight.
A system-wide overlay that records from any other app. Brainstorming in your browser, on a call, in another note tool — one tap and it’s captured.
Pick existing audio files from the device — old voice memos, lectures, podcasts — and run them through the same transcription + AI pipeline.
Built-in, headset or Bluetooth — the input device list is enumerated so you choose which mic actually captures the audio.
Recording and transcription run under an Android foreground service with a CPU wake-lock — safe to lock the screen, switch apps, take a call.
No signup, no email, no payment. The 14-day Pro trial begins automatically. You can record within seconds of opening the app for the first time.
Every new recording lands in the Inbox unfiled. You don’t have to pick a project at capture time.
Every recording starts unfiled, on your home screen, until you (or the AI) move it. No mandatory categorisation at capture time.
Free on every tier. Custom color and emoji icon per project — flat list, no nested hierarchy to fight with.
AI tag suggestions reuse your existing taxonomy so tags don’t multiply. Global tag filter to slice across projects.
Deleted notes move to a system Trash project with a configurable retention window. Nothing disappears the moment you tap.
Multi-select N notes → merge into one. Audio files join with a 1-second gap, transcripts concatenate. Free on every tier.
On every app launch an audit reconciles audio files on disk with notes in the database. Broken references can’t leak storage over time.
The post-capture pipeline runs five LLM steps, each gated by your settings. Pick a different model for each step if you want.
Long transcripts are chunked and merged so hour-long lectures don’t blow context windows. The AI suggests project moves — it never silently re-files behind your back.
Every BYOK API call goes straight from your device to the provider you chose. No VoiceJar middleware in the path. Per-task model routing: short, medium, long, fast, strong, rewrite slots — mix providers freely.
Pick the engine that matches the moment — private and offline, or cloud-fast with your own key.
Fully on-device, no internet. Choose a model from tiny to large-v2. No cost, no quotas, no metadata leaving your phone.
The whisper-1 cloud endpoint. Bring your own OpenAI key — billed on your account, not ours.
Cloud STT with optional speaker diarization (Scribe v2). Speaker labels are localized in 14 languages.
Groq’s LPU runs Whisper Large v3 at very high speed — long recordings transcribe in seconds.
If a cloud transcription call fails and you have a Whisper model downloaded, the request retries on-device. The result is flagged so you know.
Failed cloud calls don’t consume your daily quota slot. Cancel an in-progress transcription directly from the notification.
The surfaces that turn a pile of recordings into a working system.
Speak a to-do — VoiceJar files it as a task with the date parsed from natural language. Optional daily reminder at the hour you choose.
Read-aloud playback via ElevenLabs TTS or OpenAI TTS. Useful for re-listening to summaries on a commute or while walking.
Every note from the last 24 hours, pending tasks, total minutes recorded — rolled up on-device. No extra AI call, no email send.
Each speaker talks in their own language. When a turn ends, VoiceJar transcribes it, translates it, and — with a TTS key — reads it back aloud in the other language. Back and forth, alternating.
Voice-activity detection ends a turn when you stop speaking — no button to hold. The reply lands on the other side, rotated to face them.
Speech-to-text can run offline (Whisper.cpp), but translation uses a cloud LLM and the spoken read-back uses a cloud TTS key. It’s cloud-assisted — not a fully-offline translator.
Conversations are stored locally — reopen any past chat as “EN ↔ SK · 12 turns” with both sides, the translated text and the original audio per turn.
On Free BYO Keys, translations share the 5/day cloud cap with transcription — a failed turn never burns a slot. Unlimited on Pro or with your own LLM key.
No internet, no translation. The transcription step works on a plane, but translating and speaking a turn both need a connection. Without a TTS key, the translation still appears as text — it just isn’t read aloud.
VoiceJar doesn’t tap your call audio itself. It listens for “call ended” events and imports the recording BCR (Basic Call Recorder) just produced.
Each imported call gets a clean header: phone number, contact name, date, duration, direction — all extracted from the BCR filename pattern.
Once imported, the call flows through the same transcription + AI pipeline as any other recording. Summary, title, tags, tasks — just like a voice note.
Call recording isn’t gated behind Pro. T0 Trial, T1 Free BYOK and T2 Pro all import BCR calls.
Disclosure. Two-party-consent laws apply — the legal responsibility for recording a call lies with you. VoiceJar imports what BCR captured; it does not tap the call audio itself.
Hardware-backed key stores hold your API credentials. We don’t proxy your requests, we don’t log your prompts, we never see your keys.
Every BYOK key (Anthropic / OpenAI / Gemini / Groq / ElevenLabs) lives in the OS secure storage. Never written to SharedPreferences or the SQLite database.
Mark a project as locked — titles and transcripts are encrypted with AES-256-GCM, key derived with PBKDF2-SHA256, 100k iterations. Biometric unlock optional.
Optional Face ID, fingerprint or PIN fallback for the entire app, not just the vault. Toggle in Settings.
Zero VoiceJar backend in your AI calls — every request hits the provider directly. We can’t see your prompts because we’re not in the network path.
A verifier hash rejects an incorrect vault password — it can’t silently return garbage and pretend it decrypted.
This website runs Umami — no third-party tracking, no cookies, no Google Analytics. Your visit isn’t a tracked event.
Built for the polyglot inbox. The interface speaks your language, the transcription speaks the language you spoke.
English, Slovak, Czech, German, French, Spanish, Portuguese, Italian, Hungarian, Polish, Japanese, Chinese, Russian, Ukrainian.
Whisper is multilingual out of the box; cloud providers support 50+ languages. AI summaries and titles respect the language you spoke.
Requires Android 7.0 (Nougat) or newer. Floating bubble, foreground transcription service and BCR call import are Android-only.
Codebase compiles for iOS and the design is ready; App Store distribution is planned for a follow-up release.
No signup, no payment, no email. After 14 days you soft-land on Free BYO Keys, with every recording you ever made still on your device.