How does voice capture work?

Hold the mic, speak briefly — KeptMind turns your words into structured tasks in under 12 seconds. No forms, no typing, no project selection required.

Voice capture is the core interaction in KeptMind. The design goal: get a thought from your head into a structured task faster than any other method, with zero executive function required beyond speaking. The target is under 12 seconds from thought to saved task.

How to capture

From the lock screen: tap the KeptMind widget (iOS) or notification shortcut (Android). The mic activates immediately — no unlock required.
From inside the app: tap the microphone icon on the Capture tab.
Speak naturally: one sentence is enough. "Call dentist tomorrow morning" or "the budget thing for Marek, can we do it Wednesday?" — both work. Messy is fine; the AI handles interpretation.
Stop speaking: the capture auto-saves after a brief silence (about 1.5 seconds). Or tap the stop button manually.
Review (optional): the parsed task appears immediately. You can edit the title, date, or energy level before confirming — or just let it save as-is.

What happens after you speak

The audio uploads to our EU servers, gets transcribed into text, and then the AI parsing layer interprets the text into a structured task:

Task title: extracted from the core action in your sentence
Date/time: if you said "tomorrow", "Friday", "next week", etc., the task gets scheduled accordingly
Energy level: inferred from context — "quick email" gets tagged low-energy; "write the proposal" gets tagged high-energy
Priority: if you said "important", "critical", "urgent", or similar, the task gets flagged

The parsed task lands in your Today list (if scheduled for today) or Backlog (if scheduled later). You do not need to assign a project, pick a category, or fill any form.

Tips for better captures

Say the action verb first: "Schedule dentist" works better than "I should probably schedule the dentist sometime"
Name people directly: "Text Marek about budget" beats "text him about it" — the AI uses names for context
Include dates naturally: "Friday at 3pm" or "next Monday" — the date parser handles most natural-language formats
Keep it short: the median successful capture is under 12 seconds. Longer captures (60+ seconds) work but produce less precise parsing
Do not worry about grammar: "uh, the thing for the meeting, the slides, can I do those Wednesday" is a perfectly valid capture

When voice does not fit

Voice capture is not always the right mode. In meetings where speaking aloud is awkward, in quiet shared offices, or when you need to capture something very precise (a URL, a phone number), text capture is better. The Capture tab has a text input alongside the mic — same AI parsing, just typed instead of spoken.

The lock-screen widget is what makes voice capture genuinely fast. Without it, you need to unlock → find app → open → tap mic (4 steps, ~8 seconds). With it, you tap once from the lock screen (1 step, ~2 seconds to start recording).

iOS: Long-press the Lock Screen → Customize → Add Widget → KeptMind → Mic widget. Place it where your thumb naturally rests.

Android: Add the KeptMind widget to your home screen. On Android 13+, you can also add it to the lock screen via Settings → Lock screen → Widgets.

Frequently asked questions

How accurate is the transcription?

For clear English speech in a quiet environment: 95-98%. For ADHD-style speech (fast, hesitant, trailing off) in normal environments: 85-92%. The AI parsing tolerates common transcription errors — even at 85% accuracy, the resulting task is usually recognizable and actionable.

Does it work in Estonian?

Yes. Set capture language to Estonian in Settings → Voice. Accuracy is slightly lower than English (80-90%) but sufficient for daily use.

What if I capture something private?

Audio is deleted within 24 hours of transcription. The text transcript stays in your account (encrypted at rest) until you delete it. We do not train AI models on your captures. See Are my recordings secure? for the full privacy architecture.

Can I capture multiple tasks in one recording?

Yes — if you say "call dentist and also email Marek about the slides", the AI often splits this into two separate tasks. It works about 80% of the time for two items; for three or more, it is more reliable to do separate captures.

What is the maximum capture length?

There is no hard limit, but captures over 60 seconds produce less precise parsing. The sweet spot is 5-15 seconds per thought. If you have a longer brain dump, the 30-second voice dump technique works better than one long recording.