Voice transcripts
Captured with iOS on-device speech recognition. Raw audio is discarded immediately. Never leaves your device. Enforced in code.
Most ADHD apps route the stuff you hate most about yourself through someone else's servers. KickMint does not. Every sensitive piece stays on your device, enforced in code you can audit.
Voice transcripts
Captured with iOS on-device speech recognition. Raw audio is discarded immediately. Never leaves your device. Enforced in code.
Cycle, sleep, medication patterns
Listed in nonSyncableTables. Excluded from every sync payload by construction. Never leaves your device. Enforced in code.
Task titles and AI breakdowns
Prompts and outputs are generated locally by llama.cpp. No network call during inference. Never leaves your device. Enforced in code.
No streak shame, no dashboards no one opens, no AI upsell for features that should be free. Six things that earn their place on your home screen.
Tap any task. Get 3 to 5 micro-steps in seconds, sized for how you actually work. Model runs on your phone.
"When you sit at your desk, open the document." Implementation intentions that nudge you into the next action.
Mic on the home screen widget, lock screen, or in-app. Speak the thought, keep walking.
Patterns surfaced when useful, never weaponized against you. No red. No scolding. No dark patterns.
7-day pause you can take without losing progress. Gentle nudges, not guilt trips.
Past the deadline? AI re-breaks the task into a smaller first step, adjusted for the time and energy you have now.
Privacy is not a vibe. For ADHD adults it is a hard constraint that shapes whether the app is usable at all.
77% of ADHD adults rate privacy "very important" or "mandatory" in recent survey work on productivity-AI adoption (cf. arXiv 2603.17258). We built for that 77%, not the other 23%.
Works on a plane, a subway, a bad hotel wi-fi. Every feature, including AI breakdown, runs without a network round trip.
Cycle, sleep, medication context is local. The app can personalize suggestions based on your patterns without any of that context ever being transmitted.
No API costs means no rate limits. Personalization is continuous, not budgeted. You will not be asked to ration your own suggestions.
HIPAA-exempt by design. Medical-adjacent data never crosses a server, so it is not in a covered exchange to begin with.
Open model, auditable stack. Qwen 2.5 1.5B under Apache 2.0, running on llama.cpp. No vendor lock-in, no silent model swaps.
Free is genuinely useful. Pro pays for the on-device AI that does the expensive work, not for the features you should already have.
Genuinely useful without any trial clock.
or $59.99 / year, save 44%.
30-day free trial. Cancel anytime.
Founder's pricing, first 500 buyers only.
Just launched. Limited time.
Pricing and IAP products are live on the App Store. See pricing details →
These are the questions real users ask. If yours is not here, email [email protected].
Open the app, tap any AI feature, get 30 days of full Pro access. Cancel anytime in Settings. No charge until the trial ends.
ADHD brains dislike subscriptions. If you want to own KickMint forever with no monthly friction, Lifetime is for you. First 500 buyers get Founder's pricing at $89.99 instead of $129.99.
Yes. Every feature works without a network connection, including AI task breakdown. The only time the app uses the network is a one-time ~1GB AI model download on first launch and optional encrypted cross-device sync.
No. Tasks, voice transcripts, AI prompts, AI outputs, cycle, sleep, and medication data all stay on your device. Optional sync uses AES-256-GCM encryption with keys stored only on your device, so our server sees ciphertext it cannot read.
Qwen 2.5 1.5B Instruct in GGUF format, quantized to Q4_K_M, served by llama.cpp. The model file is roughly 1GB and Apache 2.0 licensed.
iPhone 14 and newer for full on-device AI features. iPhone 11 through iPhone 13 run the rule-based fallback for task breakdown with all other features available.
A single task breakdown takes a few seconds and under 1% of typical battery. The app caps background inference and batches work with iOS efficiency cores so continuous use stays within normal productivity-app battery draw.
No. Voice capture uses the iOS on-device speech recognizer. The raw audio buffer is discarded as soon as it is transcribed. Only the resulting text is kept, and only on your device.
The app still works. Task capture, voice capture, the inbox, rule-based breakdown, streaks, and basic insights all run without the AI model. AI breakdown is the feature that requires the model.
Goblin Tools and Todoist AI send your task text to a third-party LLM API. KickMint runs inference entirely on your phone using a model we bundle. That means no API costs, no rate limits, no data leaving your device, and offline functionality.
KickMint launches on the App Store soon. Put your email on the waitlist and we will ping you the day it ships. We only use the address for launch notification. No marketing, no drip, no resale.
We do not use forms on this page because forms usually mean cookies and trackers. A plain email does the job and keeps the site at zero third-party scripts.