You don't have to choose
between losing the thought
and leaving the moment.
WristBear turns your Apple Watch into a presence-preserving capture device. Five seconds. Voice. Haptic. Back to the dinner table. The phone never comes out.
Always on. Voice note in 5 seconds. Works at dinner, in the car, mid-conversation. No structure required.
30-min mode pings. Watch shutdown button. Phone morning briefing. For when you want the full picture.
The phone is not a tool
for your brain type.
You're walking the dog. Between sets. Mid-drive. And you get a genuinely good idea — the kind that connects two things you've been thinking about for weeks. You think: I'll remember this.
You don't.
Or you fumble for the phone, unlock it, navigate to Notes or ClickUp, and by the time you're typing the thread of the idea has collapsed. Two seconds of friction is enough to lose it. Your working memory doesn't buffer — it evaporates.
ADHD is a deficit in executive function: working memory, time blindness, task initiation, task switching, distraction tolerance. The phone's interaction model — unlock, navigate, choose, type, close — requires four executive-function steps before anything gets captured.
The phone is not a tool for your brain type. It's a trap disguised as one. Every notification is a potential 20-minute hole. Every app open is a decision tax. The phone is structurally aligned against every deficit you're already fighting.
You're at the table. Kids are talking. You made the deliberate call: phone stays in the other room. You're present. This is the goal.
Then a thought fires. Something personal — "oh I need to look into that" — or something that came up in the conversation. Your brain knows it's going to evaporate.
So you have three options, all bad: let it evaporate, go get your phone and lose the moment, or interrupt the conversation to hold it in working memory until you can write it down.
Option A: Let it go. It evaporates. Costs you the thought.
Option B: Get the phone. Costs you the presence. Every time.
Option C: Hold it in working memory. ADHD working memory doesn't buffer — it drops.
Option D: Wrist up. Five seconds. Back to the conversation. Thought is safe.
Every Monday you think: I worked all weekend. On what? On Thursday you know you were "building" — but you can't reconstruct the shape of the week.
You feel busy but the compounding isn't there. You don't know if you're spending 80% of your cognitive hours on the thing that actually needs you this quarter — or drifting across five things and finishing none.
"I had a great idea in the car and I cannot for the life of me remember what it was."
"I know I worked hard this week. I have nothing to show for it and I don't know why."
"I thought I was heads-down on the thing that matters. Wasn't. Still don't know where the time went."
Capture requires the phone
Every note app, task manager, and voice recorder requires unlocking the phone. By the time you're in the app, the ADHD working memory has already moved on. The bottleneck isn't the app. It's the four steps before the app.
ADHD fails at maintenance
Streak-based systems, daily check-ins, inbox-zero rituals — all require executive function to maintain. ADHD is a deficit in exactly that. Every system that punishes you for missing a day is designed for someone else's brain.
Time blindness is neurological
You worked three hours and thought it was seven. Or worked seven and thought it was three. Time blindness isn't laziness — it's the ADHD clock running at a different speed. You cannot self-monitor what you cannot perceive.
The phone is a trap disguised as a tool.
It's not your fault you reach for it. It was designed by teams of engineers to be impossible to put down. Every notification, every app icon, every infinite scroll is structurally aligned against every deficit you're already fighting. You can't discipline your way out of a device designed to addict you.
The Watch is the only device that's on you 16 hours a day. It doesn't require an unlock. It doesn't have an app grid. It doesn't have infinite scroll. It's the only surface that can actually solve this.
Three steps.
All on your wrist.
No app to open. No unlock. No decision tax. The Watch does the work — you get the thought captured, the time tracked, and the week visible.
Raise your wrist. Speak.
A thought fires. Tap the Watch and dictate — ten words or thirty. WristBear routes it automatically: task to ClickUp, idea to Notion, note to Scratch Pad. Haptic confirms. Phone never comes out. Total interruption: under 8 seconds.
Voice → Haiku 4.5 → routed → haptic
Tap your mode every 30 minutes.
Watch buzzes. One tap: Build, Family, Foil, Lift, Drive, Talk, or Rest. One second. Optional voice if there's something worth saying. That tap is your time ledger. And critically: skipping a ping is not failure. It logs as data. A cluster of dismissals = high-focus block. The system doesn't punish you for being in flow.
Randomized within window · skips are signal, not shame
Review Sunday. Adjust Monday.
Thirty taps across the week become a map. You see that Build was only 9 hours, not the 20 it felt like. You see which project got your cognitive hours and which got none. Adjust next week against real data — not memory, not vibes.
Mode × project breakdown · open loops · capture volume
The Watch face shows three things: current mode, current project, time elapsed. That's the entire display. It answers “what am I supposed to be doing right now” without requiring a decision.
What the day actually
looks like.
Not a use-case scenario. This is the real session log from the day WristBear was first tested live.
Coffee. Watch on wrist. Phone face-down. Rich taps the WristBear complication. Mode chip shows Build · FoilBear · 0:00. That's the last decision he has to make about what he's supposed to be doing.
Deep in the sensor pipeline when a thought fires: call the accountant about Q2 before Friday. Wrist up. "Call Ben before Friday — Q2 estimates." Haptic confirm. Back in code. Task is in ClickUp. Total interruption: 8 seconds.
Watch buzzes. Thirty-minute ping. Taps: Build. Long-presses: "Still on FoilBear sensor integration, almost at a natural break." Four seconds. Mode and project logged. Back.
Switches to Statterbox prep — meeting in 20 minutes. Taps mode chip: Build · Statterbox. Raises wrist: "Pull debrief structure doc before 9:30." Task in ClickUp. No Slack open. No email check. Stays on task.
End of day. Presses the Watch shutdown button. Speaks a 30-second voice summary. The Watch mode map + voice note feed the Claude shutdown skill. Tomorrow's 3 things are ready before he sleeps.
The week that felt chaotic has a shape. The project that got the morning hours is visible. The task that fired mid-flow is in ClickUp, not lost.
What you see.
What you do.
Three things only: current mode, current project, time elapsed in this block. Answers "what am I supposed to be doing right now" without requiring a decision. No readiness score. No runway number. Those are look-once, then obsolete — they belong on the phone morning briefing.
Tap → dictate → haptic. The classifier (Haiku 4.5) routes automatically: task to ClickUp, idea to Notion Idea Bank, note to Scratch Pad. You never choose the destination — just speak.
Watch buzzes every ~30 min. One tap to select your mode — Build, Family, Foil, Drive, Lift, Talk, Rest. Done in one second. Dismissals log too; a cluster of skips means high-focus, don't lose that signal.
Mode × Project breakdown by hour across the week. Open loops not yet actioned. Capture volume by day. And one second-order metric: context-shift score — how many times you switched modes within an hour. 10 Build↔Family switches in 60 minutes isn't a scheduling problem. It's an environment problem. The map tells you which.
Honest constraint: WristBear does not tell you how to fix your time allocation. It tells you what's actually happening. The insight is yours to act on.
A voice OS
you actually own.
Watch capture is where it starts. Three input surfaces sharing one backend is where it goes. The voice layer plus an intent classifier turns every dictation into a routable action — not just text in a field.
Wispr Flow is dictation. WristBear is a voice OS.
Same starting price. Five times the product surface. Owns your data. Knows your stack. Ships actions, not just transcripts. The category doesn't exist yet — that's the opportunity.
They found each other.
Both have ADHD.
Rich and Paige both have ADHD. They found each other — as ADHD people often do. Which means the dinner table problem, the evaporating thought problem, the phone-coming-out problem — it's not one person's struggle. It's both.
She doesn't need mode pings or Sunday maps. She needs the capture. A thought fires at dinner — something personal, something the kids said, something she needs to follow up on. Five seconds, wrist up, done. She stays in the conversation.
WristBear doesn't require her to use it like Rich does. Layer 1 stands completely on its own. She installs the app, sets up the Shortcut, and that's it. No pings. No mode tracking. Just capture.
Uses Layer 1 everywhere — desk, car, dinner, gym. Uses Layer 2 for work structure: mode pings, project tracking, shutdown ritual, morning briefing. The full picture of how the day actually distributed.
The two layers don't require the same level of engagement. You can use as much or as little structure as your day demands — and that can change day to day.
When both people at the table have a way to capture without pulling out a phone, the phone stops being the default interrupt. The table stays the table. The car stays the car. The conversation stays the conversation.
This isn't a side effect of WristBear. It's the main thing.
Why another builder
would want this.
The viral hook isn't features. It's recognition.
"I built a Life OS for my Apple Watch because the phone was making my ADHD worse" is a sentence that lands for a specific person — technically capable, already wearing a Watch, already frustrated with the phone's interaction model, already suspicious that they're losing hours to drift they can't prove.
That person does not need to be sold to. They need to see it working for someone like them.
"Thought I was heads-down on FoilBear all week. Was actually on it for 9 hours. Switched projects 14 times. No wonder it felt scattered."
That's the share. Not a screenshot of a feature — a screenshot of a week made visible.
Honest constraint: WristBear is not an ADHD intervention. It works with how ADHD actually operates — it does not fix it, treat it, or diagnose it. It's a time-awareness tool for people who already have coping strategies and need external scaffolding.
The same Watch,
everywhere you go.
Mac productivity trackers capture 3-4 hours of a builder's day. WristBear is on your wrist for 16 hours. No other tool can see the full shape of the day.
Captures tasks and ideas mid-flow without breaking focus. Mode chip shows current project. Ping logs cognitive hours by project.
Voice capture hands-free, haptic confirm. No looking at phone. Ideas that fire on the commute land in ClickUp before you park.
Mode tap at the start. Voice note mid-rest if something fires. Ping skips log as lift block — no interruption, still tracked.
Watch is waterproof. Pings during a session are expected to skip. Session boundary logged by mode change on shore.
Mode = Family. Pings confirm presence — or absence. No phone pull needed for captures. The Watch doesn't become a distraction.
Watch stays silent if pings are dismissed. Clusters of skips are signal, not failure. Every context produces real data.
After 90 days, you have something no journal, no app, and no memory gives you — a real map of how your cognitive hours distributed across projects, contexts, and life domains over an entire quarter. You can see the drift before it becomes a wasted season.
What works right now.
Not promises. Production.
Everything below is shipped, deployed, and in daily use. Pipeline metrics are tracked per capture in Supabase.
Capture surfaces
LiveThree places to start a dictation. All hit the same cleanup pipeline.
- Mac menubar app — hold ⇧⌘X, speak, release, transcript pastes at cursor
- iOS Shortcut — Dictate Text → POST /capture/text → haptic confirm. Pinnable on Watch face
- iOS Shortcut (audio) — Record Audio (tap-to-stop) → POST /capture/audio for rambles
- CarPlay pattern — Watch records → cleaned text → Universal Clipboard → paste on iPhone later
Cleanup pipeline
LiveWhisper → 4 layers → text. Each layer is editable as a config file, not vendor-locked.
- Layer A — shortcode expansion: 'btw' → 'by the way', 'fb' → 'FoilBear', regex pass before LLM
- Layer B+C — Haiku 4.5 cleanup: verbal edit commands, disfluency removal, punctuation, capitalization
- Layer D — per-app profiles: Slack, Mail, VSCode, ClickUp, iMessage, Notion get different style/format
- Per-app bundle ID detection — captured at hotkey-down before WristBear takes focus
- Graceful degradation — Haiku failure returns raw transcript with degraded=true, never blocks paste
- Prompt-injection hardened — transcripts wrapped in <transcript> XML tags
Verbal commands while you talk
LiveChange your mind mid-sentence. Apply formatting with your voice.
- "scratch that" / "actually no" / "actually make that" / "delete that" — removes preceding clause and replaces with what follows
- "scratch the last sentence" / "delete the last sentence" — removes the prior full sentence
- "new line" — inserts a newline
- "new paragraph" — inserts a paragraph break
- "all caps X end caps" — uppercases X. Variants: "in all caps", "make this all caps"
- "in quotes X end quote" — wraps X in double quotes
Learning
LiveTeach the system on the fly. No prompt edits, no redeploys.
- Teach-by-voice — say "WristBear, when I say X and I'm talking about Y, mean Z". Stored in Supabase, applied to all future captures
- Context guards — "lyft" → "Lift Foils" only when talking about water/foiling/knots; "lyft to the airport" stays Lyft
- Vocabulary biasing — 54-term personal dictionary injected into both Whisper (ASR) and Haiku (cleanup preserve list)
- Voice framework synced to 8K iMessage corpus — per-recipient register matrix, signature phrases, anti-patterns
Cross-device history
LiveEvery transcript is durable. Pull the last 10 from any surface, any device.
- Mac menubar shows last 10 dictations — each individually copyable
- Server endpoint GET /captures/recent — seeds the menu on launch from Supabase
- Global across surfaces — Mac, iOS Shortcut, Watch all share one history feed
- Supabase wristbear_captures is source of truth — every capture logged with full pipeline metadata
Kanban + laptop board
LiveTriage captures on the watch, iPhone, or any browser — same lifeos_captures, three surfaces.
- iPhone Kanban — group by Status or App, drag-reorder lanes, P0 red-dot, swipe Done/Push
- Life filter — show only non-project captures (Turo, school, appointments), separated from Bear-app dev work
- Tappable email links — email-triage cards deep-link back to the source Gmail message via metadata.gmail_url
- Web board at /rich/board — laptop-accessible queue behind the PepStar login, grouped Now / This Week / Next / Later, one-click ✉️ email + 🔗 action links
- Done / Push actions on the web board via the lifeos_update_status RPC (user JWT, RLS-scoped)
Speech-to-text
LiveBest-of-both providers. Whichever is up, wins.
- Groq Whisper large-v3-turbo — primary, ~700ms typical, $0.002/min
- OpenAI Whisper — fallback on any Groq error, ~800ms typical, $0.006/min
- Same vocabulary prompt biases both — your terms get spelled right regardless of provider
- Provider field logged per capture for visibility
Infrastructure
LiveEverything you own. Nothing you rent that can't be swapped.
- Cloudflare Worker — wristbear.rpeplin.workers.dev. Stateless. Sub-second cold start.
- Supabase catchall — Pro tier, daily auto-backup + PITR, RLS on learned-vocab table
- Bearer-token auth — rotatable via wrangler secret. v1 plan: per-surface tokens for individual revocation
- Cost tracking — input/output tokens logged per cleanup, MTD vs Wispr $15/mo benchmark via wristbear-costs CLI
From capture to a life that runs itself.
One layer at a time, observed before built.
The capture layer is live. Everything above it gets observed first — real usage shapes the design before code touches it. Especially the ping layer, which could be genuinely useful or genuinely annoying. Won't know until you wear it.
Layer 0 — Capture (shipped)
LiveWhat's working today. The voice-on-wrist piece. Already in your daily flow.
- Fn-hold on Mac → cleaned text on clipboard
- Action Button on iPhone → Shortcut → cleaned text on clipboard
- Apple Watch tap-to-record → cleaned text on Universal Clipboard
- Last-10 history view, cross-device, per-user isolated
Layer 1 — Voice routing (next, observation phase)
ObservingDon't build the router yet. Start prefacing dictations with intent words ('work log', 'idea', 'task', 'note') and see what patterns emerge from real usage. After ~2 weeks of natural use, scan the captures, identify which prefixes are sticky, build routing only for those.
- WristBear, work log: 'Spent 2 hours on the FoilBear coaching pipeline.' → routes to Notion Work Log DB
- WristBear, idea: 'What if the watch sent a haptic at every hour?' → routes to Notion Idea Bank
- WristBear, task: 'Email Nick by Friday.' → routes to ClickUp Inbox
- WristBear, note: 'Paige's gym session was 45 min.' → stays in WristBear, taggable
- Auto-routing only after data shows what prefixes you actually use
Layer 2 — Mode pings (optional, opt-in, possibly insane)
Exploratory30-min Watch haptic asking 'what mode are you in?' Tap a chip: Build, Family, Foil, Drive, Lift, Talk, Rest, Other. Plus optional Project tag. Maps your day in 5-second taps. Honest caveat: the cadence may drive you nuts. Make it killable in one tap, with adjustable interval (15/30/60/off).
- Buzz at 30 min (randomized within window — standard ESM method)
- Tap a chip → 1 second cost. Dismiss → 0 seconds, logged as 'skipped'
- Long-press → voice note ('what just changed')
- Sunday map: time-per-mode × project, plus context-shift score (mode switches/hr — environment signal)
- Total daily cost ≈ 4 minutes if you tap every ping
Layer 3 — Shutdown ritual + morning briefing
ExploratoryEnd-of-day Watch button: 30-second voice summary. Routes to gbrain. Next morning iPhone shows: last night's note + today's 3 things, derived from the day's mode map + open-loop captures + calendar. Standup-ready in 60 sec.
- Shutdown button on Watch → 'how did today actually go?'
- Voice summary writes to Supabase + indexes in gbrain
- Morning briefing surfaces uncompleted captures from yesterday
- Integrates with Claude /shutdown skill so terminal sessions + WristBear share one timeline
Capture works. Beyond that, every layer is a hypothesis. Voice routing only ships after the prefix data tells us what we actually do. Mode pings only ship if the wearer wants them — and ship killable. Shutdown ritual only ships once the morning briefing has real signal to surface. Build less, observe more, ship the layer when it's earned.
Roadmap.
Planned, not promised.
v0 ships first — proving the capture loop before building the native app. Everything below depends on the 14-day gate passing.
- Native watchOS app with Watch face complication (mode + time elapsed)
- Built-in ping system with mode chip picker on Watch
- Voice capture native to Watch — no iOS Shortcut dependency
- Shutdown button on Watch — manual trigger, fires when YOUR day ends (never a timer)
- Sunday report generated automatically (Supabase → notification)
- Phone morning briefing: last night's shutdown note + today's 3 things, standup-ready in 60 sec
- Shutdown integrates with Claude shutdown skill — Watch mode map + voice summary + Claude sessions = full day picture
- Open-loop tracker: captured thoughts not yet actioned surface in the morning briefing
- Context-shift score on Sunday map: mode switches per hour reveal environment problems, not just schedule gaps — 10 Build↔Family switches in an hour means the environment broke focus, not the calendar
- Mode × Energy: cross-reference WHOOP readiness against mode breakdown
- Multi-user: both people in a household use Layer 1 independently, no shared account needed
- Capture routing preferences — teach the classifier your personal patterns
- iOS widget: live mode on the home screen for desk reference
Before building v1, the v0 Shortcut must pass: ≥3 captures/day average, ≥40% from phone-unreachable scenarios, ≥80% Haiku classification accuracy. If the gate fails — two dev days saved, one real lesson learned.
Under the hood.
One level down.
Proves the capture loop works before building anything custom. Ships in 4-6 hours.
Built after the v0 gate passes. Same Cloudflare Worker backend.
v0 uses a zero-shot Haiku 4.5 prompt. Expected ~80% accuracy on cold start. Misclassified captures still land somewhere — worst case, an idea goes to Scratch Pad instead of Idea Bank. Nothing is lost. Accuracy improves as routing data accumulates.
Every voice capture — raw transcript, classified type, routed destination, mode at capture, timestamp. Plus: classified_correct (bool, nullable) + corrected_type (text, nullable) — marked during Sunday review to compute gate accuracy
Every 30-min ping — mode tapped, project tagged, timestamp, skipped boolean
Mode sessions — start/end, mode, project, duration
Five ADHD builder types.
All of them.
Most builders are some combination of all five. WristBear's data layer surfaces which one is running on any given day.
"Build" alone is too coarse. The Mode × Project two-axis system captures that FoilBear and Audiobook Empire both counted as Build but are completely different cognitive states. Sunday map shows project distribution, not just mode totals.
A ping at :23 that logs "Family" when the builder thought he was in "Build" is the data point. Not a judgment — a fact. Over time, drift patterns become visible: Tuesdays after 4pm always drift, regardless of intent.
5-second voice capture is the only friction that works. Voice → haptic → done. No app to open, no note to navigate to, no context switch. The idea lands in Notion without breaking flow. Never lost to the 30-second window.
The ping system is time-blindness scaffolding — an external clock that doesn't require checking a clock. The Sunday report is the first objective evidence of where hours actually went. Not memory. Not perception. Data.
Long consecutive ping dismissals log as a block. WristBear learns — manually, via Sunday review — that skip clusters are flow, not broken habit. The builder labels them retroactively. The system honors the flow while still capturing it.
The cost of
not changing anything.
The idea you lost last Tuesday was worth something. You don't know what because you don't have it. Over a year of ADHD idea-evaporation, the cumulative loss is real — product insights, connection moments, financial reminders that became expensive because they surfaced two weeks late.
You cannot improve what you cannot see. Every productivity system Rich has tried assumes he can self-monitor. He cannot — ADHD makes self-monitoring unreliable by definition. Phone-based systems require executive function to maintain them. WristBear is the first system designed around not having that in the first place.
The Watch is already there. Already paid for. Already on your wrist 16 hours a day. Already building apps on it. Not using it as your operating system isn't a neutral choice — it's leaving the most contextual, always-on sensor you own running at 10% of its potential.
Estimated cognitive hours lost to distraction, task-switching debt, and evaporated ideas in a typical ADHD builder week. Not measurable until you have a baseline.
Average error between perceived and actual time spent on a project in a given day. You think you worked seven. You worked four. Or vice versa.
Hours per day the Watch is on your wrist. Hours per day you currently use it as an operating system for your brain: approximately zero.
Feeling busy and being productive are different. Without a time ledger, the feeling wins. Six months from now, you could look back at Q3 and know exactly what got 80% of your cognitive hours — or you could guess, like you do now.
FAQ.
Hardest first.
Not a product yet.
Here's what comes next.
WristBear is in v0 development — proving the capture loop works before building the native app. The 14-day gate has to pass first.
Proves the capture habit is real and sticky
Proves the Watch solves the problem phones can't reach
Tasks to ClickUp, ideas to Notion, notes to Scratch Pad
Build v1 with confidence — native watchOS app, Watch face complication, ping system, shutdown button. Classifier-tuning data already in hand.
Two dev days saved and one real lesson learned. The v0 data tells you exactly why it didn't work — no guessing.