Jake
26c7307607
Phase 9 of the rebrand cascade. Sweep covers everything the Phase 8
frontend pass deliberately skipped: docs/, root markdown, scripts,
Cargo.toml descriptions, code comments that survived earlier
word-boundary sed, plus a handful of identifiers caught on the final
verify pass.
transcription-app changes:
- README.md, HANDOVER.md, KNOWN-ISSUES.md, run.sh — magnotia/Magnotia
-> lumotia/Lumotia.
- docs/ — sweep across all subdirs except docs/handovers/ (preserved
as immutable audit trail). Includes architecture-map references
to magnotia_core::*, magnotia_storage::*, etc. now pointing at
lumotia_*; dev-setup.md tracing output examples (lumotia_startup
target); brief/ + superpowers/ + issues/ + whisper-ecosystem/ +
audit/.
- Cargo.toml descriptions on 9 crates (core, audio, cloud-providers,
hotkey, llm, mcp, plus referenced others).
- crates/core/src/{error,hardware,recommendation,paths}.rs +
crates/audio/src/wav.rs + crates/llm/src/model_manager.rs +
crates/cloud-providers/src/keystore.rs + crates/mcp/src/lib.rs —
doc comments and a model-manager user-agent string.
- Caught on final pass: BroadcastChannel("magnotia_task_sync") -> ...
("lumotia_task_sync"); magnotia_locale i18n localStorage key
renamed + migration shim added; CSS keyframe names
magnotiaPulse / magnotiaBar / magnotiaFade renamed in the design-
system kit; magnotia_viewer_item / magnotia_viewer_mode handoff
keys renamed in HistoryPage + viewer/+page.svelte; src/assets/
wordmark.svg text.
- src-tauri/src/lib.rs comment cleanup ("magnotia era" was sed'd
to "lumotia era" earlier — restored).
Preserved (intentional):
- crates/core/src/paths.rs — keeps "magnotia" / "Magnotia" / ".magnotia"
legacy detection strings in legacy_and_target_paths() so the
migration shim can still find user data from the magnotia era.
- src/lib/stores/{page,focusTimer}.svelte.ts + src/lib/i18n/index.ts
— migration call sites reference the legacy magnotia keys
deliberately.
- docs/handovers/ — historical audit trail.
cargo build --workspace passes. npm run check: 0 errors / 0 warnings
(3958 files). cargo test --workspace: 339 pass / 0 fail.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
7.1 KiB
7.1 KiB
4. Feature Set
Core MVP (shipping with beta)
- Local AI transcription (Whisper, on-device)
- Auto-populating to-do lists from transcriptions
- Visual time representation. Tasks displayed as visual blocks of time or countdowns, not just text lists. Traditional text-based to-do lists trigger overwhelm — visual timelines directly combat time blindness. This is the #1 community-requested feature and Tiimo's primary strength. Lumotia must match or exceed it from day one. Time should be externalised using visual countdown timers (e.g. shrinking colour disks, filling progress rings) rather than standard digital clocks — making the passage of time concrete and anchoring focus for users with time agnosia.
- WIP limits. The main screen must mathematically restrict how many active tasks are visible at once. A "Now" column showing only 1–3 items maximum. Auto-generated task lists that dump 30 items onto a screen will instantly trigger the freeze response. The AI can prioritise; the UI must constrain.
- History of past voice notes and transcriptions
- Light/dark mode
- Templates with local AI agent (contextual text under headings with associated metadata)
- Vocabulary profiles (custom dictionaries for specialist terms — e.g. DND NPC/location names, technical jargon)
- Transcription of uploaded voice notes and media files
- Open data format. All transcripts and task lists stored locally in plain text, JSON, or Markdown. Essential for the privacy-first and PKM audience. Enables the Lumotia → Obsidian workflow promised in the distribution strategy. Users must be able to export, move, and own their data without vendor lock-in.
Post-MVP features (validated, designed, not yet prioritised)
- AI-powered micro-stepping with "just start" timer. Decomposing abstract goals into hyper-specific actionable steps. The local AI agent must generate micro-steps that begin with highly specific, low-friction action verbs. Linguistic rules: every generated step must start with a concrete physical verb, target one single action, and be completable in under 5 minutes. Example: "Clean room" → "Pick up one shirt from the floor." NOT "Organise your bedroom" (still abstract, still paralysing). The goal is to bypass executive dysfunction by removing all ambiguity about what "starting" means. Paired with a 2-minute or 5-minute "just start" focus timer. Committing to a task for just five minutes bypasses internal resistance and builds micro-momentum — users frequently work past the timer. The timer should be a single tap from any micro-step, visually prominent, and use a shrinking colour disk or similar visual countdown (not a digital clock) to externalise the passage of time and combat time blindness.
- Implementation intentions / if-then automation ("If 9am and at desk, then start project X")
- Forgiving gamification (non-punitive progress indicators, no streak-shaming, grace days)
- Soft-touch nudging system ("Margot" protocol). Reminders must not function as standard push notifications (anxiety-inducing noise). Instead, design as "anticipatory guidance" — context-aware interventions that respond to behavioural signals (e.g. inactivity, time of day, task proximity) rather than rigid schedules. Tone must invite the user back without inducing guilt: "Your list is still here when you're ready" not "You missed your 2pm task!" Rhythmic voice anchoring: Case studies on custom ADHD AI coworkers (the "Margot" project) show users don't need complex avatars — they need rhythm and presence. Simple intermittent voice prompts (calm voice stating "Hey, time to move on" when a timer ends) reduce default-mode network activity, anchoring focus and restoring temporal structure without visual clutter. Delivery mechanisms: ambient visual cue within the app, OS-native notification via tauri-plugin-notification (platform-specific sounds: 'Glass' on macOS, 'message-new-instant' on Linux, 'Default' on Windows), discreet haptic nudge on mobile (Web Vibration API on Android). Context-aware suppression: no nudge if user typed within last 5 seconds or is actively speaking (detected via AudioContext analyser). All notifications fully customisable or disableable.
- Human-in-the-loop feedback. Users must be able to easily correct, rate, or override the AI's task organisation and micro-stepping output. ADHD manifestations vary wildly between individuals — the system must adapt to individual cognitive rhythms over time rather than remaining static. Simple thumbs up/down on AI-generated steps, plus ability to edit and retrain. This feedback loop is essential for the AI to improve and for users to feel ownership, not dictation.
- Start/shutdown rituals (transition scaffolding). ADHD brains struggle immensely with transitions — starting work and turning "off" at the end of the day. Implement guided rituals: a 2-minute morning triage (AI surfaces yesterday's incomplete tasks, user picks 1–3 realistic goals for today) and an evening shutdown sequence (review what was done, close mental open loops, consciously separate work from rest). Borrowed from Sunsama's proven model but adapted for neurodivergent users — must be optional, gentle, and never guilt-inducing if skipped.
- Energy-aware task sequencing. Allow users to tag transcription dumps or tasks with an energy level (High / Medium / Brain-Dead). The AI surfaces low-friction, easy tasks when the user is in an afternoon energy dip, and reserves high-cognitive-load tasks for peak energy windows. This replaces temptation bundling (which was cut due to OS limitations) with a less invasive mechanism that achieves the same goal: getting low-dopamine tasks done by matching them to the right moment.
- Read Page Aloud (text-to-speech). A simple TTS function that reads transcriptions, task lists, or AI-generated micro-steps aloud. Engages auditory processing alongside visual, which improves retention and comprehension for ADHD users. Particularly valuable during the "Clarify" stage when reviewing a brain dump. Use OS-native TTS engines (available on all target platforms) to avoid additional dependencies. Should be a single-tap action from any text view.
Parked / future consideration
- AI body doubling (low-fi implementation). Research strongly validates the concept (rated #1 ADHD workplace strategy in 2025 ADDitude survey; 12-week study showed focus doubling, 30% anxiety reduction, £37 public value per £1 invested). Body doubling doesn't require high-fidelity interaction — simple ambient presence and shared monitoring work. A "low-fi" version could be a "Focus Room" interface showing abstract statuses ("AI is sorting your tasks…", "3 other Lumotia users are in deep work right now") to provide the feeling of parallel presence without complex engineering. This sidesteps the need for video, voice, or real-time communication. Potential future subscription feature. Not in MVP scope but worth prototyping early — the implementation cost is low relative to the validated demand.
- Temptation bundling — cut (OS-level integration nightmare across platforms, essentially impossible on iOS). Replaced by energy-aware task sequencing (see post-MVP features).