Echoic — Task List

Phase 1: Foundation (M0 — Week 1)

T1 · Create Xcode project with SwiftUI App lifecycle, menu bar target (NSStatusItem), macOS 13 deployment target · L · blocked-by: none
T2 · Add SPM dependencies: WhisperKit, SpeakerKit, GRDB.swift, Sparkle · S · blocked-by: T1
T3 · Set up GRDB database with v1 migration (meeting, transcriptSegment, transcriptFts, summary, meetingSpeaker tables) · M · blocked-by: T2
T4 · Create Swift data models conforming to GRDB Record protocol (Meeting, TranscriptSegment, Summary, MeetingSpeaker) · M · blocked-by: T3
T5 · Add .gitignore (Xcode, Swift, macOS, CoreML models, build artifacts) · S · blocked-by: T1
T6 · Set up CI pipeline (GitHub Actions: build + test on macOS runner) · M · blocked-by: T1

T7 · Implement ScreenCaptureKit audio capture service (SCStream, excludesCurrentProcessAudio, 48kHz mono) · L · blocked-by: T1
T8 · Implement PCM ring buffer (30-second, thread-safe) with 48→16 kHz downsampling · M · blocked-by: T7
T9 · Implement AAC encoder service (AudioToolbox, 64kbps, 30-second segments to disk) · M · blocked-by: T7
T10 · Implement AVAudioEngine microphone capture on separate channel · M · blocked-by: T1
T11 · Implement meeting audio file management (segment concatenation → single .m4a on meeting end) · M · blocked-by: T9
T12 · Write audio pipeline tests (buffer management, format conversion, encoding verification) · M · blocked-by: T8, T9

T13 · Implement WhisperKit integration service (model loading, streaming transcription, 10s chunks with 2s overlap) · L · blocked-by: T2, T8
T14 · Implement transcript segment deduplication (timestamp alignment for overlapping chunks) · M · blocked-by: T13
T15 · Wire transcription output to GRDB (insert segments as they arrive, live search during recording) · M · blocked-by: T4, T14
T16 · Implement model download manager (fetch CoreML model bundles on first run, progress reporting) · M · blocked-by: T13
T17 · Implement Intel Mac detection + batch-mode fallback (detect hardware, queue audio for post-meeting transcription) · S · blocked-by: T13
T18 · Write transcription tests (known audio → expected text within WER tolerance, dedup correctness) · M · blocked-by: T14, T15

T19 · Implement SpeakerKit diarization service (post-meeting, full audio, 2–6 speakers) · L · blocked-by: T2, T11
T20 · Implement transcript-speaker merge (assign speaker ID to each transcript segment by timestamp overlap) · M · blocked-by: T19, T15
T21 · Implement local speaker identification (mic channel embedding → cluster matching → "You" label) · M · blocked-by: T10, T19
T22 · Write diarization tests (known multi-speaker audio → DER measurement, merge correctness) · M · blocked-by: T20

T23 · Implement Keychain service (store/retrieve/delete Anthropic API key via Security.framework) · M · blocked-by: T1
T24 · Implement Anthropic API client (URLSession, claude-sonnet-4-6, structured JSON response parsing) · L · blocked-by: T23
T25 · Implement retry logic (exponential backoff with jitter, 3 attempts, max 5 min) · S · blocked-by: T24
T26 · Implement summary prompt template (system prompt + transcript formatting + JSON schema) · M · blocked-by: T24
T27 · Implement long meeting chunking (>3 hours → 90-min segments → meta-summary consolidation) · M · blocked-by: T26
T28 · Wire summary to GRDB (store rawJson, extract title + executiveSummary for display) · S · blocked-by: T4, T24
T29 · Write summarization tests (API client formatting, response parsing, retry behavior, Keychain operations) · M · blocked-by: T24, T25

T30 · Implement menu bar (NSStatusItem) with idle/recording/processing icon states · M · blocked-by: T1
T31 · Implement menu bar popover (recent meetings list, Start/Stop Recording button, Settings gear) · M · blocked-by: T30, T4
T32 · Implement recording flow (Start → capture + transcribe → Stop → diarize + summarize → ready) · L · blocked-by: T7, T13, T19, T24, T30
T33 · Implement Meeting Library view (chronological list, cards with title/date/duration/preview, FTS search bar, date filter) · L · blocked-by: T4, T15
T34 · Implement Meeting Detail view (transcript with speaker colors, structured summary panel, editable speaker labels) · L · blocked-by: T33, T20
T35 · Implement audio playback (AVAudioPlayer, play/pause, scrub, transcript-segment-to-timestamp linking) · M · blocked-by: T11, T34
T36 · Implement Settings view (API key entry, model selection + download, mic toggle + device picker, storage path) · M · blocked-by: T23, T16

T37 · Implement first-run onboarding flow (Screen Recording permission explanation, Mic permission, API key validation, model download) · M · blocked-by: T36, T7
T38 · Implement meeting status state machine with error handling (recording → processing → ready / error, with user-visible messages) · M · blocked-by: T32
T39 · Performance tuning (profile RAM/CPU, lazy model loading, optimize FTS queries) · M · blocked-by: T32
T40 · Create app icon and menu bar icon assets (idle, recording, processing states) · S · blocked-by: T30
T41 · Set up code signing, notarization, and .dmg packaging · M · blocked-by: T1
T42 · Integrate Sparkle for auto-updates · S · blocked-by: T41
T43 · Add recording consent reminder feature (configurable notification to meeting participants) · S · blocked-by: T32
T44 · End-to-end testing with real meetings (10 test recordings across the audio corpus) · L · blocked-by: T32
T45 · Private beta release to 10 users, collect feedback · M · blocked-by: T41, T44