iOS app that records meetings, transcribes audio, and generates structured AI notes — powered by local LLMs running entirely on your device. Runs Qwen 2.5 1.5B and Phi-2 on-device via GGUF inference. No cloud, no accounts, no data leaves the phone.

Tap to start a recording from the Home screen. The recorder shows a live amplitude visualizer and elapsed timer. Pause and resume at any point, or discard the session entirely. Meetings are auto-titled by the AI after processing.

During recording, open the transcript overlay to see speech-to-text updating in real time. Transcription runs on-device via Apple's SFSpeechRecognizer.

After saving a recording it enters the queue. The queue runs transcription then LLM summarization in sequence, in the background, even when the screen is closed. Progress is visible per item with retry and cancel controls.

A searchable, sortable list of all completed meetings. Supports filtering by keyword across titles, topics, and summaries.

Each meeting shows:

• AI-generated title and summary
• Action items (with assignees where detected)
• Key decisions
• Open questions
• Topics covered
• Full transcript
• Audio playback of the original recording

From the menu you can copy the transcript, copy the full note as plain text, re-run summarization, or delete the meeting.

Download and manage local GGUF summarization models:

Models are downloaded once and stored on-device. You can switch the active model or remove installed models at any time.

• Swift + SwiftUI (iOS 17+)
• SFSpeechRecognizer — on-device speech recognition (Apple Speech framework)
• llama.swift — on-device GGUF model inference
• AVAudioEngine — microphone capture and chunk rotation

Feature-first MVVM with a strict four-layer hierarchy:

SwiftUI View → ViewModel → AppStore → Services

• View — layout and bindings only, no business logic
• ViewModel — UI-ready @Published state and named action methods per screen
• AppStore — single source of truth, owns the recording lifecycle and queue
• Services — pure and injectable, no knowledge of AppStore or ViewModels

All processing (transcription, LLM inference, file I/O) runs off the main thread. No data ever leaves the device.

# Build for simulator
xcodebuild build \
  -project Debrief.xcodeproj \
  -scheme Debrief \
  -destination 'platform=iOS Simulator,name=iPhone 16'

# Run tests
xcodebuild test \
  -project Debrief.xcodeproj \
  -scheme Debrief \
  -destination 'platform=iOS Simulator,name=iPhone 16' \
  -only-testing:DebriefTests

# Build and run on simulator or device
./run-ios.sh sim
./run-ios.sh device

• Tests — .github/workflows/ios-tests.yml runs on every push to main and on all pull requests