1. Access Recording — After stopping, you’ll see it in the Recordings list.
2. Start Transcription — Tap “Generate Transcript” on the recording row, or open the audio player and tap it there. The app processes your audio using your selected transcription engine.
3. Progress — You’ll see processing status and an estimate. You can continue using the app while it processes.
4. View Results — Open the Transcripts tab to read the full transcript with editable text and confidence indicators.

1. Prerequisites — You need a transcript first.
2. Start Summary — Open the transcript (Transcripts tab) and tap “Generate Summary”. A second “Generate Summary” button is also available directly on the transcript row.
3. AI Processing — The app analyzes your transcript and creates:
   • Enhanced Summary: Main content overview
   • Action Items: Tasks with priority levels
   • Reminders: Time-sensitive items with urgency indicators
   • Alternative Titles: AI-generated recording names
4. Review Results — Expandable sections, priority indicators, interactive maps, attachments, and integration options (Apple Reminders, Apple Calendar, Google Calendar).

• “Enable iCloud Sync” — Eligible recordings, transcripts, summaries, and selected settings sync across your devices through your private iCloud account. BisonNotes shows a HIPAA notice before enabling this.
• “Keep Local Only” — Content stays on this device only. Better for privacy-sensitive content. Can be changed later.
• Per-recording exclusions — Use Keep on This Device from a recording row or audio player to keep that recording, transcript, and summary out of BisonNotes iCloud sync and backup.

Long press on any recording, or tap and use the “…” menu. Storage options:

• Archive to iCloud Drive — Copies audio to iCloud Drive, keeps transcript/summary and a saved restore location, and can optionally remove only the local audio file.
• Audio File Only — Keeps transcript/summary, removes audio. Good for saving storage.
• Everything — Removes audio, transcript, and summary. Cannot be undone.
• Summary Only — Keeps audio and transcript. Useful for regenerating with a different engine.

1. Open the Settings app on your iPhone.
2. Navigate to Action Button.
3. Select Shortcut as the Action Button function.
4. Tap “Choose a Shortcut” → Search for “Start Recording” or “BisonNotes AI” → Select it.
5. Press the Action Button to test!

What happens: The app launches, switches to the Recordings tab, and starts recording immediately. Recording continues even if you switch apps or lock your phone.

1. Select recordings — Go to Recordings tab → tap “Select” → check two recordings.
2. Tap “Combine” — A “Combine” button appears when two are selected.
3. Choose order — The app recommends chronological order. Tap the “First” card to swap if needed.
4. Review & confirm — Check combined duration preview, tap “Combine Recordings”.
5. After combining — The new recording appears in your list. Original recordings remain unchanged. Generate new transcripts and summaries for the combined file.

Tips:

• Make sure recordings are in the correct chronological order
• Combined file size = sum of both originals
• You’ll need to generate new transcripts/summaries afterward

Type: Local processing • Cost: Free • Privacy: 100% local • Internet: Only for initial model download

Requirements

• Transcription: iOS 17.0+, 4GB+ RAM. Uses Parakeet (default) for on-device transcription.
• AI Summary (MLX Swift): 4GB+ RAM. Uses Ternary Bonsai models scaled to your device.
• Legacy AI Summary (llama.cpp): 6GB+ RAM. Available as On Device AI (Legacy).
• Storage: MLX models are about 470 MB to 2.3 GB on iPhone and iPad. Native macOS also offers an optional 27B model of about 8.5 GB. Legacy llama.cpp models are about 1.3 GB to 4.5 GB; Parakeet transcription downloads separately.

MLX Models (by RAM)

• 4GB+: Ternary Bonsai 1.7B (~470 MB) — Compact local summaries for lower-memory devices
• 6GB+: Ternary Bonsai 4B (~1.1 GB) — Default model for most supported devices
• 8GB+: Ternary Bonsai 8B (~2.3 GB) — Slower but higher-quality summaries
• Native macOS, 16GB+: Ternary Bonsai 27B (~8.5 GB) — Laptop-class reasoning available only in the native Mac app

Legacy llama.cpp Models (6GB+)

• 8GB+: Gemma 3n E4B (4.5 GB) — Best overall quality
• 6GB+: Gemma 3n E2B (3.0 GB) — Good quality, smaller size
• 6GB+: Granite Micro (2.1 GB) — Very fast processing
• 8GB+: Granite (4.3 GB) — Reliable and accurate
• 6GB+: Ministral (2.1 GB) — Best for tasks/reminders
• 6GB+: Qwen3.5 2B (1.3 GB) — Thinking mode, summary only
• 8GB+: Qwen3.5 4B (2.7 GB) — Thinking mode, excellent detail

Best for: Privacy-conscious users, offline use, and local summaries without sending transcript text to a cloud provider.

Type: Cloud • Cost: Pay-per-use • Privacy: Data sent to OpenAI • Internet: Required

Setup

1. Get API key at platform.openai.com
2. Go to Setup → AI Settings → OpenAI
3. Enter your API key, select model, test connection.

Summarization Models

Transcription Models

Best for: High-quality results, advanced features

Type: Cloud • Cost: Free tier, then pay-per-use • Privacy: Data sent to Google • Internet: Required

Setup

1. Get API key at aistudio.google.com
2. Go to Setup → AI Settings → Google AI Studio
3. Enter API key, select model, test connection.

Available Models

• Gemini 3 Flash Preview (Default) — Fast and efficient
• Gemini 3.1 Flash Lite Preview — Lightweight variant for quick processing

Best for: Balanced performance and cost, with free tier for getting started

Type: OpenAI-compatible endpoint • Cost: Varies • Privacy: Depends on provider • Internet: Required unless local

Supported Providers

• LiteLLM — Base URL: your server (e.g., http://localhost:4000/v1)
• llama.cpp — Base URL: http://localhost:8080. API key: “no-key” or empty.
• Groq — Base URL: https://api.groq.com/v1. API key from console.groq.com
• Nebius — Your Nebius endpoint URL
• Custom — Any OpenAI-compatible endpoint

Setup

1. Go to Setup → AI Settings → OpenAI API Compatible
2. Enter API key and base URL for your provider.
3. Select model (use your provider’s exact model name, e.g., llama-3.1-70b-versatile for Groq).
4. Test the connection.

Best for: Using LiteLLM, llama.cpp, Nebius, Groq, or other compatible services

Type: Cloud • Cost: Free tier available (no credit card); paid tiers for higher rate limits • Internet: Required

Guided In-App Setup (Recommended)

BisonNotes AI includes a step-by-step onboarding wizard that walks you through the entire process in about 2 minutes:

1. Launch the wizard from any of these places:
   • First-time setup: Select “Mistral AI (Free)” and tap “Save & Configure”
   • AI Settings: Tap Mistral AI (look for the orange “Free” badge), then “Configure” → “Set Up Free Account”
2. Create account — The wizard opens console.mistral.ai in an in-app browser. Sign up with email and verify your phone number.
3. Generate API key — The wizard opens the API keys page. Create a key named “BisonNotes” and copy it.
4. Paste & validate — Paste your key (or use the clipboard button), then test the connection.
5. Done! — The wizard auto-configures Mistral Medium for summaries, Voxtral Mini for transcription, and enables speaker diarization.

Summarization Models

• Mistral Large (25.12) — Most capable, 128K context (Premium)
• Mistral Medium (25.08) — Balanced, 128K context (Standard) — default after guided setup
• Magistral Medium (25.09) — Economy, 40K context

Transcription

• Voxtral Mini Transcribe — $0.003/min with optional speaker diarization
• Supports MP3, MP4, M4A, WAV, FLAC, OGG, WebM
• Automatic language detection or explicit language code
• Files over 24MB automatically chunked

Manual Setup (Alternative)

1. Get API key at console.mistral.ai
2. Go to Setup → AI Settings → Mistral AI → Configure
3. Enter API key, select model, test connection.
4. For transcription: Setup → Transcription Settings → select “Mistral AI”.

Best for: Free cloud AI with fast summaries and affordable transcription with speaker diarization

Type: Cloud • Cost: Pay-per-use • Internet: Required

Available Models

• Claude 4.5 Haiku (Default) — Fast, efficient (Standard tier)
• Claude Sonnet 4.5 — Advanced reasoning and analysis (Premium tier)
• Llama 4 Maverick 17B — Meta’s latest, enhanced reasoning (Economy tier)

Setup

1. Create an AWS account and enable Bedrock service.
2. Create IAM user with Bedrock permissions and generate access keys.
3. Go to Setup → AI Settings → AWS Bedrock.
4. Enter AWS credentials, select region and model, test connection.

Best for: Enterprise features and compliance

Type: On-device • Cost: Free • Privacy: 100% local • Internet: Not required

Requirements

• iOS 26.0 or later
• iPhone 15 Pro or newer (Apple Intelligence–capable device)
• Apple Intelligence enabled in Settings → Apple Intelligence & Siri

Setup

1. Go to Setup → AI Settings.
2. Select Apple Native from the engine library.
3. No API key or model download required — uses the built-in Foundation Model.

Limitations

• Context window is 4,096 tokens. Long transcripts are automatically chunked and summarized in segments.
• Only available on devices that support Apple Intelligence.

Best for: Maximum privacy, no setup, and users with Apple Intelligence–capable devices on iOS 26+

Type: Local server • Cost: Free • Privacy: 100% local • Internet: Not required

Setup

1. Install Ollama on your machine.
2. Download models:

   ollama pull qwen3:30b
   ollama pull gpt-oss:20b
   ollama pull mistral-small3.2

3. Go to Setup → AI Settings → Ollama.
4. Set server URL and port (default: http://localhost:11434).
5. Tap refresh to scan for models, select one, test connection.

Best for: Privacy, customizable models, offline use

Type: On-device llama.cpp • Cost: Free • Privacy: 100% local • Internet: Only for initial GGUF model download

Setup

1. Go to Setup → Additional Settings → AI Engines.
2. Choose On Device AI (Legacy) if it is available on your device.
3. Download one of the listed GGUF models.
4. Optional: turn on experimental summary models to show Qwen or Granite H models that match your RAM tier.

Best for: Users who already downloaded llama.cpp models or want a different local model lineup than the default MLX Swift path.

Initial Setup

1. Enable: Setup → Transcription Settings → select “On Device” → Parakeet is selected by default.
2. Choose a model: Parakeet v2 is English-only with stronger long-form English recall. Parakeet v3 supports multilingual transcription across 25 European languages.
3. Download the model: Tap “Download” to get the selected Parakeet model. Progress now shows checking, downloading, preparing, and ready states.
4. Requirements: iOS 17.0+, model download required.

1. Install Whisper Server:

   # Using Docker (recommended)
   docker run -d -p 9000:9000 \
     -e ASR_MODEL=base \
     -e ASR_ENGINE=openai_whisper \
     onerahmet/openai-whisper-asr-webservice:latest

2. Configure: Setup → Transcription Settings → Whisper (Local Server). Set server URL and port (default: 9000). Choose protocol (REST API or Wyoming) and model size.

1. Create an AWS account and enable Transcribe service.
2. Create IAM user with transcribe:StartTranscriptionJob and transcribe:GetTranscriptionJob permissions.
3. Setup → Transcription Settings → AWS Transcribe.
4. Enter access key, secret key, region, language. Test connection.

Recording Title

You can rename a recording from three places:

• The audio player view (tap the title to edit)
• The transcript editor (tap the title at the top)
• The summary view — “Titles” section → “Edit” → enter a new title or pick one of the AI-generated alternatives → “Use This Title”.

Custom Date & Time

Open summary → “Recording Date & Time” → “Set Custom Date & Time” → use pickers → “Save”. BisonNotes AI also persists the explicit recording start timestamp so date sorting matches when the audio was captured, not when it was imported.

Location

Open summary → location section → “Add Location” or “Edit Location”. Choose: Current Location (GPS), Map Selection, or Manual Entry → “Save”.

• No Audio: Check microphone permissions in Settings → Privacy
• Poor Quality: Adjust audio quality settings, ensure proper mic distance
• Background Recording: Enable in app settings
• Mac meeting audio missing: Enable Record Meeting Audio in BisonNotes Settings → Recording, grant BisonNotes access in macOS System Settings → Privacy & Security → Screen & System Audio Recording, then quit and reopen BisonNotes. If permission or mixing fails, the app still saves microphone audio.
• Preferred Mac microphone disappeared: Reconnect the USB, Bluetooth, or interface input. BisonNotes temporarily uses the system default while it is absent and reselects the saved microphone when it returns.

• Connection Failed: Check internet and API keys
• Timeout Errors: Increase timeout settings
• Authentication Errors: Verify API credentials

• No Transcription: Check engine configuration
• Poor Quality: Try a different engine or model
• Large File Issues: Enable chunking for files over 5 minutes
• Parakeet says a model is missing: Open Transcription Settings and re-download the selected Parakeet model. BisonNotes will reuse valid cached files when they exist, but clears stale settings if the files were removed.

• Missing Recordings: Use Data Migration tools
• Corrupted Data: Clear and re-import data
• Sync Problems: Check iCloud settings

• Job Queue — Multiple operations queued and processed in order
• Progress Tracking — Real-time updates with accurate status display
• Error Recovery — Automatic retry for failures
• Stale Job Detection (v1.7) — Intelligent detection and recovery of stuck jobs, with timeouts based on actual processing start time
• Job Cancellation (v1.7) — Properly cancel and interrupt running jobs when the app is backgrounded or closed
• On Device AI Handling — Local AI jobs are gracefully interrupted when the app is backgrounded, since GPU processing requires the app to stay open
• Live Transcription (v1.8) — On-device speech-to-text streamed in real time during recording via SFSpeechRecognizer; transcript auto-saved when recording stops
• Duplicate Summary Cleanup (v1.8) — Automatically detects and removes duplicate summaries; manual cleanup available in Database Tools
• Explicit credential resolvers (v1.11) — Background AWS jobs always pull fresh credentials from the Keychain so stale process-environment values can’t leak in
• Temporary audio cleanup (v2.1) — Conservative startup cleanup removes stale scratch, merge, and chunk files left by failed or interrupted audio workflows without touching active recordings.

• Import/Export — M4A, MP3, WAV, CAF, AIFF, AIF, plus transcript documents such as TXT, MD, VTT, SRT, PDF, DOC, and DOCX
• Import From Link (v2.2) — Import direct HTTPS links for audio/video files and transcript documents; YouTube links can import public captions or fall back to pasted transcript text when captions are blocked
• Audio Export (v1.8) — Share any recording as an audio file via the iOS share sheet
• Clean Audio Export (v1.10) — Exports use a clean, user-friendly filename derived from the recording title
• Audio Archive — Offload selected or older audio files to iCloud Drive, keep an in-app restore pointer, and remove the archived cloud copy after restore
• Video Import (v1.8) — Import video files; audio is automatically extracted to M4A
• Audio Cleanup (v1.8) — Optional pre-transcription DSP: high-pass filter, noise gate, dynamic normalization, peak limiting
• Share Extensions — Import supported audio and transcript files from iPhone, iPad, and Mac Share menus with one-time token authorization. If another import is already active, staged files remain available for a later retry instead of being discarded.
• Combining Recordings — Merge two into one
• PDF Export — Professional reports with metadata, maps, page numbers
• File Relationships — Audio/transcript/summary links maintained
• Orphaned File Detection — Find and clean disconnected files
• Selective Deletion — Choose what to keep
• File Protection (v1.11) — Recordings, attachments, notes, and the Core Data store are stored with explicit iOS file protection

• Mac Recording Widget (v2.2) — Add the small or medium BisonNotes widget to the Mac desktop or Notification Center and select Record to open the app and start a new audio note.
• Apple Watch Complications (v1.9) — Surface live recording state on watch faces. Tap the complication to open the watch app.
• Control Center Recording Widget (iOS 18+) — Add the BisonNotes recording control to Control Center to start/stop recordings without opening the app.
• Watch App (v2.0) — Full standalone recording on Apple Watch with one large tap target: tap to record, tap to stop, and use mute to pause/resume the same file.
• Watch Sync (v2.0) — The watch sends the finished audio file to iPhone with WatchConnectivity. The phone imports it, avoids duplicate imports, and sends queued success/failure confirmations back to the watch.

• GPS Integration — Automatic capture
• Reverse Geocoding — Coordinates to addresses
• Smart Search — 3-tier fallback system
• University Database — Built-in campus mapping
• Interactive Maps — View recording locations
• Manual Location — Add locations after recording

• iCloud Backup — Full backup of all data to iCloud Drive
• Audio Archive to iCloud Drive — Space-saving audio offload with tracked restore locations. Third-party Files providers are paused for archive targets for now.
• Auto-Backup (v1.7) — Automatically backs up when new recordings are created
• Automatic cross-device sync — When iCloud Sync is enabled, BisonNotes reconciles eligible recordings, transcripts, and summaries when the app launches or becomes active. The Include audio files checkbox controls whether audio files are uploaded and restored.
• CloudKit Sync — Optional summary synchronization across devices with paginated queries and schema-safe fallback for reliability
• Mac, iPhone, and iPad sync — iOS, iPadOS, and native macOS use the shared iCloud container iCloud.Bison-Networking.BisonNotes-AI. Devices must use the same Apple ID and CloudKit environment. Local Debug builds use the CloudKit development environment, while TestFlight and App Store builds use production, so a Debug Mac build will not see production iPhone or iPad records until the build channel/environment matches.
• Production schema required — TestFlight and App Store builds require the CloudKit production schema to include the app-managed backup record types before sync can save new backup records. Production clients cannot create new CloudKit record types at runtime.
• Review iCloud Items — Current app versions mark synced content as active before it is automatically restored on other devices. Older cloud-only items that are not marked active are held in Settings, where they can be restored or deleted from BisonNotes iCloud sync records.
• Keep on This Device — Mark an individual recording from its recording row or audio player to exclude that recording, transcript, and summary from BisonNotes iCloud sync and backup. When enabled, BisonNotes skips future app-managed iCloud sync/backup for that item and removes known app-created iCloud records when iCloud is available.
• Deleted item cleanup — Deleting a recording writes an iCloud deletion marker and removes known app-created iCloud records. Other devices apply that marker before uploading so deleted items are not silently re-created. Active cloud-only records without a deletion marker are restored, while older untrusted cloud-only records are held for review.
• Summary-only cleanup — Deleting a summary while keeping its recording/transcript queues removal of the summary’s live record, backup record, and content-index reference. If iCloud is unavailable, BisonNotes retries the queued removal later.
• Sensitive data excluded (v1.11) — API keys and AWS credentials are never written into iCloud settings backups; if a previous backup contained them, they are restored into the Keychain instead.
• Prompted after generating your first summary

• Legacy Import — Migrate from old file-based storage
• Data Integrity — Validate and repair data relationships
• Batch Operations — Process multiple files at once
• Progress Tracking — Monitor migration progress