Add Discord voice channel recording access

[Peyton-Spencer]

Problem

Agents can't access Discord voice channel recordings from team calls. This prevents agents from understanding context from voice discussions and limits their ability to provide continuity across async + sync communication.

Proposed Solution

Integrate Discord voice recording capabilities so agents can:

1. Access recordings from Discord voice channels/calls
2. Transcribe audio to text for analysis
3. Extract action items and key discussion points
4. Link recordings to chat context (connect voice → text channels)

Technical Challenges

Discord Recording Options

Option 1: Discord Bot Voice Receive (Hard)

• Use Discord.js voice receive to record in real-time
• Requires bot to join voice channels
• Need audio processing pipeline (PCM → MP3/OGG)
• High complexity, resource-intensive

Option 2: Third-Party Recording Service (Easier)

• Use existing services (Craig, Mee6, etc.) that already handle recording
• Bot triggers recording start/stop via API
• Fetch recordings after call ends
• Lower complexity, may cost money

Option 3: Manual Upload Flow (Simplest)

• Users manually upload recordings to Discord chat
• Bot watches for audio file attachments
• Processes uploaded recordings automatically
• No real-time recording needed

Recommended Approach: Hybrid

1. Phase 1: Manual upload flow - get transcription working first
2. Phase 2: Integrate third-party service - automate recording
3. Phase 3: Native recording - only if needed for privacy/cost

Implementation Ideas

Discord Bot Permissions Needed

• CONNECT - Join voice channels
• SPEAK - Participate in calls (optional)
• VIEW_CHANNEL - See voice channels
• USE_VAD - Voice activity detection

Audio Processing Pipeline

1. Capture: Record audio stream (PCM/Opus from Discord)
2. Convert: Transcode to standard format (MP3/WAV)
3. Transcribe: Use Whisper API or similar (OpenAI, AssemblyAI)
4. Store: Save recording + transcript
5. Index: Make searchable for future context

Integration Points

• Add to existing Discord backend (src/discord/)
• Store recordings in shared storage (B2, S3)
• Link recordings to Discord message context
• Trigger transcription jobs automatically

Use Cases

• Call summaries: Auto-generate summaries after team calls
• Action item tracking: Extract TODOs from discussions
• Async catchup: Team members can review call context
• Agent context: Agents understand what was discussed in voice

Related Work

• Complements Google Meet recording access (sibling issue)
• Could share transcription infrastructure
• May need audio storage (Backblaze B2 already configured)

Libraries/Services to Consider

• Discord.js with voice support (@discordjs/voice)
• Prism Media for audio processing
• FFmpeg for audio conversion
• Whisper (OpenAI) for transcription
• Craig Bot as reference implementation

Privacy & Legal Considerations

• Consent: Notify participants that recording is active
• Storage: Secure audio file storage with access controls
• Retention: Define how long to keep recordings
• Compliance: Check local laws (two-party consent, etc.)

Example Flow

1. Bot joins voice channel when call starts
2. User: "@PeytonOmni record this call"
3. Bot: "🔴 Recording started"
4. [Call happens]
5. User: "@PeytonOmni stop recording"
6. Bot: "✅ Recording saved, transcribing..."
7. Bot: [Posts summary + transcript to text channel]

[Peyton-Spencer]

Feature request for Discord voice recording. Recommend the Hybrid approach starting with Phase 1 (manual upload). Get transcription working first, then explore automation. The Craig Bot approach is a good reference for Phase 2.