clio/clio-infra/TODO.md

Clio - Consolidated TODO

Architecture Overview

┌─────────────────────────────────────────────────────────────────────────┐
│                          TARGET ARCHITECTURE                            │
├─────────────────────────────────────────────────────────────────────────┤
│                                                                         │
│   ┌──────────┐        ┌──────────────────────────────────────────────┐  │
│   │  Phone   │───────▶│         Server (10.0.0.45)                   │  │
│   │ (Android)│  push  │  - Storage hub (recordings)                  │  │
│   └──────────┘        │  - clio-api (FastAPI)                        │  │
│                       │  - clio-ui (React frontend)                  │  │
│   ┌──────────┐        │  - Database (highlights, transcripts)        │  │
│   │    Pi    │───────▶│  - Forgejo (code.hallocks.xyz)               │  │
│   │(portable)│  push  └──────────────────────────────────────────────┘  │
│   └──────────┘                           │                              │
│                                          │ pulls recordings             │
│                                          ▼                              │
│                       ┌──────────────────────────────────────────────┐  │
│                       │       Desktop (10.0.0.130)                   │  │
│                       │  - GPU: RTX 3090                             │  │
│                       │  - clio-ai/transcription (Whisper)           │  │
│                       │  - clio-ai/real_time (wake word detection)   │  │
│                       │  - Off at night (non-critical)               │  │
│                       └──────────────────────────────────────────────┘  │
│                                                                         │
└─────────────────────────────────────────────────────────────────────────┘

Immediate Tasks

Infrastructure Setup

• Import Pi recorder code (systemd setup) → clio-pi/
• Migrate ~100s GB of recordings from Pi to server storage
• Pi: Add sync script to push recordings to server
• Pi: Auto-detect audio device (currently hardcoded to hw:1)
• Set up Forgejo repos for all clio-* projects
• Configure Forgejo Actions for server deployments (10.0.0.45)
• Create backup strategy for viewer, highlights, and database

Deployment Pipeline

• clio-api: Forgejo Actions → deploy to 10.0.0.45
• clio-ui: Forgejo Actions → deploy to 10.0.0.45:/var/www/recordings-viewer
• clio-ai: Manual deploy to desktop (10.0.0.130) - not via Actions

---

Feature Backlog

High Priority

• QR code for key import (avoid email/typing)
• Server URL configuration in app UI
• Manual sync trigger button
• View sync history/status in app
• Display last successful sync time on phone
• Server status page/dashboard:
  • When phone/pi last synced
  • Total recordings count
  • Recent uploads
  • Disk space usage

Wake Word / Voice Recognition (clio-ai)

• Manage wake words in UI (currently hardcoded)
• Centralize voice recognition training data
• Real-time detection improvements (currently sucks even on desktop)
• Future: Real-time detection on phone/pi (low priority)

Audio Analysis & Intelligence

• Speech detection bar in recording detail view
• Sleeping/silence detection for overnight recordings
• Auto voice recognition / speaker identification
• Auto highlight generation (audio-based, not text search)

Viewer UX

• Background audio playback (mini player across views)
• Highlight duration (currently point-only, add end time)
• Auto-scroll timeline to recordings

Medium Priority

• Configurable recording quality (bitrate, sample rate)
• Configurable segment duration (currently 2 hours)
• Delete synced files automatically (with retention)
• Server authentication for upload endpoint
• Server HTTPS with Let's Encrypt (partially done via Caddy)

Low Priority / Nice to Have

• Multiple server destinations
• Compression before encryption
• Resume interrupted uploads
• Docker container for server

---

Questions to Resolve

Key Rotation

• What happens when server keypair changes?
• Options:
  1. Keep old private key archived
  2. Sync all files before rotating
  3. Store key version in encrypted file header

Manual Decryption

• Document steps to decrypt files copied via USB
• CLI tool needed

Desktop Worker Architecture

• How does desktop pull recordings from server?
• How does it push transcriptions back?
• Queue system needed?

---

Known Limitations

1. No streaming encryption: Entire temp file must complete before encryption
2. No partial sync: If upload fails, whole file retries
3. Single server: Can only sync to one destination at a time
4. No authentication: Relies on encryption for confidentiality
5. Desktop off at night: Real-time detection unavailable

---

Repository Structure

/work/projects/clio/
├── clio-api/          # FastAPI server (recordings, highlights, transcripts)
├── clio-android/      # Android recording app
├── clio-ui/           # React frontend (viewer)
├── clio-ai/           # ML/AI components
│   ├── real_time/     # Wake word detection, voice classifier
│   └── transcription/ # Whisper transcription scripts
├── clio-pi/           # Raspberry Pi recorder
│   ├── bin/           # Recording scripts
│   ├── systemd/       # Service units
│   └── setup.sh       # Pi setup script
└── clio-infra/        # Deployment, configs, docs
    ├── nixos/         # NixOS modules
    ├── scripts/       # Deploy scripts
    └── TODO.md        # This file

Hosts

Host	IP	Role
Server	10.0.0.45	Storage, API, UI, Forgejo
Desktop	10.0.0.130	GPU transcription, real-time detection
Phone	(mobile)	Recording source
Pi	(TBD)	Portable recording source