PRODUCTAGENTharrisonhjohnson/productagent · /harnesses/todo/bot/VOICE_SETUP.md
000%

VOICE_SETUP.md

view on github ↗112 lines · markdown

Voice Message Setup for Navi

Overview

Navi now supports voice message transcription! Send a voice message via Telegram and it will be automatically transcribed and processed as text.

Setup Options

You have two options for voice transcription:

Pros:

  • Fast and accurate
  • No local installation needed
  • Works on any machine

Cons:

  • Requires OpenAI API key
  • Small cost per transcription (a fraction of a cent per minute; see OpenAI pricing)

Setup:

  1. Get an OpenAI API key from https://platform.openai.com/api-keys
  2. Add to your environment: 
    export OPENAI_API_KEY="sk-..."
  3. Add to your ~/.zshrc or ~/.bashrc for persistence: 
    echo 'export OPENAI_API_KEY="sk-..."' >> ~/.zshrc
source ~/.zshrc

Option 2: Local Whisper (Fallback)

Pros:

  • Free
  • No API key needed
  • Works offline

Cons:

  • Requires ~1-5GB model download
  • Slower transcription
  • Requires powerful CPU/GPU

Setup:

  1. Install OpenAI Whisper: 
    pip install openai-whisper
  2. First transcription will download the model (~1GB for base model)

How It Works

  1. Send a voice message to Navi via Telegram
  2. Navi downloads the audio file
  3. Transcription happens via:
    • OpenAI API (if OPENAI_API_KEY is set)
    • Local Whisper (if installed)
  4. Transcribed text is processed like a normal text message:
    • Can be a Claude query
    • Can be a TODO command ("add to my todo...")
    • Can be any natural language input

Usage Examples

Voice TODO Addition

🎤 "Add to my TODO list: review Q1 performance metrics" → ✅ Added to TODO.md

Voice Claude Query

🎤 "What's the status of the Canada launch?" → Claude processes and responds

Voice Commands

🎤 "Show my TODO list" → Displays TODOs

Testing

To test if voice is working:

  1. Restart the bot: ./manage_bot.sh restart
  2. Send a voice message saying "Hello Navi"
  3. You should see:
    • "🎤 Transcribing voice message..."
    • Then Navi's response to "Hello Navi"

Troubleshooting

"Transcription failed: no API key or local model available"

  • Neither OpenAI API nor local Whisper is configured
  • Set up one of the two options above

"Voice messages not supported - voice handler not initialized"

  • Voice handler failed to initialize
  • Check bot logs: tail -f ~/tools/navi/bot.error.log

Slow transcription

  • Using local Whisper (no API key)
  • Consider setting up OpenAI API key for faster transcription

Cost Estimate (OpenAI API)

  • Average voice message: 10-30 seconds
  • Cost: a fraction of a cent per message
  • 100 voice messages/day: single-digit dollars per month

Privacy Note

  • OpenAI API: Audio sent to OpenAI for transcription
  • Local Whisper: Everything stays on your machine
  • Choose based on your privacy preferences!

FULL INDEX

flow-designglobal
SKILL.md74 lines
prototype-swarmglobal
SKILL.md162 lines
pm-strategistglobal
pm-strategist.md51 lines
todosystem
bot31 items
agent_executor.py344 lines
behavior_state.py106 lines
BOT.md175 lines
claude_permission_hook.py290 lines
config.py76 lines
FORMATTING_GUIDE.md308 lines
git_sync.py87 lines
hyrule_server.py284 lines
manage_bot.sh47 lines
mcp_server.py324 lines
mobile_mode.py229 lines
navi_response_hook.py202 lines
navi_tab.py95 lines
prework_engine.py426 lines
requirements.txt4 lines
run_bot.sh18 lines
run_hyrule.sh38 lines
scheduler_state.py86 lines
scheduler.py958 lines
session.py264 lines
telegram_bot.py3328 lines
todo_manager.py1130 lines
ventures_signal.py314 lines
voice_handler.py203 lines
VOICE_SETUP.md112 lines
warroom_manager.py1004 lines
warroom_questions.py228 lines
README.md44 lines
night-orderssystem
README.md84 lines
loops-spec.md212 lines
settings.json116 lines
rev-intel-harness ↗external
README.md59 lines