cli

Fly on the Wall (name is subject to change)

A flexible CLI tool that records meetings from your computer's audio, transcribes speech using Whisper, and generates summaries using a LLM via Ollama. The tool emphasizes the local-first approach with AI.

This records and summarizes meetings at FartLabs, allowing participants to focus on contributing to discussions without wasting mental effort to take effective meeting notes.

Demo

TBA

Quick Start

NOTE: "uv run" can be substituted for "python" in case you do not want to use uv for package management.

# List available audio devices
uv run recorder.py --list-devices

# Start recording (press Ctrl+C to stop)
uv run recorder.py

# Record from a specific device
uv run recorder.py --device 2

Commands & Options

Option	Description
--list-devices	List available audio input devices
--device <index>	Specify audio input device by index
--whisper-model <model>	Whisper model to use (tiny, base, small, medium, large)
--ollama-model <model>	Ollama model for summarization (default: llama3)
--no-summary	Skip LLM summarization, output raw transcription only
--participants "A,B,C"	Comma-separated list of participant names
--output-dir <dir>	Directory to save meeting notes (default: notes)
--keep-audio	Keep the recorded audio file after processing

Examples

# Use a more accurate Whisper model
uv run recorder.py --whisper-model medium

# Skip summarization (transcription only)
python run recorder.py --no-summary

# Specify participants for better meeting notes
python run recorder.py --participants "Alice,Bob,Charlie"

# Keep the audio file for reference
python run recorder.py --keep-audio

# Use a different Ollama model
python run recorder.py --ollama-model mistral

Technology

This takes the local-first approach for using AI.

1. sounddevice for capturing system audio
2. OpenAI's Whisper model for transcribing speech to text
3. llama3 via Ollama for summarizing the text

TODO

• Add pause/resume functionality
• Support for system audio loopback (capture all audio, not just mic)
• Better speaker diarization
• Support multiple languages
• Real-time transcription mode
• Export to different formats (markdown, PDF)

Set up Development

Prerequisites

Set up the technologies below:

1. Python 3.12+
2. uv
3. ffmpeg
4. Ollama

Setup

After setting up prerequisites, check if Ollama is running at http://localhost:11434.

Once it is running, install llama3 (or any LLM of your choice):

ollama pull llama3 

Then, clone this repository.

Create a .env based on .env.example using your settings (optional):

# .env
OLLAMA_API_URL=http://localhost:11434

Finally, run the recorder:

uv run recorder.py

Tools

Run the below to format code with ruff:

uv run ruff format .