mfman

mfman is a CLI tool that uses AI to clone Morgan Freeman's voice and encourages you to "vibecode" harder. It automatically generates a new, uncensored prompt based on examples, clones the voice of Morgan Freeman to read it, and plays the resulting audio.

Prerequisites

• Python >= 3.13
• uv
• mpv (for audio playback)
• A Replicate account and API token.

Setup

1. Clone the repository.
2. Set your Replicate API token:

   export REPLICATE_API_TOKEN='your_api_token_here'

   Alternatively, you can create a .env file (which is ignored by git) and add it there:

   REPLICATE_API_TOKEN=your_api_token_here
   

Usage

Run the following command to generate a new prompt and hear the encouragement from Morgan Freeman:

uv run mfman

Configuration

mfman supports user configuration via a config.toml file.

• Linux: ~/.config/mfman/config.toml
• macOS: ~/Library/Application Support/mfman/config.toml
• Windows: %AppData%\mfman\config.toml

See config.toml.example for available options, including:

• prompt_model: Change the LLM used for prompt generation.
• tts_model: Change the TTS model used for voice cloning.
• playback_command: Change the command used to play audio (e.g., ["ffplay", "-nodisp", "-autoexit"]).
• sleep_interval: Adjust the delay between API calls.

Overriding Assets

You can override the default assets by placing your own files in the assets directory within your configuration folder (e.g., ~/.config/mfman/assets/).

Supported overrides:

• prompt_examples.txt: Custom examples for the prompt generator.
• reference.wav: A different reference voice for cloning.
• transcription.txt: The transcription of your custom reference.wav.

How It Works

1. Prompt Generation: Uses an LLM (default: meta-llama-3-8b-instruct) on Replicate to create a new, uncensored "vibecoding" encouragement prompt based on examples in prompt_examples.txt.
2. Voice Cloning: Uses a TTS model (default: qwen/qwen3-tts) to clone the voice from reference.wav and read the generated prompt.
3. Playback: Saves the generated audio in the user data directory and plays it in the background (default command: mpv).

Development

• Formatting: uv run ruff format
• Linting: uv run ruff check
• Type Checking: uv run ty check
• Testing: uv run pytest (uses mocks to avoid API costs)