VMD Audio/Text control with natural language

This repository is a proof of principle for performing Molecular Dynamics analysis, in this case with the program VMD, via natural language commands.

It contains code to generate VMD commands from natural language queries given as text commands. It also contains code to control VMD via audio commands contained in a .wav recording, and via a live speak-to-text interface.

Converting text queries into VMD commands requires an OpenAI API key that allows access to the GPT-3 model. Live speak-to-text requies a google cloud speech-to-text authentication


Here is an example in text mode


  • Installation can be performed with pip. We recommend doing this inside of a virtual environment.
pip install [email protected]+git://
  • Transcription of live audio requires the additional installation of the pyaudio package (in conda, this can be achieved by conda install pyaudio).

Use of the two APIs above requires setting environment variables. In BASH, one would do

    export GOOGLE_APPLICATION_CREDENTIALS=path_to_google_api_license_file.json
    export OPENAI_API_KEY=sh-....


After installation, the commands marvis, marvis-text will be added to your path.

To convert text commands to VMD commands, use the marvis-text. For example, marvis-text --mock brings up an interactive prompt:

    $ marvis-text --mock 
    ==== MARVIS IS READY =====
    marvis-text> rotate the protein around the z axis by 45 degrees
    Mock Sending: ";rotate z by 45 1"
    (VMD Command)|rotate z by 45 1

To interface with VMD in live mode, use marvis or marvis-text. The --vmd flag can be used to specify the path to VMD, which by default is vmd.

In the simplest case, simply type

$ marvis 

The flag --safe_mode can be used to ask for confirmation of voice to text before sending the command to OpenAI

To run marvis on a Mac, where VMD is in the applications folder, one might want to create a script such as:

    /Applications/VMD\ $*

And then run:

    $ marvis --vmd ./