Plugins

Polymind supports adding extra functions that the model can access. Included is an example plugin that allows the model to ask chatgpt questions.

Another example is DocShotgun's pubmed search plugin

Adding / Developing plugins

The format of plugins is a folder under the plugins directory, a main.py file and manifest.json. the name of the plugin should match the "module_name" under manifest.json and is the name that will be used to enable the plugin in the config.json.

The manifest.json file contains metadata about the plugin, such as the module name, name, description, and parameters. Here is an example manifest.json file:

{
    "module_name": "chatgpt",
    "name": "askchatgpt",
    "description": "This sends a message to chatgpt.",
    "params": {
        "message": "The message to send to the ChatGPT API. Should be 1:1 with the message requested by the user, for example: 'ask chatgpt if cats or dogs are better' would give a message of: 'are cats or dogs better?'"
    }
}

• module_name: The name of the plugin.
• name: The internal name of the plugin which the model will be seeing / calling.
• description: A description of what the function does, also meant for the model.
• params: The parameters that the model can include with its function call. Should include a description of what to be expected so the model can use it properly.

The main.py file should contain the following format:

def main(params, memory, infer, ip, Shared_vars):

    return f"This will be sent to the model"
if __name__ == "__main__":
    main(params, memory, infer, ip, Shared_vars)```

    
params is a dict contained any parameters from the function call, memory is polymind's context and infer is a function to do inference using the main model. [See inference.py](https://github.com/itsme2417/PolyMind/blob/main/inference.py) ip is the ip of the user who sent the request.