OpenPlugin Toolset

The openpluginclient package allows you to outsource requests to an OpenPlugin API. This enables you to use specific plugins without managing your own OpenAI API key directly.

To get started, install the package via PyPI and use the openplugin_completion function to interact with plugins.

from openpluginclient import openplugin_completion

plugin_name = "ByByAI" # enter plugin "Namespace" as written on PLUGINS.md 
completion = openplugin_completion(
    early_access_token = "__extra__-c22a34e2-89a8-48b2-8474-c664b577526b", # (required)
    plugin_name = plugin_name, # optional
    model = "gpt-3.5-turbo-0613", # optional
    messages = [ # regular ChatGPT messages argument (required)
        {
            "role": "user",
            "content": "Show me a Best Amazon products for puzzles"
        }
    ],
    temperature = 0, # optional
    # ...other openai.ChatCompletion.create arguments
)

print(completion)

Follow these steps to develop and register a new plugin within the OpenPlugin repository:

1. Prepare your branch: Fork the repository, clone it locally, and create a new branch using the naming convention plugin/<PLUGIN_NAME> (e.g., git checkout -b plugin/todo).
2. Configure OpenAI Credentials: Generate a new OpenAI secret key and add it to your environment variables as OPENAI_API_KEY.
3. Environment Setup:
   • Create and activate a Python virtual environment.
   • Navigate to the openplugin/migrations/plugin_store directory.
   • Install dependencies: pip install -r requirements.txt.
4. Registration:
   • Run the registration script: python add_plugin.py.
   • Follow the interactive prompts provided by the script.
5. Submission:
   • Commit your changes (ensuring you exclude your virtual environment).
   • Create a pull request to merge your plugin/<PLUGIN_NAME> branch into the main branch of CakeCrusher/openplugin.

# Example branch creation
git checkout -b plugin/todo

# Environment setup
python -m venv venv
source venv/bin/activate  # or venv\Scripts\activate on Windows
cd openplugin/migrations/plugin_store
pip install -r requirements.txt

# Run the plugin addition script
python add_plugin.py

The openpluginCompletion function allows you to outsource requests to an OpenPlugin API. It accepts arguments similar to the OpenAI ChatCompletion.create method.

Required Arguments:

• earlyAccessToken: A required access token string.
• messages: An array of message objects (e.g., { role: 'user', content: '...' }).

Optional Arguments:

• pluginName: The "Namespace" of the plugin to use (refer to PLUGINS.md for valid names, e.g., "ByByAI").
• model: The model identifier (e.g., "gpt-3.5-turbo-0613").
• temperature: Sampling temperature.
• Other standard OpenAI ChatCompletion.create arguments.

import {openpluginCompletion} from 'openpluginclient'

const pluginName = "ByByAI" // enter plugin "Namespace" as written on PLUGINS.md 
const completion = await openpluginCompletion({
  earlyAccessToken: "__extra__-c22a34e2-89a8-48b2-8474-c664b577526b", // (required)
  pluginName: pluginName, // optional
  model: "gpt-3.5-turbo-0613", // optional
  messages: [ // regular ChatGPT messages argument (required)
      {
          "role": "user",
          "content": "Show me a Best Amazon products for puzzles"
      }
  ],
  temperature: 0, // optional
  // ...other openai.ChatCompletion.create arguments
})

// print as prettified JSON
console.log(completion)

The classification pipeline processes raw plugin data (e.g., from OpenAI responses) to determine their compatibility and support status within OpenPlugin. The process follows these stages:

1. Base Info Generation: Extracts metadata like namespace, auth, domain, and openapi_url from the raw plugin manifest.
2. Whitelisting/Fetch Test: Attempts to instantiate an OpenPlugin class using the plugin's root_url. If successful, the plugin is marked as whitelisted.
3. Stimulation Test: Uses an LLM (e.g., GPT-3.5) to generate a natural language prompt designed to trigger the plugin. It then calls openplugin_class.fetch_plugin() with that prompt. If the plugin responds correctly, it is marked as supported.

Plugin Statuses:

• supported: Passed the stimulation test (should work).
• tentative: Passed basic whitelisting/fetch tests but failed or did not undergo stimulation (may work).
• unsupported: Failed whitelisting, is blacklisted, or requires complex authentication that isn't handled.

# Example of the high-level classification call
composite = classify_plugin(openplugins_info, openai_res_ex)