The ChatGPT Plugin simplifies chatbot creation by reverse engineering OpenAI's ChatGPT project. To build your own LLM chatbot, all you need is an OpenAPI specification. LLM chatbots offer powerful features out of the box such as:
- Responds and understand multiple languages
- Able to generate responses by utilising past messages
- Good understanding of user intentions
- ...
pip install chatgpt-plugin
The framework generates for you an JSON that contains:
- The endpoint that needs to be called
- The endpoint body & params that needs to be passed to fulfill the users wish
- A response message for each documented case
For the case of an TODO app:
from chatgpt_plugin.plugin import ChatPlugin
with open("todo_openapi.json") as f:
openapi_specs = f.read()
todo_plugin = ChatPlugin(openapi_specs=openapi_specs)
todo_plugin.chat(message="Add buying frozen pizza as an todo")
The plugin framework will generate following python object:
endpoint='/todos/'
endpoint_params={}
endpoint_body={'title': 'Buy frozen pizza', 'description': 'Remember to buy frozen pizza', 'completed': False}
message=[MessageResponse(response_message='Todo created successfully', response_code=200),
MessageResponse(response_message='Validation Error', response_code=422)]
You can find more examples under the examples
folder.
Here are some recommendations to make full use of this framework.
- Add a description to your query parameter to make the usage more clear
- Make the language specific e.g. if your query parameter only accept English then specify it in the description
- Add all possible responses. If your API can fail for a reason, document it.
Apache License Version 2.0, see LICENSE