Create ChatGPT plugins from Python code
Project description
AutoPlugin
AutoPlugin is a Python package that makes it easy to convert Python functions into ChatGPT plugins. With just a couple lines of code, you can:
- Automatically create an OpenAPI spec with custom endpoints for your registered Python functions, telling ChatGPT how to use it. Pull endpoint descriptions from the function docstring or generate them automatically with the OpenAI API.
- Generate the
ai-plugin.json
file to register your plugin with ChatGPT. - Launch a local server that can be used by ChatGPT for development.
Installation
To install AutoPlugin, simply run the following command:
pip install autoplugin
To install with the ability to generate endpoint descriptions for the OpenAPI specification automatically from source code, install with
pip install 'autoplugin[gen]'
Basic Usage
To get started with AutoPlugin, follow these steps:
- Import the necessary functions from AutoPlugin:
from autoplugin import register, generate, launch, get_app
- Create an app instance, backed by FastAPI:
app = get_app()
- Use the register decorator to register your functions as API endpoints. AutoPlugin will automatically generate descriptions if needed.
@register(app, methods=["GET"])
async def get_order(name: str) -> str:
order = await get_order_from_db(name)
return f"Order for {name}: {order}"
# Generated description: "Retrieves an order from the database for a given name."
- Generate the necessary files (
openapi.yaml
andai-plugin.json
) for your ChatGPT plugin. Optionally, specifyout_dir
to change where they're saved to:
# generated files saved to `.well-known/` directory
generate(app, name="Example", description="Plugin to add numbers or greet users")
- Launch the server. Optionally, specify
host
andport
:
launch(app) # API hosted at localhost:8000
- Follow the instructions to run a custom plugin:
- On ChatGPT, make a new chat.
- Under "Models" select "Plugins"
- In the Plugins dropdown, select "Plugin store"
- Click "Develop your own plugin"
- Enter the URL you're running the server at ("localhost:8000" by default) and hit enter.
- Click "Install localhost plugin"
Example
Here's a complete example that demonstrates how to use AutoPlugin to create API endpoints for two functions, hello
and add
.
It also generates the openapi.yaml
and ai-plugin.json
files, by default in the .well-known
directory. :
from autoplugin.autoplugin import register, generate, launch, get_app
app = get_app()
@register(app, methods=["GET", "POST"])
async def hello(name: str, age: int = 5) -> str:
return f"Hello, {name}! Age {age}."
@register(app, methods=["GET"])
async def add(a: int, b: int) -> int:
""" Adds two numbers """
return a + b
# Generate the necessary files
generate(app, name="Example", description="Plugin to add numbers or greet users")
# Launch the server
launch(app)
This example creates a FastAPI server with two endpoints, /hello
and /add
, that can be accessed using GET or POST requests.
AutoPlugin will use the docstring for the OpenAPI description of /add
and generate an automatic description for /hello
by passing the source code of the function to OpenAI's API.
The @register
Decorator
The @register
decorator is used as follows:
@register(app: FastAPI,
methods: List[str], # which HTTP methods to support
description: Optional[str], # if provided, used as is
generate_description: Optional[bool]) # whether to autogenerate a description
def my_func(...):
...
AutoPlugin generates function descriptions in the OpenAPI spec so that ChatGPT knows how to use your endpoints. There are a few keyword arguments to customize the behavior of this generation
By default, the description is fetched from the docstring. If there's no docstring, or if you specify generate_description=True
, AutoPlugin will generate one automatically from OpenAI's API (requires the LangChain package and setting the OPENAI_API_KEY
environment variable).
Finally, you can override the description generation behavior by specifying a description (e.g. if the docstring contains extra information not needed in the OpenAPI description) in the description
keyword argument.
The generate
Function
The generate
function has the following signature:
def generate(app: FastAPI, out_dir: str=".well-known", **kwargs):
The out_dir
keyword argument determines where the ai-plugin.json
and openapi.yaml
files are saved upon generation.
All other keyword arguments are used to customize fields of the plugin manifest file.
The name
keyword argument can be used for convenience to update both name_for_human
and name_for_model
at once. Same for description
. In a future update, these can be automatically generated to further streamline the deployment process. Keep in mind the best practices for descriptions.
Testing
AutoPlugin also provides a testing_server
utility (courtesy of florimondmanca) for testing your endpoints. Here's an example of how you can use it to test the /hello
and /add
endpoints from the example above:
from autoplugin.testing import testing_server
from os.path import join
import requests
def test_api():
host = "127.0.0.1"
port = 8000
server, base_url = testing_server(host=host, port=port, app_file="path/to/example.py", app_var="app")
with server.run_in_thread():
# Server is started. Do your tests here.
response = requests.post(join(base_url, "hello"), json={"name": "John Doe", "age": 31})
assert response.json() == {"result": "Hello, John Doe! Age 31."}
response = requests.get(join(base_url, "hello"), params={"name": "Jane Smith"})
assert response.json() == {"result": "Hello, Jane Smith! Age 5."}
response = requests.get(join(base_url, "add"), params={"a": 6, "b": 8})
assert response.json() == {"result": 14}
# Server will be stopped.
test_api()
Project details
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
Hashes for autoplugin-0.1.3-py3-none-any.whl
Algorithm | Hash digest | |
---|---|---|
SHA256 | 9a54156ad01a19927bcb9b13dc24b4ad8be0db1d8ca30eb0c5660c9bf102cde1 |
|
MD5 | 146972aa3c7fa2ddab7ba224302fd65c |
|
BLAKE2b-256 | 9ec6cd1fe6cea437b1b83894cdef68feb346ae771a9f72b7af2b0f3613f8cfee |