Skip to content

API for interacting with ChatGPT and GPT4 using Python and from Shell.

License

Notifications You must be signed in to change notification settings

frankkusters/chatgpt-wrapper

 
 

Repository files navigation

🍬ChatGPT (and GPT4) Wrapper🍬

Welcome!

What would you like to do?

ChatGPT Wrapper is an open-source unofficial Power CLI, Python API and Flask API that lets you interact programmatically with ChatGPT/GPT4.

Highlights

🤖 Programmable ChatGPT. The ChatGPT Wrapper lets you use the powerful ChatGPT/GPT4 bot in your Python scripts or on the command line, making it easy to leverage its functionality into your projects.

💬 Runs in Shell. You can call and interact with ChatGPT/GPT4 in the terminal.

💻 Supports official ChatGPT API. Make API calls directly to the OpenAI ChatGPT endpoint (all supported models accessible by your OpenAI account)

🐍 Python API. The ChatGPT Wrapper is a Python library that lets you use ChatGPT/GPT4 in your Python scripts.

🐳 Docker image. The ChatGPT Wrapper is also available as a docker image. (experimental)

🧪 Flask API. You can use the ChatGPT Wrapper as an API. (experimental)

How it works

Run an interactive CLI in the terminal:

kod

Or just get a quick response for one question:

kod(1)

See below for details on using ChatGPT as an API from Python.

Requirements

To use this repository, you need setuptools installed. You can install it using pip install setuptools. Make sure that you have the last version of pip: pip install --upgrade pip

To use the 'chatgpt-api' backend, you need a database backend (SQLite by default, any configurable in SQLAlchemy allowed).

Installation

Notes for Windows users

Most other operating systems come with SQLite (the default database choice) installed, Windows may not.

If not, you can grab the 32-bit or 64-bit DLL file from https://www.sqlite.org/download.html, then place the DLL in C:\Windows\System32 directory.

You also may need to install Python, if so grab the latest stable package from https://www.python.org/downloads/windows/ -- make sure to select the install option to Add Python to PATH.

For the /editor command to work, you'll need a command line editor installed and in your path. You can control which editor is used by setting the EDITOR environment variable to the name of the editor executable, e.g. nano or vim.

Code

From packages

Install the latest version of this software directly from github with pip:

pip install git+https://github.com/mmabrouk/chatgpt-wrapper

From source (recommended for development)

  • Install the latest version of this software directly from git:
    git clone https://github.com/mmabrouk/chatgpt-wrapper.git
  • Install the the development package:
    cd chatgpt-wrapper
    pip install -e .

Backend

The wrapper works with several differnt backends to connect to the ChatGPT models, and installation is different for each backend.

Playwright (browser-based)

  • Pros:
    • Free or paid version available (as of this writing)
    • Fairly easy to set up for non-technical users
  • Cons:
    • Slow (runs a full browser session)
    • Clunky authentication method
    • No model customizations
    • Third party controls your data

Install a browser in playwright (if you haven't already). The program will use firefox by default.

playwright install firefox

Start up the program in install mode:

chatgpt install

This opens up a browser window. Log in to ChatGPT in the browser window, walk through all the intro screens, then exit program.

1> /exit

Restart the program without the install parameter to begin using it.

chatgpt

API (REST-based)

  • Pros:
    • Fast (many operations run locally for speed)
    • Simple API authentication
    • Full model customizations
    • You control your data
  • Cons:
    • Only paid version available (as of this writing)
    • More commplex setup suitable for technical users

Grab an API key from https://platform.openai.com/account/api-keys

Export the key into your local environment:

export OPENAI_API_KEY=<API_KEY>

Windows users, see here for how to edit environment variables.

Run the program with the 'config' command:

chatgpt config

This will show all the current configuration settings, the most important ones for installation are:

  • Config dir: Where configuration files are stored
  • Current profile: (shown in the 'Profile configuration' section)
  • Config file: The configuration file current being used
  • Data dir: The data storage directory

Find the 'Config file' setting, and copy the config.sample.yaml there:

On Linux:

mkdir -p ~/.config/chatgpt-wrapper/profiles/default
cp config.sample.yaml ~/.config/chatgpt-wrapper/profiles/default/config.yaml

Then edit the settings in that file to taste. You'll want to make sure backend is set to chatgpt-api in order to use the API.

Database configuration

The API backend requires a database server to store conversation data. The wrapper leverages SQLAlchemy for this.

The simplest supported database is SQLite (which is already installed on most modern operating systems), but you can use any database that is supported by SQLAlchemy.

Check the database setting from the config command above, which will show you the currently configured connection string for a default SQLite database.

If you're happy with that setting, nothing else needs to be done -- the database will be created automatically in that location when you run the program.

Initial user creation and login

Once the database is configured, run the program with no arguments:

chatgpt

It will recognize no users have been created, and prompt you to create the first user:

  • Username: Required, no spaces or special characters
  • Email: Optional
  • Password: Optional, if not provided the user can log in without a password

Once the user is created, execute the /login command with the username:

/login [username]

Once you're logged in, you have full access to all commands.

IMPORTANT NOTE: The user authorization system from the command line is 'admin party' -- meaning every logged in user has admin privileges, including editing and deleting other users.

Configuration

From a running chatgpt instance, execute /config to view the current configuration.

The output will show the location of the configuration directory, the name of the configuration file (called a 'profile'), and the current configuration.

Configuration is optional, default values will be used if no configuration profile is provided. The default configuation settings can be seen in config.sample.yaml -- the file is commented with descriptions of the settings.

NOTE: Not all settings are available on all backends. See the example config for more information.

Command line arguments overrride custom configuration settings, which override default configuration settings.

Templates (alpha, subject to change)

The wrapper comes with a full template management system.

Templates allow storing text in template files, and quickly leveraging the contents as your user input.

Features:

  • Per-profile templates
  • Create/edit templates
  • {{ variable }} syntax substitution
  • Five different workflows for collecting variable values, editing, and running

See the various /help template commands for more information.

Template builtin variables

The wrapper exposes some builtin variables that can be used in templates:

  • {{ clipboard }} - Insert the contents of the clipboard

Tutorials:

Usage

Shell

Command line arguments

Run chatgpt --help

One-shot mode

To run the CLI in one-shot mode, simply follow the command with the prompt you want to send to ChatGPT:

chatgpt Hello World!

Interacive mode

To run the CLI in interactive mode, execute it with no additional arguments:

chatgpt

Once the interactive shell is running, you can see a list of all commands with:

/help

...or get help for a specific command with:

/help <command>

Python

To use the ChatGPT class as an API for talking to ChatGPT, create an instance of the class and use the ask method to send a message to OpenAI and receive the response. For example:

from chatgpt_wrapper import ChatGPT

bot = ChatGPT()
success, response, message = bot.ask("Hello, world!")
if success:
    print(response)
else:
    raise RuntimeError(message)

The say method takes a string argument representing the message to send to ChatGPT, and returns a string representing the response received from ChatGPT.

You may also stream the response as it comes in from ChatGPT in chunks using the ask_stream generator.

To pass custom configuration to ChatGPT, use the Config class:

from chatgpt_wrapper import ChatGPT
from chatgpt_wrapper.config import Config

config = Config()
config.set('browser.debug', True)
bot = ChatGPT(config)
success, response, message = bot.ask("Hello, world!")
if success:
    print(response)
else:
    raise RuntimeError(message)

Flask API (experimental)

  • Run python chatgpt_wrapper/gpt_api.py --port 5000 (default port is 5000) to start the server
  • Test whether it is working using python -m unittest tests/api_test.py
  • See an example of interaction with api in tests/example_api_call.py

Docker (experimental)

Build a image for testing chatgpt-wrapper with following commands.

docker-compose build && docker-compose up -d
docker exec -it chatgpt-wrapper-container /bin/bash -c "chatgpt install"

Then, visit http://localhost:6901/vnc.html with password headless and login ChatGPT.

Then, turn back to terminal and enjoy the chat!

chat

GPT4

Backend notes

Playwright (browser-based) backend

To use GPT-4 with this backend, you must have a ChatGPT-Plus subscription.

API backend

To use GPT-4 with this backend, you must have been granted access to the model in your OpenAI account.

Using GPT-4

From the shell

Follow one of the methods below to utilize GPT-4 in the shell:

Method 1: Run the command

Enter the following command in your shell:

chatgpt --model=gpt4
Method 2: Modify the config.yaml file

Update your config.yaml file to include the following line:

model: gpt4

Via Python module

To use GPT-4 within your Python code, follow the template below:

from chatgpt import ChatGPT
from chatgpt_wrapper.config import Config

config = Config()
config.set('chat.model', 'gpt4')
bot = ChatGPT(config)
success, response, message = bot.ask("Hello, world!")

Projects built with chatgpt-wrapper

Contributing

We welcome contributions to ChatGPT Wrapper! If you have an idea for a new feature or have found a bug, please open an issue on the GitHub repository.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Acknowledgments

Star History

Star History Chart

About

API for interacting with ChatGPT and GPT4 using Python and from Shell.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Python 98.4%
  • Shell 1.1%
  • Dockerfile 0.5%