GPT4-powered python interpreter / AI assistant
Project description
Pandora
Pandora is an AI-powered Python console object, resulting from the combination of the latest GPT-4 Turbo model from OpenAI and a built-in interactive Python interpreter. This console allows users to execute Python commands/scripts in real time like a conventional Python console, but also allows to interact in natural language with the assistant and offers rich interactive and multimodal capabilities based on the real-time execution of AI-generated python scripts.
Main Features
- Implements a custom AI-powered python console using OpenAI GPT4 Turbo model.
- Usable both as a regular python console and/or an AI assistant.
- Capable of generating and running scripts autonomously in its own internal interpreter.
- Can be interacted with using a mix of natural language (markdown and LaTeX support) and python code.
- Having the whole session in context. Including user prompts/commands, stdout outputs, etc... (up to 128k tokens)
- Highly customizable input/output redirection and display (including hooks for TTS) for an easy and user friendly integration in any kind of application.
- Modularity with custom tools passed to the agent and loaded in the internal namespace, provided their usage is precisely described to the AI (including custom modules, classes, functions, APIs).
Powerful set of builtin tools to:
- facilitate communication with the user,
- enable AI access to data/file content or module/class/function documentation/inspection,
- files management (custom work and config folder, memory file, startup script, file upload)
- access to external data (websearch tool, webpage reading),
- notify status,
- generate images via DALL-e 3,
- persistent memory storage via an external json file.
Can also be used as an AI python function capable of generating scripts autonomously and returning any kind of processed data or python object according to a query in natural language along with some kwargs passed in the call, like so:
primes=pandora("return the list of first n prime numbers greater than m", n=5, m=15)
print(primes) # output: [17,19,23,29,31]
Can use the full range of common python packages in its scripts (provided they are installed and well known to the AI)
Installation
$ pip install pandora-ai
Usage
Using it with default setting is as minimal as:
from pandora_ai import Pandora
pandora=Pandora(openai_api_key=<your_api_key>)
# The Open API key can be ommited in the constructor if it exists as an environment variable.
pandora.interact() # enters a loop of interaction with the console-agent
Yet the Pandora class is designed to be highly configurable and easily integrated in any interface or codebase. Please refer to the full documentation or visit the Streamlit web app here to get a sense of how it can be used in a full setup.
License
This project is licensed. Please see the LICENSE file for more details.
Contributions
Contributions are welcome. Please open an issue or a pull request to suggest changes or additions.
Contact
For any questions or support requests, please contact Baptiste Ferrand at the following address: bferrand.maths@gmail.com.
Project details
Release history Release notifications | RSS feed
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
File details
Details for the file pandora_ai-0.0.23.tar.gz
.
File metadata
- Download URL: pandora_ai-0.0.23.tar.gz
- Upload date:
- Size: 33.9 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.10.12
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 776818556bf33fe4cf2e8f283b9f5a2f84b45f1a3c68269b6aaeb8452d9feb28 |
|
MD5 | f3c960024c5a80304ffbc9f067b520aa |
|
BLAKE2b-256 | d4a2207145fc340f79e8fa550fc3ac78587192c3e444eca0b942e99f3f914254 |
File details
Details for the file pandora_ai-0.0.23-py3-none-any.whl
.
File metadata
- Download URL: pandora_ai-0.0.23-py3-none-any.whl
- Upload date:
- Size: 34.5 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/4.0.2 CPython/3.10.12
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 170d53317c0bfb9d76cf4fed4dfa0139f2f3959436d31f9d41b65fce99db5ca7 |
|
MD5 | 2bc33f6dba9574d19d17dd706cbd3495 |
|
BLAKE2b-256 | 106d8ffa7fea5025104df2cda939df63e59647f6a308e86a2f22c250279b2dd5 |