This repository primarily contains notes and documentation about OpenAI's Code Interpreter. Important files include:
system_prompt.md: the system prompt for the Code Interpreter.system.md: details about the system environment where the code interpreter operates, including information about the OS,- processor, RAM, and environment variables.
env: the system environment variables.requirements.txt: A list of Python package dependencies required for Code Interpreter.dev_scripts: A directory containing utility scripts that assist in the exploration and understanding of Code Interpreter's capabilities.
Below is what Code Interpreter reveals about its working environment.
The Python environment in which the code is executed is based on Python 3. Python is a high-level, interpreted, and general-purpose dynamic programming language that focuses on code readability. The Python Software Foundation's standard runtime environment is used, which includes a wide variety of standard libraries.
Libraries can be found in requirements.txt.
Python packages are managed using pip, the standard package manager for Python. You can install, upgrade, and remove Python packages using pip.
The code execution takes place in a Jupyter Notebook environment. Jupyter is an open-source web application that allows the creation and sharing of documents that contain live code, equations, visualizations, and narrative text. The Jupyter notebook is an interactive computing environment that enables users to author notebook documents that include code, widgets, plots, and narrative text. The notebook extends the console-based approach to interactive computing in a qualitatively new direction, providing a web-based application suitable for capturing the whole computation process.
Python code that you send for execution will time out after 120 seconds if it doesn't complete. This is to ensure that long-running code doesn't lock up the interactive environment.
Files can be stored and persisted in the /mnt/data directory. This allows you to upload and download files from the Jupyter environment. Note that these files will be available only for the duration of your current session and will not be preserved across different sessions.
The Python environment does not have internet access. This means that you cannot make HTTP requests or API calls directly from your Python code. This is a security feature designed to prevent unauthorized network activity.
The OpenAI's Code Interpreter allows ChatGPT to interpret and execute Python code within the Jupyter Notebook environment. This can be used for running Python scripts, performing computations, generating plots, and more.
Each session operates in an isolated environment. This means your code won't have access to the underlying operating system, ensuring the safety and privacy of your data.
The Python and Jupyter environments are containerized. This means that each user session is isolated within its own environment, which includes a specific version of Python and certain pre-installed packages. This isolation ensures that the Python code execution environment is consistent across all sessions and users. Containers are lightweight and portable, which makes it easy to manage and scale the infrastructure.
Kubernetes is used to orchestrate the containers. It is responsible for deploying containers, scaling them according to demand, and managing their lifecycle. Kubernetes also handles networking between containers and provides services like load balancing, service discovery, and secret management.
Within the Kubernetes cluster, each container has its own IP address and can communicate with other containers using standard TCP/IP protocols. However, for security reasons, the containers are isolated from the internet and cannot make outgoing network requests. This means that you can't fetch data from the web or call APIs directly from your Python code.
While containers themselves are ephemeral (meaning they are destroyed when the session ends), data can be stored and persisted using volumes. In this environment, the /mnt/data directory is mounted as a volume, so you can use it to save and load files within a session.
Files can be stored and persisted in the /mnt/data directory. This allows you to upload and download files from the Jupyter environment. The /home/sandbox directory is the default working directory, where scripts are executed and where relative paths are resolved from. Note that these files will be available only for the duration of your current session and will not be preserved across different sessions.
The use of containers and Kubernetes also provides a number of security benefits. The isolation between containers helps to limit the impact of security vulnerabilities. In addition, network policies can be used to control which containers can communicate with each other, and Role-Based Access Control (RBAC) can be used to control who can access the Kubernetes API and what actions they can perform.
In the context of the Code Interpreter, each user session operates within a single, isolated container. This container includes both the language model and the Python execution environment. Importantly, containers associated with different user sessions do not communicate with each other, ensuring the privacy and security of each individual session.
Contributions to this repository are welcome. If you have additional notes or scripts that you believe would benefit others in understanding or using OpenAI's Code Interpreter, feel free to create a pull request.