# How to Set Up NVIDIA Jetson for Remote AI Development with Codex

## Executive summary

This tutorial details the secure setup of an NVIDIA Jetson device for remote AI development using Codex. The process establishes robust connectivity (SSH over USB and Wi-Fi), integrates the Codex CLI for project management, installs NVIDIA Jetson Device Skills, and validates the environment's readiness for AI workloads. Key steps include configuring non-root Docker permissions and running a CUDA container test to ensure full GPU access, all while maintaining a remote, reproducible development environment.

## Key takeaways

- Secure Remote Connectivity Setup: Established dual connectivity (USB SSH and Wi-Fi) and used Codex to automate the setup of SSH keys, ensuring secure, password-less access to the Jetson.
- Codex Project Integration: The Codex CLI was installed and authenticated on the Jetson, allowing the remote project to be registered and managed within the Codex desktop app, streamlining file access and command execution.
- Device Skills and Containerization: NVIDIA Jetson Device Skills were installed to manage device-specific instructions. The workflow was containerized using CUDA libraries, eliminating the need for the full JetPack SDK on the host machine.
- GPU Access Validation: The final step involved configuring non-root Docker permissions and running a CUDA container test. This verified that the Jetson Orin processor was correctly initialized and accessible for AI workloads.

## Technical details

- SSH Setup: Codex was used to generate and deploy dedicated SSH keys, enabling secure, password-less remote login from a Mac to the Jetson via USB.
- Network Configuration: The Jetson was configured for Wi-Fi connectivity while maintaining the USB connection, ensuring redundancy and automatic network reconnection.
- Codex Authentication: The Codex CLI was installed and authenticated using a browser-based link and one-time code, authorizing the tool without requiring local account passwords.
- Docker Permissions: To allow the Jetson user to run Docker containers, the system-level permission was updated using a specific command (run locally, then reconnected) to grant non-root access.
- CUDA Validation: A temporary CUDA container was run to detect the Jetson Orin processor, initialize CUDA, and complete a GPU memory test, confirming readiness for AI development.

## Practical implications

- The setup provides a blueprint for creating secure, containerized, and remotely accessible edge computing environments.
- Build engineers can leverage this methodology to standardize the deployment of AI models on resource-constrained hardware like the Jetson.
- The use of Codex streamlines the complex process of environment setup, reducing manual configuration steps and potential human error.

## Topics

NVIDIA Jetson, AI Development, Docker, CUDA, SSH, Codex CLI, Remote Computing, https://www.jetson-ai-lab.com/tutorials/ai-assisted-development-on-jetson/, https://developer.nvidia.com/blog/nvidia-jetpack-7-2-1-adds-agentic-video-skills-and-t3000-emulation/

Source: https://www.youtube.com/watch?v=CLGaG0JeUBI