Claude Code + Unsloth API setup on RunPod

This tutorial connects Claude Code to Unsloth's API endpoints instead of Anthropic's models. The presenter hosts Unsloth Studio on a rented RunPod GPU and explains that Claude Code can also connect from a local computer. The demonstrated inference runs remotely, and the GPU rental incurs costs.

The setup uses an RTX Pro 4500 pod with a 200 GB container disk and exposes ports 8005 and 8006. The presenter installs Studio through a Linux terminal in JupyterLab, launches it on port 8005 and sets a password. The installer creates a Python 3.13 environment.

The presenter downloads a GGUF model with the UD Q4 KXL quantization, shown as 21 GB. After creating an API token and installing Claude Code, he sets three environment variables for the base URL, authentication token and model. A short greeting checks the connection. He also edits settings.json with the stated aim of improving speed, but provides no measured comparison or coding benchmark.

The closing section proposes asking Claude Code to prepare an Unsloth fine-tuning project with one GPU. It does not show a completed training run; the presenter leaves that walkthrough for a separate video.