Cline and LM Studio: local coding setup in VS Code

This tutorial connects Cline in VS Code to LM Studio for a coding assistant that uses a local LLM. The setup covers installing the extension, downloading a model, starting LM Studio's local server and selecting LM Studio as Cline's API provider. The speaker uses Qwen 3 4B and says models trained for tool use have worked best for him, even when they are not dedicated coding models.

Hardware and context length affect the workflow. The speaker says CPU-only operation is possible, while GPU capability affects speed and available memory limits model size and context length. He enables Cline's compact prompt setting for a small context window and turns on YOLO mode, described as auto-approval for file creation and command execution.

The first task asks Cline to write and run a Python script that lists files in the current directory. In the demo, the model repeatedly restarts its reasoning. The speaker ejects it, reloads it with manual load parameters and increases the context length to the maximum available setting. The next attempt creates the script, executes it and displays the results.

A second task produces a snake game with high-score tracking. The initial game does not respond; after feedback, the speaker reports that the revised version works. These examples show his results with this setup rather than a guarantee for other hardware or models. He finishes by ejecting the model to release GPU and memory resources.