How to connect LM Studio with VS Code to program with local AI

If you're tired of relying on monthly subscriptions or having your personal data and private projects end up in the cloud, setting up your own artificial intelligence environment is the ideal solution. Connect LM Studio with VS Code Programming with local AI can be a great idea.

Thanks to tools like LM StudioToday, it is entirely feasible to run powerful language models directly on your computer's hardware, while maintaining the absolute privacy and working without an internet connection. The real magic happens when we get this local brain to communicate with our favorite code editor. By linking LM Studio with Visual Studio CodeWe transformed a simple editor into an intelligent workstation capable of generating code, explaining complex functions, and auto-completing entire lines, all by leveraging the power of your GPU or NPU without spending a single cent on external APIs.

Preparing the ground with LM Studio

The first thing we need to connect LM Studio with VS Code is to have the LM Studio application installed and configured. For those seeking programming assistance, not just any model will do; ideally, you should download versions optimized for code. A very reliable option is the Qwen2.5 CoderHowever, you must choose the size (such as 14B or 3B parameters) based on the available RAM memory on your computer to prevent the system from crashing.

If you use a Mac with Apple Silicon And if you have 32GB of RAM or more, you can try more robust models like Codestral-22BAn advanced trick for macOS users is to unlock the hardwired memory limit using the command sudo sysctl iogpu.wired_limit_mb=50000This allows for the loading of higher quality models, such as quantization models. Q8_0 without memory errors.

Once the model has been chosen, it is essential to access the developer panel and activate the local serverBy default, LM Studio starts a server at the address http://localhost:1234Make sure to adjust the Context Length (the context window) according to your needs; for extensive code tasks, increasing this value to 16.384 or even 32.768 tokens is recommended, although this will consume more of your machine's resources.

 

Connect LM Studio with VS Code using the Continue.dev extension

In order for VS Code to "talk" to the server we just set up, the most popular and versatile option is the extension Continue.devAfter installing it from the marketplace, a new panel will appear in the sidebar. To start chatting, we must add a model by clicking on the selector at the bottom and choosing LM Studio as a provider and selecting the auto-detection option so that it finds the model loaded on the server.

After connecting LM Studio with VS Code, you can activate the intelligent autocomplete (those gray suggestions that appear while you type), you'll need to edit the configuration file config.json located in the folder .continue of your user. This is where you define the tabAutocompleteModel, indicating the API URL and the specific model name, thus achieving a smooth experience similar to Copilot but completely offline.

Exploring alternatives: AI Toolkit and OpenCode

When connecting LM Studio with VS Code, it's not all about Continue.dev. Microsoft offers the extension AI Toolkitwhich is especially interesting for those who own Copilot+ certified PCs and have a Dedicated NPUThis tool allows you to manage a catalog of local and cloud models, facilitating the uploading of lightweight versions such as Phi-4, optimizing energy consumption and the performance of modern hardware.

On the other hand, if you're looking for a more agent-oriented experience (capable of editing files and executing terminal commands), there's integration with OpenCodeThis solution acts as a headless server connecting LM Studio to an advanced chat panel in VS Code. The most powerful aspect of this workflow is its support for... MCP (Model Context Protocol)which allows the agent to interact with databases, browsers, or file systems through servers configured in a file .mcp.json.

Advanced settings and BYOK

VS Code has evolved to allow the scheme BYOK (Bring Your Own Key)This means you can add custom endpoints through the language model editor (the gear icon in the chat selector). When configuring a Custom EndpointYou can define whether the API uses the Chat Completions or Messages format, allowing local models to be integrated directly into the native VS Code chat interface without heavy extensions.

It's important to mention that, although the chat works perfectly, some features of utility (such as the automatic generation of chat titles or commit messages) usually require a model configured in the settings chat.utilityModelIf you don't have a Copilot subscription, simply assign your local model These variables in the VS Code configuration will prevent you from losing any productivity functionality.

For those who prefer the terminal, LM Studio now offers an Anthropic-compatible endpoint, allowing the use of tools such as Claude Code pointing to the environment variable ANTHROPIC_BASE_URL to the local server. This versatility turns your computer into an artificial intelligence hub where you can jump between different models and interfaces depending on the complexity of the task you are solving.

It's worth it How to connect LM Studio with VS Code? Having a programming assistant that doesn't depend on the network is a huge competitive advantage, as it allows you to work in restricted environments and ensures that sensitive code never leaves your hard drive. By combining the LM Studio server with extensions like Continue or AI Toolkit, you have a flexible and free ecosystem that adapts to the power of your hardware, transforming the way you write and optimize software.