05 - Installing OpenCode and Connecting it to LM Studio

In this post, I will show you how to install OpenCode and connect it to your local LM Studio instance for private, offline AI-assisted development.

Why#

OpenCode is an AI coding agent that runs in your terminal. Unlike cloud-based agents, it can work with models from any provider—including local LLMs served by LM Studio. This keeps your code private and eliminates API costs.

Connecting OpenCode to LM Studio means you can use your locally hosted models (like Qwen3.5-35B-A3B) without sending data to external servers. You get the power of an agentic workflow with full control over which model runs and where.

How#

Installing OpenCode#

OpenCode supports multiple installation methods. On macOS, I use Homebrew for easy updates:

brew install anomalyco/tap/opencode

Alternative methods:

• npm: npm install -g opencode-ai
• Bun: bun install -g opencode-ai
• Install script: curl -fsSL https://opencode.ai/install | bash

Verify installation:

opencode --version

You should see something like 1.2.15.

Connecting to LM Studio#

OpenCode connects to models through a JSON configuration file. This approach is more persistent than the interactive wizard and lets you define multiple models upfront.

Step 1: Create the config file

Create ~/.config/opencode/opencode.json (or .opencode/opencode.json in your project root for per-project settings):

{
  "$schema": "https://opencode.ai/config.json",
  "provider": {
    "lm-studio": {
      "npm": "@ai-sdk/openai-compatible",
      "name": "LM Studio",
      "options": {
        "baseURL": "http://127.0.0.1:1234/v1",
        "apiKey": "lm-studio-very-secure-key"
      },
      "models": {
        "qwen3.5-35b-a3b@q4_k_xl": {
          "name": "Qwen3.5 35B - Q4_K_XL",
          "supportsTools": true
        },
        "qwen3.5-9b@q4_k_xl": {
          "name": "Qwen3.5 9B - Q4_K_XL",
          "supportsTools": true
        }
      }
    }
  },
  "model": "lm-studio/qwen3.5-35b-a3b@q4_k_xl"
}

Step 2: Start the LM Studio server

Launch LM Studio, load your model (e.g., Qwen3.5-35B-A3B), then click Server in the bottom-left corner and hit Start Server. The server runs at http://localhost:1234 by default.

If running LM Studio on a separate machine (e.g., a headless server), enable Serve on Local Network in Server Settings. This binds the server to all network interfaces, allowing connections from other devices on your local network. Update the baseURL in your config to use the server’s local IP address (e.g., http://192.168.1.100:1234/v1).

Step 3: Run OpenCode

Start OpenCode with your new configuration:

opencode

The model key at the top level specifies which model to use by default. Format is provider-name/model-id.

The $schema field enables auto-completion in editors like VS Code. The npm field specifies the AI SDK to use for the OpenAI-compatible API. LM Studio’s server mode implements this format natively.

Confirming the Connection#

Test your setup with a simple prompt:

What is 2+2?

You should see the model respond within seconds using your locally hosted Qwen3.5-35B-A3B model.

If it hangs:

1. Verify LM Studio’s server is running
2. Check the port matches (default: 1234)
3. Ensure the model is loaded and selected

Key Points Recap#

1. OpenCode runs in your terminal — It’s an agentic coding assistant that works with any LLM provider, including local models.
2. LM Studio serves OpenAI-compatible API — Start the server in LM Studio to expose your local models at http://localhost:1234.
3. Configure OpenCode via JSON file — Create ~/.config/opencode/opencode.json with your provider settings and models.
4. Use any local model — Once connected, your locally hosted Qwen3.5 or other GGUF models work seamlessly in OpenCode.
5. Privacy first — Your code never leaves your machine when using local models through LM Studio.