LM Studio for Copilot Chat
Run local LM Studio models inside VS Code Copilot Chat with streaming responses, tool calling, terminal execution, filesystem tools, and optional image generation.
What it does
- Adds LM Studio-hosted chat models to the Copilot Chat model picker
- Streams responses from LM Studio into VS Code chat
- Exposes a small built-in toolset for terminal and filesystem actions
- Supports tool-calling friendly local models with configurable tool budgeting
- Supports optional image generation through either A1111 or DALL-E-compatible APIs
Features
- 🔌 Auto-discovery: Automatically detects models loaded in LM Studio
- 💬 Chat Integration: Use LM Studio models directly in VS Code's chat interface
- 🔄 Streaming: Real-time streaming responses
- 🛠️ Tool Calling: Built-in terminal, file, directory, search, and image generation tools
- 🖥️ Integrated Terminal Launch: Start LM Studio server from VS Code terminal
- 🖼️ Optional Image Generation: Route image requests to Automatic1111 or DALL-E-compatible endpoints
- ⚙️ Configurable: Customize server URL, timeouts, tool budget, and more
Requirements
- VS Code 1.104.0 or later
- LM Studio running with a model loaded
- LM Studio's local server enabled (default:
http://localhost:1234)
Setup
- Install and open LM Studio
- Load a model in LM Studio
- Start the local server in LM Studio (usually enabled by default)
- Install this extension in VS Code
- (Optional) Set
lmstudio-copilot.launchCommand (for example: lms server start)
- Open the Command Palette and run "LM Studio: Start Server in Integrated Terminal"
- Run "LM Studio: Refresh Available Models"
Extension Settings
This extension contributes the following settings:
lmstudio-copilot.serverUrl: LM Studio server URL (default: http://localhost:1234)lmstudio-copilot.apiKey: Optional API key for LM Studio-compatible serverslmstudio-copilot.launchCommand: Command to launch LM Studio server in integrated terminal (example: lms server start)lmstudio-copilot.terminalName: Terminal name used to run the server (default: LM Studio Server)lmstudio-copilot.autoStartServer: Auto-run launchCommand when extension activates (default: false)lmstudio-copilot.startupWaitMs: Wait time before connection check after terminal start (default: 3000)lmstudio-copilot.enableTerminalTool: Allow model-invoked terminal commands (default: true)lmstudio-copilot.terminalToolName: Terminal name for model tool commands (default: LM Studio Tool Terminal)lmstudio-copilot.autoRefreshModels: Automatically refresh models on startup (default: true)lmstudio-copilot.requestTimeout: Request timeout in milliseconds (default: 60000)lmstudio-copilot.maxTools: Maximum number of tools exposed to the model per request (default: 20)lmstudio-copilot.enableToolCalling: Enable tool calling support when the model can use toolslmstudio-copilot.imageGenBackend: Choose dalle or a1111lmstudio-copilot.imageGenEndpointUrl: Base URL for the selected image generation backendlmstudio-copilot.imageGenModel: Image model name for DALL-E-compatible backendslmstudio-copilot.imageGenWidth, lmstudio-copilot.imageGenHeight, lmstudio-copilot.imageGenSteps, lmstudio-copilot.imageGenOutputDir: Image defaults
Commands
LM Studio: Refresh Available Models - Refresh the list of available models from LM StudioLM Studio: Start Server in Integrated Terminal - Starts LM Studio server command in VS Code terminalLM Studio: Stop Server Terminal - Stops the dedicated LM Studio terminalLM Studio: Check Server Connection - Test connection to LM Studio server
Usage
- Make sure LM Studio is running with a model loaded
- Open VS Code's Chat view (Ctrl+Shift+I or Cmd+Shift+I)
- Click on the model selector dropdown
- Select an LM Studio model (prefixed with "LM Studio")
- Start chatting!
Let models run commands in the integrated terminal
The extension now exposes a chat tool named #runInTerminal.
- In chat, ask the model to use
#runInTerminal to run commands (for example, compile or test commands).
- Commands run in a dedicated integrated terminal (
lmstudio-copilot.terminalToolName).
- Invocations are automatically executed without a confirmation prompt.
Many local models become unreliable when given dozens of tool schemas at once, especially when MCP servers add very large tool inventories.
- By default, the extension sends at most
20 tools per request.
- Tools already used in the current conversation are prioritized.
- Set
lmstudio-copilot.maxTools to 0 to disable the limit.
This helps reduce hallucinated tool names, malformed inputs, and context-window bloat.
Image generation
The built-in lmstudio_generate_image tool supports:
- Automatic1111 / Stable Diffusion WebUI via
lmstudio-copilot.imageGenBackend = a1111
- DALL-E-compatible APIs via
lmstudio-copilot.imageGenBackend = dalle
Set lmstudio-copilot.imageGenEndpointUrl to your image server URL.
Troubleshooting
Models not appearing
- Make sure LM Studio is running
- Make sure a model is loaded in LM Studio
- Check that the server URL is correct in settings
- Run "LM Studio: Check Server Connection" to verify connectivity
Slow responses
- LM Studio performance depends on your hardware and the model size
- Consider using a smaller/faster model
- Increase the timeout in settings if needed
Development
# Install dependencies
npm install
# Compile and watch
npm run watch
# Package the extension
npm run package
# Build a VSIX for local install / Marketplace submission
npm run package:vsix
Releases
- Git tags use the format
v<package.json version>
- GitHub releases are created from those tags
- The first release for this repo is
v0.1.0
VS Code Marketplace publishing
This repo is prepared for packaging with vsce.
# Build a VSIX locally
npm run package:vsix
# Publish to the Visual Studio Marketplace
# Requires a VS Code publisher and VSCE_PAT in your environment
npm run publish:vsce
Before publishing to the Marketplace for real, you may still want to add:
- a square extension icon (128x128 PNG recommended)
- a
CHANGELOG.md entry for each release
- optional CI automation for tagging and VSIX builds
License
MIT
