🍋 Lemonade for GitHub Copilot

Demo

Lemonade allows you to use local LLMs on GitHub Copilot Chat!

🚀 Getting Started

Make sure Lemonade is running
- We recommend setting the context size to at least 32k by using the Lemonade tray or --ctx-size cli option.
Install the Lemonade for GitHub Copilot extension
Open VS Code's chat interface, click the model picker, and click "Manage Models..."
Select "Lemonade" provider, confirm the server URL (defaults to http://localhost:13305/api/v1) and optional API key in the form
Choose a model (Qwen3-Coder-30B is a great start!)

You can now start chatting with your local LLM models! 🥳

NOTE: Models appear only after a server has been added through the form above. Add one entry per Lemonade server to use several.

🌟 Why Choose Lemonade for Copilot?

🔒 Complete Privacy: Your code never leaves your machine. Everything stays local and secure
💰 Zero API Costs: No usage fees, no tokens to buy - just pure local AI power
⚡ Lightning Fast: Direct connection to your local server means instant responses
🌐 Works Offline: No internet? No problem! Code assistance anytime, anywhere
🛠️ Advanced Tool Support: Full function calling capabilities for complex tasks

Requirements

VS Code 1.109.0 or higher
Lemonade server 8.1.10 or higher

🔧 Configuration

The extension connects to http://localhost:13305/api/v1 by default. You can change this by:

Opening VS Code Command Palette (Ctrl+Shift+P)
Running "Manage Lemonade Provider" command
Entering your custom Lemonade server URL
Optionally, entering your API key (if your server has LEMONADE_API_KEY configured)

Limiting visible models

The optional models setting restricts the model picker to the listed model ids. Leave it unset or empty to show all models. Names the server does not return are ignored, and matching is exact.

List the ids your server provides:

curl -s http://localhost:13305/api/v1/models

If the server form does not show a Models field, add it to the Lemonade entry in chatLanguageModels.json (open it from the Command Palette with Chat: Manage Language Models, or find it in your VS Code user data folder), then reload the window:

{
	"name": "Lemonade",
	"vendor": "lemonade",
	"url": "http://localhost:13305/api/v1",
	"models": ["Qwen3-Coder-30B-A3B-Instruct-GGUF", "Qwen3.5-2B-GGUF-Q8_0"]
}

Names must match the server's model ids exactly. If none of them match, a warning lists the names the server did not return. Remove models to show every model again. This applies to servers added through the provider form, not through the older lemonade.manage command.

Environment Variables

Variable	Default	Description
`LEMONADE_CTX_SIZE`	`128000`	Fallback context window (in tokens) for older Lemonade servers or models whose effective context cannot be resolved. Current servers use per-model context settings and runtime/auto-tuned sizes when available.

Tip: For older servers, launch VS Code with LEMONADE_CTX_SIZE set to the actual context limit. Current Lemonade servers normally determine this per model automatically.

Support & License

Open issues: https://github.com/lemonade-sdk/lemonade/issues

🙏 Acknowledgments

This plugin was originally based on the excellent work by the Hugging Face team. We're grateful for their foundational work.

Lemonade for GitHub Copilot

lemonade-sdk