Continue Coding Assistant

Continue provides open-source Integrated Development Environment (IDE) extensions, such as for Visual Studio Code and JetBrains, and an open-source CLI that lets developers leverage custom AI coding agents.

This guide walks through how to use Lemonade Server with Continue Hub and the VS Code extension for code generation, editing, and chat capabilities, all running locally on your AMD PC.

Prerequisites

Before you start, make sure you have the following:

Software Requirements

• IDE: Visual Studio Code (v1.80+) or another supported IDE.
• Lemonade Server: Installed and set up using the Getting Started guide.
• Lemonade Server Running: The server should be running at http://localhost:8000.
• Model Downloaded: At least one model from the supported models list must be installed locally. This should match the model you plan to use in Continue Hub.

Hardware Requirements

For best results, a code-tuned model with at least 20B parameters is required. To run such a model:

• Minimum spec: PC with an integrated GPU (Ryzen™ AI 7000-series or newer) and 64 GB system RAM.
• Recommended specs:
  • PC with a discrete GPU that has 16 GB VRAM or greater (Radeon™ 7800 XT or newer).
  • Strix Halo PC with 64 GB System RAM or greater.

Setup

Configuring with Continue Hub

1. Install Models Locally

   • Use the Model Manager or lemonade-server CLI to download your desired model, for example:

   lemonade-server pull <model-name>
   

   Example downloading Qwen3-Coder:

   lemonade-server pull Qwen3-Coder-30B-A3B-Instruct-GGUF
   

2. Open Continue Hub: Go to hub.continue.dev. Sign up if you are a new user.

3. Add an Assistant: In the Assistants section, click the + next to any available assistant (e.g., llama.cpp).

4. Add Models to Your Assistant

   • Browse the list of supported models and click the + next to the model name to add it to your assistant.
   • All models shown must be installed locally using the command provided in Step 1. See the supported models list.
   • When you sign in to the Continue extension in your IDE, your configuration will automatically sync.

Example configuration screen:

For more details, see the Continue Assistants Guide.

Setting Up Continue Extension in VS Code

1. Go to Extensions Marketplace: In VS Code, click the Extensions icon in the Activity bar (default is on the left).

2. Add "Continue": Type "Continue" in the search box. Click "Install" on the Continue extension entry.

   Example marketplace screen:

3. Open Continue in VS Code: After installation completes, the Continue logo appears in the Activity bar. Click it to open the extension.

4. Sign In to Continue: Click the gear icon, then the Sign In button. A browser window will open for you to log in.

5. Select Your Assistant: Once signed in, use the drop-down menu to select the assistant you added earlier (e.g., llama.cpp).

   Example extension screen:

Working with Continue.dev

Continue provides three interaction modes for different development tasks:

1. Chat: Code explanations, debugging discussions, architecture planning
2. Plan: Provides a safe environment with read-only tools for exploring code and planning changes
3. Agent: Multi-file refactoring, large-scale changes across projects

See the Continue Documentation for detailed descriptions.

Examples

Example 1: Chat Mode - Building an Asteroids Game

In this example, we'll use Qwen3-Coder-30B-A3B-Instruct-GGUF model to build a Python game.

Input: I want to create an asteroids game using PyGame. What guidelines should I follow in the code to do so?

The model provides a basic framework for an Asteroids game. You can then prompt it to provide you some sample code to get started.

Input: Provide me a basic implementation to get started.

In the top-right corner, you can click the "Create file" to move the code from the chat window to a Python file and save it. To run, install pygame and execute the code with python main.py.

Example 2: Plan Mode - Analysis of the Game

In this example, we'll use Plan mode to have the LLM analyze your code and provide feedback. Plan mode reviews your code and suggests improvements, but does not modify your files.

To use Plan mode with large files, increase Lemonade Server's context size:

1. Stop Lemonade Server: Use the tray icon to "Quit Lemonade" or close any running Lemonade Server processes.

2. Restart with higher context size: Open a terminal and run:

   lemonade-server serve --ctx-size 8192
   

3. Use Plan mode in VS Code: Select the "Plan" option in Continue, enter your prompt and press Alt+Enter to include the currently active file as context.

Input: What improvements could be made to this game?

Example 3: Agent Mode - Improving the Game

Lastly, we'll use Agent Mode to take action to change the code to implement improvements.

Here, we can see that the agent edited the code in main.py to improve the gameplay and add colors.

Best Practices

Setup & Configuration

• Install Lemonade Server: Follow the setup guide to install and configure Lemonade Server before you begin development.
• Download Models Locally: Use lemonade-server pull <model-name> to install models you want to use. Refer to the supported models list for available options.
• Pre-load Models: Start Lemonade Server and load your models before coding sessions. This can easily be done using the Lemon tray icon and Load.
• Increase Context Size for Agent Mode: For large code changes with GGUF models, start Lemonade Server with a higher context size:

  lemonade-server serve --ctx-size 8192
  

• Create Custom Assistants: Build assistants tailored to specific languages or domains (e.g., JavaScript/TypeScript for Node.js projects) for more relevant results.
• Customize Scoping: See Continue Customization for tips on effective assistant and model scoping.

Development Workflow

• Start New Conversations for Each Feature: Begin a fresh chat for every new feature or task. Clear chat history when switching topics to keep interactions focused.
• Keep Prompts Focused: Only include the code and context relevant to your current task. This helps the model provide accurate and useful responses.
• Write Clear, Detailed Prompts: Structure your requests with a clear task description, specific requirements, and any technical constraints.
• Use Agent Mode for Multi-File Changes: Invoke agent mode with the @ symbol to perform refactoring or changes across multiple files.
• Be Specific in Your Requests: Move from broad prompts ("Create a game") to detailed ones ("Create an Asteroids game in Python using Pygame, under 300 lines, with ship controls and asteroid splitting").
• Iterate and Test Frequently: Generate an initial implementation, test it right away, and refine with targeted follow-up prompts.
• Leverage Unlimited Iterations: With local models, you can iterate as many times as needed for continuous improvement.

Common Issues

Model not appearing in Continue

• Make sure Lemonade Server is running and the model is loaded locally.
• Double-check the supported models list and install any missing models with:

  lemonade-server pull <model-name>
  

Slow response times

• Pre-load your model before starting coding sessions.
• Check your system's available RAM and close unused applications to free up resources.

Missing error handling in generated code

• In your prompt, explicitly request: "with comprehensive error handling" to ensure the model adds proper error checks.

Inconsistent code style

• Provide a sample or example of your desired code style in your prompt. The model will use this as a reference for formatting.

Resources

• Lemonade Server Setup Guide
• Lemonade Server Supported Models
• Lemonade Applications
• Continue Documentation
• Lemonade on Continue Hub