Continue_config

A clean, practical configuration for using Continue (VS Code) with local AI models via Ollama — fully offline, no API keys, no paid tools.

Developer

GitHub: https://github.com/xoxxel

Project Overview

This project provides a structured configuration for building your own local AI coding assistant inside VS Code using Continue. Instead of relying on paid services like Copilot or Cursor, this setup allows you to:

• Run models locally (privacy-first)
• Customize behavior (chat, edit, autocomplete, etc.)
• Optimize performance based on your system
• Build your own lightweight or advanced AI agent

How It Works (Simple)

Each model is assigned a role:

• chat → conversation & reasoning
• autocomplete → fast code completion
• edit → modify code
• apply → apply changes to files
• rerank → improve search/context accuracy
• embed → search & indexing

You can assign different models to each role depending on your system power.

Requirements

• VS Code
• Continue Extension
• Ollama installed
• Minimum 8GB RAM (16GB+ recommended)

Install Models (Example)

ollama pull qwen2.5-coder:1.5b-base
ollama pull qwen2.5:3b
ollama pull llama3.1:8b
ollama pull deepseek-coder-v2:16b
ollama pull nomic-embed-text

ROLE-BASED MODEL SELECTION

Below are recommended models for each role based on system capability.

CHAT (Conversation / Reasoning)

System	Recommended Models	Notes
8GB RAM	qwen2.5:3b	Fast, lightweight
16GB RAM	qwen2.5:3b / llama3.2:3b	Balanced
32GB+ RAM	llama3.1:8b	Best quality

AUTOCOMPLETE (Fast Suggestions)

System	Recommended Models	Notes
8GB RAM	qwen2.5-coder:1.5b-base ✅	Best choice
16GB RAM	qwen2.5-coder:1.5b-base	Still optimal
32GB+ RAM	qwen2.5-coder:1.5b-base	No need heavier

EDIT (Code Modification)

System	Recommended Models	Notes
8GB RAM	qwen2.5-coder:1.5b-base	Only light edits
16GB RAM	qwen2.5-coder / llama3.1:8b	Medium tasks
32GB+ RAM	deepseek-coder-v2:16b	Best for large edits

APPLY (Apply Changes to Files)

System	Recommended Models	Notes
8GB RAM	qwen2.5-coder:1.5b-base	Fast & simple
16GB RAM	qwen2.5-coder:1.5b-base	Stable
32GB+ RAM	deepseek-coder-v2:16b	Safer for complex changes

RERANK (Context Optimization)

System	Recommended Models	Notes
8GB RAM	❌ Skip	Not necessary
16GB RAM	llama3.1:8b	Acceptable
32GB+ RAM	deepseek-coder-v2:16b	Best accuracy

EMBED (Search / Indexing)

System	Recommended Models	Notes
All Systems	nomic-embed-text ✅	Best option

SYSTEM-BASED QUICK SETUP

Low System (8GB RAM)

• chat → qwen2.5:3b
• autocomplete → qwen2.5-coder
• edit/apply → qwen2.5-coder
• embed → nomic
• rerank → disabled

Medium System (16GB RAM)

• chat → qwen2.5:3b
• autocomplete → qwen coder
• edit → qwen coder / llama3
• apply → qwen coder
• rerank → llama3.1:8b
• embed → nomic

High-End System (32GB+ RAM)

• chat → llama3.1:8b
• autocomplete → qwen coder
• edit/apply → deepseek-coder-v2:16b
• rerank → deepseek
• embed → nomic

Best Practices

• Use small models for frequent tasks (autocomplete)
• Use large models only when needed (edit / rerank)
• Avoid running too many heavy models simultaneously
• Prefer SSD over HDD
• Keep your config minimal and focused

Notes

• Large models (16B) may be slow on CPU-only systems
• First run may take time (model loading)
• Local setup = zero cost + full privacy

Contribution

Feel free to improve this config and submit a pull request.

Support

If this project helps you, give it a star ⭐