Use local LLMs in VS Code – 100% Local, Zero External Calls

Write, refactor, and review code with AI that runs entirely on your machine. Nothing leaves your device.

Integrated features: Chat | Inline Editing | Autocomplete | Code Review | Test Generation

Privacy-first: Telemetry and external tracking are disabled by default. Your code stays yours.

All powered by local LLMs through VS Code's Continue.dev extension—offline, private, and fast.

Note: we provide no guarantees on security, but turn off Wifi on your macbook to confirm ;)

Quickstart

Install

Use the LM Studio setup option, instead of Ollama, for superior visual control, model switching, and usage monitoring.

Prerequisites: Install Ollama or LM Studio, Python 3.10+, VS Code, 8 GB+ RAM recommended (4 GB minimum)

# clone and install (into a virtual env)
git clone https://github.com/plpxsk/vs-local.git
cd vs-local
pip install -e .

# 2. Run guided setup — prompts you to choose Ollama or LM Studio
python -m cli setup

# 3. Open in VS Code and install the Continue.dev extension when prompted
code . # or open project in VSCode

# 4. Start coding with AI: Cmd+L (chat) | Cmd+I (inline edit) | Tab (autocomplete)

Setup will: detect OS and RAM, recommend a model tier (default small = phi4-mini), check or install Ollama, pull the model, generate a Continue.dev config locked to localhost, and optionally install it to ~/.continue/ (prompts before overwriting). If ~/.continue/config.json already exists, use --force to overwrite without prompting.

For help selecting local models, see below, Model Tiers.

To verify install, run: python -m cli verify

Use in your own project

After setup, the Continue.dev config is global and works in any repo. To also copy the VS Code privacy settings into your project:

# activate venv as needed
cd /path/to/your-project
python -m cli vscode-init

Usage

Run the app with python -m cli

python -m cli setup              # Full guided setup (prompts for runtime)
python -m cli setup --lmstudio   # Skip prompt, use LM Studio
python -m cli verify             # Health check + network audit
python -m cli models             # List model tiers and local models
python -m cli models --pull qwen2.5-coder:7b  # Pull a specific model
python -m cli config             # Regenerate Continue.dev config
python -m cli firewall           # Show firewall setup instructions
python -m cli vscode-init        # Copy VS Code settings into current project

Examples

See examples/ for exercises (code generation, refactoring, code review, test generation). Troubleshooting: docs/TROUBLESHOOTING.md.

What You Get

• Chat - Ask questions, generate code, explain code (Cmd+L)
• Inline edit - Select code and describe changes (Cmd+I)
• Autocomplete - Tab completions as you type
• Code review - Find bugs, security issues, code smells
• Test generation - Generate pytest tests from your code

All powered by local models. Nothing leaves your machine (not guaranteed).

Model Tiers

Tier	Model	Download	RAM	Best For
small	phi4-mini	~2.5 GB	4 GB	Fast completions, low-end hardware
medium	qwen2.5-coder:7b	~4.5 GB	8 GB	Recommended default
large	deepseek-coder-v2:16b	~9 GB	16 GB	Highest quality

The setup command auto-detects your RAM and recommends a tier.

Security

Three layers ensure no data leaves your machine:

1. App config - All API endpoints locked to localhost, telemetry disabled
2. Network audit - python -m cli verify checks for external connections
3. Firewall templates - OS-specific rules in security/ to block outbound traffic

See docs/SECURITY.md for details.

Learn more

• Models — Model tiers, hardware requirements, and top picks
• Runtimes — Ollama vs LM Studio vs MLX
• Security
• Troubleshooting

License

MIT