Shimmy VS Code Extension
The official VS Code extension for Shimmy: The 5MB Alternative to Ollama.
Features
- One-click server management: Start, stop, and restart Shimmy from the status bar
- Instant code generation: Right-click any text selection to generate code completions
- Auto-discovery: Works with any GGUF model, with zero-friction LoRA serving
- Zero configuration: Smart defaults that just work
- Status monitoring: Real-time server status in the status bar
Quick Start
- Install Shimmy binary:
cargo install shimmy --features llama
- Install this extension
- Set your model path in settings (or use
SHIMMY_BASE_GGUF environment variable)
- Click the status bar to start the server
- Right-click selected code → "Generate from Selection"
Commands
Shimmy: Start Server - Start the Shimmy inference serverShimmy: Stop Server - Stop the running serverShimmy: Restart Server - Restart the serverShimmy: Generate Code - Open prompt dialog for code generationGenerate from Selection - Generate code based on selected text (context menu)
Configuration
| Setting |
Description |
Default |
shimmy.serverUrl |
Shimmy server URL |
http://localhost:11435 |
shimmy.binaryPath |
Path to shimmy binary |
shimmy |
shimmy.modelPath |
Path to GGUF model file |
(uses SHIMMY_BASE_GGUF env var) |
shimmy.autoStart |
Auto-start server on activation |
false |
Shimmy provides OpenAI-compatible endpoints, so it works with:
- GitHub Copilot: Configure server URL in settings
- Cursor: Point to
http://localhost:11435
- Continue.dev: Add as custom model provider
- Any OpenAI-compatible tool: Just change the base URL
Troubleshooting
Server won't start?
- Check that
shimmy binary is in your PATH
- Verify model path is correct
- Ensure port 11435 is available
No code generation?
- Check server status in status bar (should show green checkmark)
- Verify model is loaded properly
- Check VS Code developer console for errors
Performance tips:
- Use smaller models (1-3B parameters) for faster responses
- Place models on SSD for faster loading
- Consider LoRA adapters for specialized tasks
About Shimmy
Shimmy is the 5MB alternative to Ollama - a single-binary local inference server that:
- Starts in <100ms vs Ollama's 5-10 seconds
- Uses <50MB RAM vs Ollama's 200MB+ overhead
- 100% OpenAI API compatible
- First-class LoRA adapter support
- Zero configuration required
Privacy-first, cost-free, blazing fast local AI.
License
MIT License - see Shimmy repository for details.
| |
