System Requirements
Docket is designed to work on most modern computers. Here's what you need.
Minimum Requirements
| Component | Requirement |
|---|---|
| Operating System | macOS 12+, Windows 10+, or Linux (Ubuntu 20.04+) |
| RAM | 4 GB (for smaller models) |
| USB Port | Any USB port (USB-A or USB-C) |
| Storage | No additional storage needed (runs from USB) |
Recommended Specifications
| Component | Recommended |
|---|---|
| RAM | 8 GB or more |
| USB Port | USB 3.0+ for faster model loading |
| CPU | Multi-core processor (4+ cores) |
RAM Requirements by Model
Different models have different memory requirements:
| Model | Size | RAM Required |
|---|---|---|
| Micro | 0.6B | 2 GB |
| Spark | 1B | 4 GB |
| Lightning | 1B | 4 GB |
| Quick Chat | 2B | 4 GB |
| Snapshot | 4B | 6 GB |
| Unchained | 4B | 6 GB |
| Storyteller | 7B | 6 GB |
| Unrestricted | 7B | 6 GB |
| Vision | 7B | 8 GB |
| Developer | 7B | 6 GB |
| Balanced | 9B | 8 GB |
| Powerhouse | 14B | 12+ GB |
tip
If you have limited RAM, start with smaller models like Micro (0.6B) or Lightning (1B).
Platform-Specific Notes
macOS
- Works on both Intel and Apple Silicon Macs
- Apple Silicon (M1/M2/M3) provides significantly better performance
- Allow the app in System Preferences > Security if prompted
Windows
- Requires Windows 10 version 1903 or later
- The app may require administrator privileges for first run
- Windows Defender may scan the app on first launch
Linux
- Tested on Ubuntu 20.04+, Debian 11+, and Fedora 36+
- May require
chmod +xto make the executable runnable - Some distributions may require additional dependencies
Network Requirements
Docket works completely offline for local models. Internet is only needed for:
- Downloading additional models from Hugging Face
- Using OpenRouter API models
- Checking for updates
Troubleshooting
If you experience performance issues:
- Close other applications to free up RAM
- Use a smaller model that fits in your available memory
- Use USB 3.0+ port if available for faster model loading
- Check system temperature — AI inference can be CPU-intensive
See Common Issues for more help.