Shell ML Autocomplete

How It Works

• Trained on real shell command histories
• 3-gram Markov model with Trie prefix matching
• Compressed APR binary format (~9KB model)
• Zero-infrastructure: runs entirely in browser
• Sub-millisecond inference latency