5 unstable releases
| 0.3.0 | Jan 12, 2026 |
|---|---|
| 0.2.2 | Dec 7, 2025 |
| 0.2.1 | Nov 30, 2025 |
| 0.2.0 | Nov 27, 2025 |
| 0.1.0 | Nov 27, 2025 |
#42 in #autocomplete
Used in pzsh
12MB
262K
SLoC
aprender-shell
AI-powered shell completion trained on your command history.
Installation
cargo install aprender-shell
Pre-trained Base Model
A base model trained on synthetic developer commands is available on Hugging Face Hub:
# Download base model
huggingface-cli download paiml/aprender-shell-base model.apr --local-dir ~/.aprender
# Use with aprender-shell
aprender-shell suggest "git " -m ~/.aprender/model.apr
Model: paiml/aprender-shell-base
The base model includes 401 common developer commands (git, cargo, docker, kubectl, npm, python, aws, terraform) and contains no personal data.
Usage
# Train on your zsh history (creates personalized model)
aprender-shell train
# Get completions for a prefix
aprender-shell suggest "git ch"
# Install zsh widget for tab completion
aprender-shell zsh-widget >> ~/.zshrc
source ~/.zshrc
Features
- Sub-10ms suggestion latency
- Context-aware N-gram completions
- Privacy-safe: sensitive commands filtered automatically
- Supports zsh and fish shell history formats
- Incremental training on new commands
Fine-tuning
Start with the base model and fine-tune on your history:
# Download base model
huggingface-cli download paiml/aprender-shell-base model.apr -o base.apr
# Train incrementally on your history
aprender-shell train --base base.apr
License
MIT
Dependencies
~14–28MB
~365K SLoC