AutoFeat

Interactive AI-powered feature engineering CLI.

Unlike traditional AutoML tools that operate as black boxes, AutoFeat keeps the user in control. The AI proposes preprocessing workflows, explains every decision, and executes transformations through Rust tools powered by featrs, while the user can inspect, modify, accept, or reject every step.

AI proposes → User reviews → User edits → AI explains → Pipeline evaluated → Repeat

Installation

cargo install autofeat

Or build from source:

git clone https://github.com/DeathSurfing/autofeat
cd autofeat
cargo build --release

Usage

autofeat --dataset data.csv

Keyboard shortcuts

Key	Action
A	Agent — live reasoning, tool execution, conversation
D	Dataset — schema, statistics, null counts, distributions
W	Workflow — interactive DAG editor
T	Tools — execution history, runtime, outputs
S	Settings — general, LLM, pipeline, evaluation config
H	Help — keyboard shortcuts
Enter	Edit node
Space	Disable / enable node
I	AI suggestions
R	Review pipeline
Ctrl+S	Save
Q	Quit

Architecture

                    User
                      │
                      ▼
               Ratatui Interface
                      │
                      ▼
              Workflow Controller
                      │
          ┌───────────┴───────────┐
          ▼                       ▼
     AI Planning Agent      Workflow Graph
          │                       │
          ▼                       ▼
     Tool Registry          Editable Pipeline
          │
          ▼
     featrs
          │
          ▼
     Polars

The LLM only plans and reasons — every computation is performed by Rust tools through featrs.

Tech Stack

• Rust — performance and safety
• Polars — DataFrame operations
• featrs — feature engineering transformers
• Ratatui — terminal UI
• rig — LLM orchestration (OpenRouter)
• Tokio — async runtime

License

MIT