✦ Doushi AI CLI (doushi / dsh)

The official command-line interface for Doushi.ai — empowering developers, data scientists, and hackers to build, train, inspect, and deploy autonomous machine learning models directly from their terminal.

---

✨ Features

• ⚡ Zero-Boilerplate ML: Go from raw datasets (.parquet, .xls, .xlsx, .json, .csv, .tsv) to production-ready models in a single command.
• 🔒 Security-First Auth: AWS CLI-style credential management (~/.doushi/credentials with strict 0600 permissions).
• 🛡️ Pre-Flight Limit Checks: Validates dataset integrity and file limits locally before uploading.
• 🔄 Real-Time Streaming: Watch the autonomous self-healing agent loop live in your terminal.
• 🚀 Live Inference & Pipes: Predict via inline JSON, batch CSV files, or standard input streams (cat data.csv | doushi predict <id>).
• 📦 No Vendor Lock-In: Export raw scikit-learn/XGBoost training code, model.pkl, FastAPI microservice, and Dockerfile with doushi export.

---

📦 Installation

Install via pip:

pip install doushi-cli

Or run instantly without installing using pipx:

pipx run doushi demo

(Both doushi and dsh are available as CLI aliases)

---

🚀 Quickstart in 60 Seconds

1. Authenticate your terminal

doushi configure

(Prompts for your API key from app.doushi.ai)

2. Try the 30-second instant demo

doushi demo

(Trains an autonomous model on a sample customer churn dataset and shows live performance metrics)

3. Train on your own dataset

doushi train ./my_dataset.csv --goal "Predict customer churn probability"

---

📖 Command Reference

Authentication

doushi configure       # Set up API Key & verify credentials
doushi whoami          # View active organization, plan tier, and project quotas
doushi logout          # Remove stored local credentials

Model Training & Projects

# Upload dataset and train with autonomous agent
doushi train ./dataset.csv --goal "Predict house sale prices"

# List all organization projects and best metrics
doushi projects list

# Inspect detailed model parameters, KPIs, and dataset preview
doushi projects view <project_id>

# Delete a project
doushi projects delete <project_id>

Predictions & Inference

# 1. Predict with single inline JSON record
doushi predict <project_id> --data '{"age": 32, "tenure": 12, "balance": 4500.0}'

# 2. Batch predict from a test CSV file
doushi predict <project_id> --file ./test_records.csv --output ./predictions.csv

# 3. Unix standard input piping
cat records.csv | doushi predict <project_id> | jq '.predictions[]'

Observability & Code Export

# Stream real-time agent execution & self-healing sandbox logs
doushi logs <project_id> -f

# Export standalone model bundle (model.pkl + FastAPI + Dockerfile)
doushi export <project_id> --out ./model-bundle/

---

📂 Exported Model Bundle Structure

Running doushi export <project_id> gives you everything you need to run completely independent of Doushi:

model-bundle/
├── model.pkl            # Pretrained model weights
├── pipeline.py          # Complete scikit-learn / XGBoost training script
├── requirements.txt     # Locked dependencies
├── serve.py             # FastAPI REST microservice
├── Dockerfile           # Production container
└── README.md            # Local deployment instructions

---

🛠️ Development & Contributing

Local Setup

git clone https://github.com/doushi-ai/doushi-cli.git
cd doushi-cli

python3 -m venv venv
source venv/bin/activate
pip install -e ".[dev]"

Run Test Suite

python3 -m unittest discover -s tests

---

📄 License

Distributed under the MIT License. See LICENSE for more information.

---

Built with ✦ by the Doushi.ai Team