drift-ml 0.2.11

drift — terminal-first, chat-based AutoML. Open source. No tokens. No auth.

drift

Terminal-first, chat-based AutoML. Open source. No tokens. No auth.

---

Install

pipx install drift-ml

Or:

npm install -g drift-ml

(Both require pipx install drift-ml for the Python CLI.)

---

Run

drift

On first run, drift downloads and starts the engine automatically. No backend setup. No config. No tokens.

---

How it works

• Local-first — Engine runs on your machine. Data never leaves.
• Chat-based — load data.csv, predict price, try something stronger
• Auto-start — Engine downloads and starts in the background. You never start a backend manually.
• No auth — No API keys for drift. (You need an LLM for training: Ollama, Gemini CLI, etc.)

---

Example (CLI)

drift › load iris.csv
drift › predict variety
drift › try something stronger
drift › quit

---

Use as library

pip install drift-ml

from drift import Drift

d = Drift()
d.load("iris.csv")
d.chat("predict sepal length")
result = d.train()
print(result["metrics"])

Or with an existing engine:

from drift import Drift

d = Drift(base_url="http://localhost:8000")
d.load("data.csv")
reply = d.chat("predict price")
print(d.get_last_reply(reply))

---

Windows (npm / pipx)

On Windows, when using npm or pipx, the engine can inherit a limited PATH — Gemini CLI may not be found. We now auto-prepend npm/pipx bin paths. If you still see "I'm ready for your first command" or planning failures:

1. Add GEMINI_API_KEY — Create .env in project root with GEMINI_API_KEY=your-key. The backend will use the API instead of the CLI.
2. Run backend manually — cd backend && python -m uvicorn main:app --host 0.0.0.0 --port 8000, then set DRIFT_BACKEND_URL=http://localhost:8000.

---

Philosophy

drift should feel like git, docker, brew — a tool you trust immediately. Zero friction. Open source.

---

Web UI (optional)

This repo also includes a web UI. For development:

./start.sh

Then open http://localhost:3000

---

Contributing

See CONTRIBUTING.md.

---

License

MIT