v0.1.0 · Open Source

Your own AI agent.
Running locally.

Tell Lucy what to do. She'll open apps, click buttons, fill forms, and get it done — inside a sandboxed Ubuntu VM.

Currently available for macOS
View on GitHub
lucy — task
lucy> "Open LibreOffice Writer, draft a cover letter for Acme Corp, and save it as PDF"

▶ Launching LibreOffice Writer...
▶ Typing cover letter content...
▶ File → Export as PDF...
▶ Saved: ~/Documents/cover-letter-acme.pdf

✓ Task completed — recipe saved as cover-letter.yaml
Lucy controlling LibreOffice Calc inside a sandboxed Ubuntu VM
Lucy filling a spreadsheet in LibreOffice Calc — live inside the VM
How it works

Three steps. Zero boilerplate.

No scripting, no APIs, no record-and-playback. Just say what you need.

01

Describe

Type your task in plain language. Lucy understands context — from simple file operations to complex multi-app workflows.

02

Watch

Lucy opens a sandboxed Ubuntu VM and works through your task in real-time. You see every click, every keystroke, live.

03

Replay

Every task is auto-saved as a YAML playbook. Edit it, schedule it, share it with your team — or let Lucy improve it over time.

Features

Built for real work.

Not a demo. Not a wrapper. A full automation platform you own and control.

Live VM Screen

Watch the AI agent work inside the VM in real-time. Every action is visible — no black-box execution.

Multi-Agent

Run multiple agents in parallel, each with their own isolated VM, API key, and task history.

Recipes & Playbooks

Every task is auto-recorded as YAML. Replay, edit, export, import — build a library of automations.

Scheduled Execution

Run recipes on a schedule — daily reports, weekly data pulls, or custom intervals. Set it and forget it.

Learning System

Give feedback when Lucy makes mistakes. Lessons persist and improve future task execution automatically.

Privacy-First

Fully local and self-hosted. Your data never leaves your machine. No cloud, no telemetry, no compromise.

Software Auto-Install

Lucy installs what she needs — apt, snap, flatpak. No manual VM setup. Just describe the task.

Shared Storage

Mount a persistent folder per agent to exchange files between the VM and your host machine. Drop files in, get results out.

Per-Agent Settings

Different API keys and models per agent. Mix Anthropic with local AI. Full control over every instance.

Examples

What can Lucy do?

Real prompts, real tasks. Switch between beginner-friendly and power-user mode.

Developer

Set Up a Local Git Repo

Prompt

Developer

API Health Check

Prompt

Developer

Prepare Node.js Environment

Prompt

Developer

Run a Python Script

Prompt

Developer

Check GitHub Issues

Prompt

Sysadmin

Disk Space Report

Prompt

Sysadmin

System Monitoring

Prompt

Sysadmin

Run Package Updates

Prompt

Sysadmin

Network Ping Test

Prompt

Sysadmin

Log Active Processes

Prompt

Data & Research

Generate Mock Data CSV

Prompt

Data & Research

Wikipedia Research

Prompt

Data & Research

Stock Price Lookup

Prompt

Data & Research

Download a Dataset

Prompt

Data & Research

JSON Formatting

Prompt

Office

Formal Business Letter

Prompt

Office

Daily To-Do List

Prompt

Office

Draft an Email

Prompt

Office

Invoice Template

Prompt

Office

Prepare Meeting Notes

Prompt

Marketing & SEO

Competitor Screenshot

Prompt

Marketing & SEO

Create a Content Calendar

Prompt

Marketing & SEO

SEO Meta Check

Prompt

Marketing & SEO

Trend Research

Prompt

Marketing & SEO

Campaign Folder Structure

Prompt

Students

Collect Random Facts

Prompt

Students

Literature Research

Prompt

Students

Book Summary Template

Prompt

Students

Build a Study Schedule

Prompt

Students

Simple Calculations

Prompt

Architecture

Three layers. Clean separation.

Each layer runs independently, communicating over WebSocket.

Layer 1

Host

  • Electron + React
  • TypeScript
  • Docker management
  • Scheduler
  • Recipe library
Layer 2

Agent

  • Python 3.12
  • Claude API
  • Screenshot loop
  • Action executor
  • WebSocket server
Layer 3

VM

  • Ubuntu 24.04
  • XFCE desktop
  • VNC server
  • xdotool / xclip
  • Docker or QEMU
Quick start

Up and running in 2 minutes.

1

Clone & build

Clone the repo from GitHub and build from source with Docker + Node.js. Currently macOS only.

2

Create an agent

Click New Agent, enter your Anthropic API key, and Lucy spins up an isolated Ubuntu VM.

3

Describe your task

Type what you want in plain language and hit Cmd+Enter. Watch Lucy work.

Open source. Free forever.

Lucy is built in the open. Star the repo, file issues, submit PRs — or just use it.

View on GitHub