Wendy

Wendy is an open-source OS, CLI and cloud that lets coding agents simulate, ship and operate robots and drones, moving one project from a laptop simulation to Jetson, Raspberry Pi and other real hardware.

Wendy is a full-stack platform for physical AI that covers the path from a simulated robot on a laptop to a fleet of real machines. It combines an open-source operating system (WendyOS), a command line tool, a device agent and a cloud service, all designed so that a coding agent such as Claude Code or Codex can drive each stage. The same project moves from simulation to hardware without being rewritten. The workflow has three stages. In simulation, a robot runs inside a WendyOS virtual machine on the developer's computer, with MuJoCo physics, a locomotion policy and ROS 2 beside the app. Profiles exist for the Unitree Go2 and G1 (both marked preview), and there is a twelve-drone formation example. When the simulation behaves, the same `wendy run` command targets a real device: the CLI builds the app into a container, finds the board over USB-C or the local network, and sends only the missing chunks. Logs stream back so the agent can see what the device sees. Once deployed, the Wendy Agent provides telemetry through an OpenTelemetry endpoint, pinned deploys without SSH, over-the-air updates and, in beta, fleet tools through Wendy Cloud. Agent integration is central. Claude Code and Codex use a plugin, while Cursor, Windsurf and Claude Desktop connect through an MCP server exposing 44 tools, each tagged read-only, mutating or destructive. Most CLI commands answer in JSON when no terminal is attached, and a `wendy.json` manifest declares the hardware an app may use. Provisioned devices accept API calls only over mutual TLS. Supported targets include NVIDIA Jetson, Raspberry Pi, Qualcomm Dragonwing boards, generic arm64 or x86_64 Linux machines, the Unitree G1 computer, Apple Silicon Macs (beta) and ESP32 microcontrollers (preview). The audience is robotics and embedded teams in manufacturing, energy, medical devices, defense and research who want agent-assisted development rather than hand-built deployment tooling. It sits closer to a robotics deployment stack than to a general ROS distribution, and several pieces are still in preview, beta or planned status.