Europe Embodied hackathonAn autonomous Go2 inspector.
- ROS 2
- NVIDIA Jetson
- Unitree Go2
By Anirudh Jupudi, Adyansh Gupta, and Manas Reddy Arumalla. Shared with the team’s permission.
Wendy is the open-source OS and toolchain that gives Claude Code, Codex, and other coding agents a fast, reliable, and repeatable way to ship to robots, drones, and edge devices. Your agent deploys over USB-C, your network, or the cloud, and a provisioned device takes commands only over mutual TLS.
curl -fsSL https://install.wendy.dev/cli.sh | bashRuns on
Example session: a coding agent is asked to make a Unitree Go2 wave and then patrol. It starts wendy run --watch, which deploys the greeter app to the robot over USB-C, then edits main.py twice. Each edit redeploys in a few seconds, and the second one starts the dog navigating a route of four waypoints.
01 · Fast
<200 msTo rebuild a Python edit with a Stagefile, versus about 30 s with a Dockerfile. Only the changed chunks then go to the device.
02 · Reliable
A/BOS updates install to a spare partition and roll back on their own if the new system fails to boot or fails its post-update checks.
03 · Repeatable
wendy.jsonOne manifest declares the app and its hardware access, with a published schema your agent and CI can check.
04 · Secure
mTLSOnce a device is provisioned, the CLI, the device, and Wendy Cloud talk only over mutual TLS with per-device certificates.
Claude Code and Codex install the Wendy plugin, and Cursor, Windsurf, and Claude Desktop connect through the Wendy MCP server. Your agent sets up the device, deploys the app, reads the logs, and fixes what broke, using the same commands and authenticated connections you do.
01 · Fast
A Stagefile rebuilds a Python edit in under 200 ms, and Wendy sends only the chunks the device is missing, so each edit is running on the hardware within seconds. Your agent sees the result as the logs stream back.
02 · Reliable
When no terminal is attached, most CLI commands answer in JSON, and the CLI stops with a clear error instead of waiting on a prompt. MCP results are size-capped, with a hint on how to narrow the query.
03 · Repeatable
wendy.json declares the app and the hardware it may use, checked against a published schema. The same project deploys the same way from your agent, a teammate, or CI.
04 · Secure
Your agent works through the Wendy API, which a provisioned device serves only over mutual TLS, and release Jetson and Raspberry Pi images have no SSH server. The CLI refuses a device whose identity doesn’t match, apps get only the hardware wendy.json declares, and every MCP tool is tagged read-only, mutating, or destructive.
Then run /reload-plugins and start with /wendy-agentic-coding:wendy-install.
Then ask your agent
Find my Jetson, deploy the camera-feed-yolo template, and tell me what the camera sees.
Your coding agent calls the wendy CLI, which builds the app into a container, finds the device, and sends only the chunks the device is missing. Logs stream back over the same link, so the agent sees what the device sees.
wendy CLI
Plug in one cable. The device appears as a USB network adapter at a fixed link-local address, so there is nothing to set up on your laptop.
Each device advertises a readable name, like wendyos-zestful-stork.local, over mDNS.
Enrolled devices dial out, so there are no inbound ports or static IPs to manage.
Wendy Agent
$ wendy install
Flash WendyOS from your laptop with its Wi-Fi, name, and optional Wendy Cloud enrollment set in the same command, or add the agent to a Linux machine you already run.
$ wendy init
Scaffold an app from a template, or bring any project with a Dockerfile. Add --assistant claude or codex to hand the new project to your agent.
$ wendy run
Build, deploy, and stream logs in one step. With --watch, every change your agent saves redeploys.
Wendy covers the work that sits between a working prototype and a device in the field, with one toolchain from the first build to the fleet.
Write Python, Swift, Rust, TypeScript, or C++, start from a template, or bring any project with a Dockerfile or Compose file. Hand a new project to Claude Code or Codex with one flag, or work in VS Code or Cursor with the Wendy extension.
Read the docs$ wendy init --template# start from a sample app
$ wendy init --framework ros2# ROS 2 project
$ wendy init --assistant claude# hand off to Claude Code
Flash WendyOS to a supported board, or install Wendy Agent on the Linux computer or Mac you already run.
A community robot, our own field hardware, and notes from teams piloting WendyOS today.
Europe Embodied hackathonBy Anirudh Jupudi, Adyansh Gupta, and Manas Reddy Arumalla. Shared with the team’s permission.
Wendy BoxSupports your qualified fire watch and existing site procedures.
“If we found Wendy two years ago, we would have saved $4M in product development costs.”
CEO & Co-Founder
Autonomous farm and lawncare equipment
“I’m testing WendyOS on a Jetson to enable in-field AI to tune old, dumb control loops. I call it Automated Dynamic Local Loop Control.”
Automation Manager
Refinery production improvement
“We stopped tech projects because they weren’t practical or helpful. But in testing, the Wendy Box is a game changer. We’re already changing how we plan to run our crews.”
Site Manager
Turnaround resource management
Teams use Wendy to run perception, inspection, and control on devices far from a data center, including places with no network at all.
Install the Wendy CLI, add the Wendy plugin to Claude Code or Codex, and ask your agent to deploy a sample app to your hardware. You can also start with a MuJoCo simulation if you don’t have a robot yet.
curl -fsSL https://install.wendy.dev/cli.sh | bashPlanning a pilot or a fleet?
Talk with our engineers about your hardware, security review, and rollout. We also build and ship field-ready devices like Wendy Box.