quackd
The communication layer between people and robots, and between robots. Talk to any robot in your own words, with an LLM for its brain. Robots that have never me
What is quackd?
The communication layer between people and robots, and between robots. Talk to any robot in your own words, with an LLM for its brain. Robots that have never met check each other's passports and share the work. Microduck, Huebot, Open Duck Mini, LeRobot arms, XLeRobot carts, AlohaMini, ToddlerBot, MOSS, ROS bases and industrial arms on ROS 2.
Overview of quackd
What it is
quackd is a command-line interface for robots. It allows users to connect multiple robots, command them, and let them work together, each with a large language model (LLM) as its brain. The name is pronounced “quacked”. It began as the brain daemon for the Microduck robot. In quackd, a task is a .duck file and a group of robots is called a flock. Nine robots are supported: Microduck, Open Duck Mini v2, SO-101 class arm through LeRobot, any wheeled base over rosbridge, XLeRobot dual-arm cart, AlohaMini, ToddlerBot, Huebot, and MOSS. One of these bodies (the SO-101 arm) runs on real hardware; the others are in a simulator or mock. Each robot is registered once by name. A goal can be given from a terminal or from a chat with Claude, to one robot or to a flock. A contract enforces which skills the model may use, how many steps it gets, and when it must ask a human for approval.
Key features
- Supports multiple LLMs: Claude, OpenAI, Gemini, Grok, Mistral, DeepSeek, Cohere, Qwen, Kimi, GLM, Meta via their APIs; OpenRouter; and open-source models via Ollama, vLLM, llama.cpp, or LM Studio.
- Decision LLMs (e.g., TypeSafe's Jev, OpenAI's Decisions API) can be used to choose among robot skills without generating text.
--decision-mode shadowrecords what a decision LLM would have chosen without affecting the run.- On the SO-101 arm, learned policies from Hugging Face Hub, NVIDIA GR00T, and Physical Intelligence pi0 can drive the arm for short segments.
- Robots communicate with each other to divide work when given a flock goal.
- A contract system limits skill usage, step count, and requires human approval when needed.
Use cases
- Command a single robot to perform a task, e.g., “find the ball and kick it”.
- Give a goal to a flock of robots, e.g., “I want to eat and drink something”, and let them coordinate: one checks the fridge, another grabs utensils, a third brings the food.
- Use at home or in an office to manage multiple specialized robots (walking, grabbing, carrying).
- Run robots offboard from a laptop, with the LLM brain living in the cloud or on a local machine.
Getting started
See the repository README for installation and quick-start instructions.
License and pricing
The license is Apache-2.0. The project is open source.
Based on the project's public README and repository metadata.