Agenticros

AgenticROS is an open source toolkit that lets AI agents control ROS 2 robots through natural language. A user types or messages a request, an AI agent calls AgenticROS tools, and the toolkit turns that request into ROS 2 actions over rosbridge, returning robot feedback to the conversation. It works with real robots or with simulators. It is built for robotics developers and researchers who want to drive mobile bases, arms and sensors from agents they already use. Supported agents include OpenClaw, NemoClaw, Claude Code and Claude Desktop, OpenAI Codex CLI, Hermes Agent and Gemini, with MCP setup handled by a single command, and a local vision language model through Ollama is an option for keeping chat and camera reasoning on the machine. Robot skills are exposed as named capabilities that can be chained into multi step missions, paused, resumed or cancelled, with transcripts saved to shared memory for handoff between agents. It integrates with Nav2 and MoveIt2, supports 2D webcams and RealSense depth cameras, and offers fleet discovery. A safety validator enforces velocity limits and workspace bounds, and an emergency stop command bypasses the AI entirely. AgenticROS is free and published as an open source npm package and GitHub repository. No paid plans are listed. Getting started takes one npx command on Node.js 20 or later; driving a real robot or running the Gazebo simulations also requires ROS 2 Humble or Jazzy.