Menlo Stack

Menlo Stack enables developers to embody AI agents into humanoid robots, turning software into physical labor. It consists of five interconnected components that together enable a humanoid labor force:

• Agent SDK  — Deploy agent payloads to compliant humanoids with safety guarantees
• Edge API  — Interface for low-level robot control and telemetry
• Uranus  — World simulator and digital twin engine for scenario testing
• Cyclotron  — Sim-to-real motor control pipeline for locomotion and manipulation
• Data Engine  — Telemetry and continuous improvement system

You get a closed deployment loop from agent definition to physical execution and back, compressing iteration from weeks to minutes.

Agent Native

The software is designed for agent-native robotics:

• Real-time kernel — Handles low-latency reflex loops for balance and manipulation
• Cloud connectivity — High-level reasoning happens in the platform; the robot handles execution
• Standardized interfaces — Compatible with any humanoid that meets hardware requirements
• Open API — Agent developers interact through well-documented interfaces, not motor controllers

Programmable at every layer

The software is open at every layer:

• Agent-to-hardware translation — Agents express high-level intentions (navigate to location, manipulate object, respond to human) to coordinate physical action
• Sensor fusion — Depth cameras, force-torque sensors, and IMUs feed into a unified perception layer
• Motor control — Low-level motor controls
• Safety enforcement — Hard boundaries prevent actions that could damage hardware or harm humans
• Telemetry collection — Real-time performance data streams