horstos

Concepts · Architecture

One controlled execution stack

Programs, agents, and external controllers enter through different contracts but share the same HorstOS Runtime and PLEXA safety boundary.

01Preview

Program mode

A defined HorstScript or PLEXA program executes functions, conditions, loops, and parameters reproducibly.

Input
Code · taught motion · datasets
Feedback
Program state · command result · robot state
02Planned

Agentic mode

Task intelligence selects governed capabilities, skills, or tools without bypassing execution controls.

Input
Goal · instruction · image
Feedback
Context · telemetry · perception
03Planned

Realtime streaming

An external controller continuously provides bounded targets through a dedicated session contract.

Input
VLA · teleoperation · external controller
Feedback
Pose · joints · sensors · watchdog

Shared boundary

From client intent to controlled motion

Each layer owns a distinct responsibility. Clients never become the robot safety system.

L3

Clients

SDKs, ROS 2, programs, agents, external controllers

API

Governed interfaces

Runtime APIs, capabilities, bounded sessions

L2

HorstOS Runtime

Validation, execution, feedback, watchdogs

L1–L0

PLEXA control

Realtime motion, drives, sensors, I/O, safety

System guardrail

Agent tools, ROS nodes, simulators, and external controllers must not bypass Runtime, motion-control, safety, or hardware boundaries.

Open interface overview