Software

Anyone can ship you a robot. Software puts it to work.

Every RaaSX robot arrives with your task already recorded, trained and tested. We run the software and keep it running. Your engineers can open it up, but nothing on this page asks them to.

1:1
Your environment rebuilt in simulation
100+
Automated tests before every deployment
96%
Accuracy in multilingual dialogue
24/7
Monitored and updated over the air

From your task to a robot that does it.

Four steps. Only the first one happens on your floor.

/ 01

We record the task

Our team films the job in your own building, the way your people do it now.

Up to 1,000 recordings a day
/ 02

We train the model

Those recordings teach the robot the movement. Nobody on your side writes code.

About 48 GPU-days per task
/ 03

We test it in simulation

Your environment is rebuilt in software and the task runs there until it holds up.

Under 5% deviation from reality
/ 04

The robot arrives ready

The trained task is loaded before delivery, and stays current over the air.

No site visit for an update

What comes with the robot.

Configured during deployment, not built from scratch.

Watched over by RoSa, our own fleet AI

RoSa is the RaaSX layer on top of the Agibot stack. She reads the telemetry of every unit we have out, schedules maintenance before something breaks, pushes updates over the air and drafts the first answer to every SLA ticket.

Meet RoSa
/ module_01

Pick & place

Picking, sorting and placing, trained on your own SKUs and hooked into your WMS.

/ module_02

Vision

Object recognition, quality inspection and scene understanding.

/ module_03

Reception

Welcoming, guiding and promotional interaction, multilingual and in your own house style.

/ module_04

Dashboard

Control, scheduling and status for the units on your floor, in the browser.

For your engineers

Open, if you want to go in.

Everything below this line is optional. Nothing above it depends on your team touching it.

Layer 01 · Interaction

WorkGPT & ActionGPT

Full-duplex multilingual dialogue at 96% accuracy, plus language-to-motion generation. Company knowledge bases through RAG.

96% accuracyFull-duplexRAG knowledge basesFace wake-up 99%
Layer 02 · Intelligence

GO-1 VLA + GE-1 world model

GO-1 turns an instruction into a physical action, GE-1 predicts what happens next, and Genie RL picks up new tasks fast.

GO-1 VLAGE-1 world modelGenie RLJetson Thor on-device
Layer 03 · Middleware

AimRT framework

A lightweight C++20 runtime, open source on GitHub, speaking the protocols your systems already use.

Open sourceROS 2HTTP · gRPCMQTT · Zenoh
Layer 04 · Developer

SDKs and low-code tools

AimDK for Python and C++, Isaac Sim support, and LinkCraft and LinkSoul plus VR teleoperation for teams without a line of code.

AimDKIsaac SimLinkCraft · LinkSoulVR teleoperation
On film

Genie Studio 2.0 on film.

Genie Studio 2.0 · from data to deployment in one platform

Bring us the task. We bring the stack.

Talk software →
0