ROBOT ROOMS FOR AI AGENTS
We're building an API that lets your agent run physical experiments in staged robot rooms. We're preparing the first physical pilot. Join the waitlist to help shape it.
Join the waitlistScroll to follow one experiment
01 / FIND A ROOM
Each room is a staged environment: two grippers and laundry, an arm with objects and bins, a humanoid in a living room. Your agent searches them and picks the one that fits.
Illustrative inventory
02 / APPROVE A BUDGET
Your agent quotes the session and asks before it spends. Approve $10 and it cannot spend more. No signup forms, just an API key and payment.
Illustrative quote
03 / SETUP IN ABOUT A MINUTE
The session starts, the robot homes, the cameras stream and the agent looks at how the room was left. After that, the experiment time is yours.
Target for ready, compatible rooms
04 / RUN THE EXPERIMENT
Your agent sends each step through the SDK. Our robot adapters run it inside the room's safety limits while the cameras record everything.
Scripted illustration
05 / REVIEW THE EVIDENCE
Your agent analyzes the observations, shows you the recording and proposes the next experiment, inside the budget you set.
How your agent connectsIllustrated facility concept · loading motion
EARLY ACCESS
IN DEVELOPMENTWe're building robot rooms that your agent can use through an API. Join the waitlist for pilot access and launch updates.
The rooms, prices and workflow on this site are concepts. Physical bookings and a hosted MCP connection are not available yet.
PLANNED AGENT WORKFLOW
LIKE RUNPOD OR FAL.AI, FOR ROBOTSDevelopers ask Claude Code or Codex to build a robot. When the plan needs real hardware, the agent reaches Actuator Science through the tools it already uses, books a room and brings back the evidence.
Your agent lists rooms, reads their equipment and limits, quotes a session and runs it, all from the conversation you are already having.
A Python SDK for controllers and policies that stream steps, read cameras and measurements, and export the run when it ends.
A plugin for coding agents bundles the tools with guidance on choosing a room, asking for a budget and analyzing what comes back.
"Test this grasp controller on a real arm. You can spend up to $10."
THE WAREHOUSE
ROOMS OF DIFFERENT SIZESCompare the robot, props, observations and control interface. Start with a tabletop arm, or explore the larger rooms we intend to build. Your agent checks the exact published configuration before booking.
PLANNING CATALOGUE · All physical setups below are planned. Rates are illustrative USD hypotheses, not live offers. The current executable service is local software simulation.
One room-hour includes the listed robot and cameras. Alternative layouts may share equipment; these cards do not represent six independently available stations. Final hardware, rates and support require admission. Room images are AI-generated concepts, not photographs or exact hardware models.
Planning catalogue · JSON ↗ How agents check compatibility ↗PLANNED SESSION TERMS
SIMPLE, BOUNDED, FAST TO STARTTry a duration, start-state option and execution location. Nothing here reserves a room or spends your budget.
ILLUSTRATIVE ESTIMATE / USD
Within your example cap
Illustrative room time and preparation only. An actual all-in quote must include any other charges before owner approval.
Read the approval flow ↗Arm examples use US$80/hour. Other rates come from saved planning research and remain unvalidated. The current local software-simulation API uses a separate US$20/hour pricing preview and development credits; this page changes no service rate. Pricing notes.
Longer term, autonomous agent research teams with their own budgets will run physical experiments here instead of building a lab.