Look. Ask the bench what is on it. The model sees the pieces before it decides anything.
The only robot bench you plug into your AI chat with one URL
Made for AI tinkerers who use Claude or ChatGPT every day and want to see how well a model plans and acts in the physical world.

Four tools your AI picks up on the bench
Move. Steer the robot around the miniature watchmaker’s bench, one action at a time.
Pick and place. Lift a piece, set it down somewhere else. Every stacking puzzle comes down to these two.
From one URL to a first move
Use the chat you already pay attention to. The bench is one connector away.

Sample caption
Connector added. Tools ready: look, move, pick and place.
What a run looks like
Sample content, not a live run
Hand it the mission
Type a physical goal into the chat you already use. Claude or ChatGPT takes it from there.
Move the tower from peg A to peg C. A big disc never sits on a small one.
Looking at the bench first.
Read it caption by caption
Each look, move, pick and place shows up as a caption, so you follow the plan as it happens.
- lookthree discs on peg A
- picksmallest disc, peg A
- placesmallest disc, peg C
- pickmiddle disc, peg A
- placemiddle disc, peg B
Try the classic: Tower of Hanoi
Three pegs, a stack of discs and one rule about order. A plain test of whether the plan holds up.
One token pays for one robot action
Token plans, as on the prototype. Every look, move, pick or place spends a token.
One token, one action. Look, move, pick and place each cost one.
A Tower of Hanoi run. It takes about 60 tokens.
CHF 0per month
300 tokens, about 5 Tower of Hanoi runs.
Plans are easy to write. Let your AI pick something up.
One connector URL, one miniature watchmaker's bench, and a model that has to act on what it sees.