Agent interfaces
Seven ways compute actually gets consumed
Every one of these is a front end onto the same network and the same settlement contract: it holds a $ROBO balance, escrows against a job before it runs, and settles when the work is verified. What differs is the shape of work each is built around, and therefore what it optimises for.
01
Halomesh
The gateway agent
What does an inference endpoint look like when the agent behind it pays for its own compute?
Read the story02
Loomcell
The training agent
What has to be true before a research team hands a 400-node run to an agent and walks away?
Read the story03
Orrery
The render agent
What happens to a pipeline when the agent is allowed to wait for a better price?
Read the story04 · In build
Proofwork
The verifying agent
When an inference result settles money, how much proof is enough — and how much is theatre?
05 · In build
Cindergrid
The operator agent
What does the earning side look like when an agent manages your capacity, bond and payouts?
06 · In build
Stratabase
The scheduling agent
How should an agent place work when the machines underneath it may leave on Tuesday?
07 · In build
Tidemark
The treasury agent
If capacity is a transferable position, what does it mean to let an agent manage it?