build an agent loop the way you'd read one
you can build an agent loop in your dashboard now. chain the agents, drop in a human gate, and hit go. each step hands to the next on its own, and you only step in where you chose to.
no yaml, no config file. you assemble the loop the way you’d read it, top to bottom.
chain the agents
a loop is a relay. the research agent finishes, the writer picks up its output. the writer finishes, the reviewer picks up. you wire the order once in the builder, then it runs the handoffs for you. when one step is done, the next one unblocks and the right agent gets pinged. no copy-pasting between tabs, no babysitting the baton.
the gate is where you step in
most of the loop is hands-off. the exception is a human gate: a single step where the loop pauses, pings you, and waits for a tap. approve and it keeps moving. tap “do better” and it bounces back to the agent to rework.
you don’t sit in every step. you sit in the one that matters, and the loop runs itself the rest of the way.
watch it run
once a loop’s going, the “your loops” view shows you where each one is: which step it’s on, how far along it is, and a status chip when one needs your tap. you can tell at a glance which loops are running, which are waiting on you, and which are done.
runs are on-demand for now. you hit go, it goes. scheduled triggers are next on the list.
one thing about your box
loops run on your server’s own agent runtime. if your box hasn’t picked up the latest update yet, you’ll see an “updating” state instead of the builder. it clears within a day on its own, or you can pull the update now from the Server tab and start building right away.
go build one
we said a while back that the loop itself is the easy part. this is the part you can finally see.
head to your loops dashboard. pick two agents, chain them, drop a gate, hit go. you’ll have a loop running before your coffee’s cold.