Concept
Your steps. Plus a definition of done.
A Loop is a graph you design — exactly the way you'd build a workflow — that repeats under a small brain until a success condition you wrote is genuinely met. Reach for one when you know the job but one pass rarely nails it.
Three ways to automate
Where a Loop sits — and where it doesn't.
Destaris has three process objects, and a Loop is not a midpoint between the other two. A workflow and a Loop are authored the same way: you build the steps. The difference is that a Loop also carries a definition of “done” and reruns itself until that's met. An Agentic Workflow is the one where you don't author the steps at all.
For steps you know
Workflow
A fixed set of steps you build once. It runs the same way every time, for next to nothing.
- Who designs the graph
- You
- Who defines “done”
- Nobody — it just runs
For steps that take a few passes
Loop
You build the steps, same as a workflow, and you also write down what “done” means. Each pass, a small brain checks the result against that and sends specific stages back to try again — until it passes.
- Who designs the graph
- You
- Who defines “done”
- You — a success condition
For figuring it out
Agentic Workflow
Give it a goal instead of steps. A local AI brain plans, runs, and adapts until the goal's met — building a workflow as it goes.
- Who designs the graph
- The brain, every iteration
- Who defines “done”
- The brain
So the three don't sit on one line. Workflow and Agentic Workflow split on do you already know the steps? A Loop doesn't sit on that axis at all — it asks the second question: does running them once actually finish the job?
Anatomy
Two pinned nodes, and everything you put between them.
A Loop's graph starts at a pinned input and ends at a pinned brain. In between is an ordinary Destaris graph — the same palette, the same nodes, the same canvas you'd build a workflow on.
input and brain are pinned. Everything between them is yours.
Not done — correction, addressed by stage
- research:
- surface §4.2 more fully next pass
- implement:
- AC-3 is still missing
verify gets no correction — it did its job. The whole graph runs again.
Every pass re-runs your whole graph — there's no resuming from halfway. When it finishes, the brain either calls it done or routes a correction addressed by stage, so the stage that needs to change is the one that hears about it.
The success condition
Write the bar, not the steps.
“Every acceptance criterion in the spec is implemented and its tests pass” is a condition. “Run the test suite” is a step — and it's already in your graph.
The brain never redesigns your graph
It only judges and corrects. The steps that run on pass five are the steps you drew on pass one — which is exactly what separates a Loop from an Agentic Workflow.
It judges what a person would check
The success condition can name files in your workspace — a spec, an acceptance checklist — and the brain re-reads them fresh off disk before every single pass, never a snapshot from when you wrote it.
It can't run forever
An iteration cap is always the backstop. A Loop that ran cleanly but never met the bar comes back stuck, which is a different answer from failed — and it tells you which one you're looking at.
It works itself out of a job
A Loop that stops needing a brain becomes a workflow.
A brain costs tokens. Once a Loop stops needing one, Destaris offers to drop it — export the stages as a plain workflow and run them for close to nothing from then on.
The bar for that offer is deliberately high: the Loop has to have met its condition on the first pass, five runs running, on an unchanged graph, across at least three distinct inputs. Repetition alone isn't enough — five clean runs against one spec show the brain has that spec's answer, not that your graph is deterministic. That's deterministic-first applied to a whole process: intelligence is scaffolding you remove once the structure stands.
Build your first Loop
The full model — the pinned nodes, the success condition, how a correction reaches a stage, and the one trap that turns a red test suite into a dead run.