Flow Control¶

Real work rarely runs in a straight line. A paid order goes one way and an unpaid one another; a list has to be worked item by item; three lookups can run at once; a step waits for a payment to clear; a call to a flaky service has to survive a hiccup. Flow control is how you shape a run's path to fit the work - the decisions, loops, parallel steps, waits, and recoveries you reach for.

The tools below are peers, and a real flow often uses several. Find the one that answers your question.

Send a run down the right path¶

When the next step depends on a test - is the order paid, which plan is the customer on - you split the run so each case is handled on its own terms. A yes/no test sends it one of two ways; a value picks among several. See Yes/No Branching and Routing on a Value.

Repeat a stretch of work¶

When the same steps have to run more than once - over and over until something becomes true, or once for every item in a list. See Repeating Steps and Working Through a List.

Run steps at the same time¶

When independent steps do not depend on each other, run them together so the flow finishes at the speed of its slowest step, not the sum of them all. See Running Steps in Parallel.

Wait before continuing¶

When a step should not happen yet - hold for a set time, or until an outside system reports back. See Adding a Delay and Waiting on an External System.

Keep going when a step fails¶

By default a failure stops the run. Instead you catch it and send the run down a recovery path you build - log what went wrong, fall back to a default, alert someone, or retry. See Handling Errors.

AI Router - route a run on a decision only an AI can make, like free-text intent or an image.
Triggers - what starts a flow, such as an outside event.
Scheduling - run a flow on a timetable.
Subflows and Call Flow - hand a stretch of work to another flow.