Reuse a step's output instead of paying twice
Tick “Reuse output” on an expensive step and an identical input serves the cached result at $0 — so you can iterate on what comes after it.
In this article (5)
The frustrating shape of building a chain: step three is wrong, so you fix step three — and pay for steps one and two all over again. Pinning removes that.
How it works
On an AI step in the Builder (Simple or Map), tick Reuse output. From then on, when that step runs with the same input, FlowArray serves its stored result instead of calling the provider — cost $0, and the result carries a Reused · no charge badge.
"Same input" is exact and per item: the provider, the model, the step, and the fully resolved input (including anything an upstream step produced) all have to match. Change the prompt, change the item, change an earlier step's output, and it re-runs normally.
Why it is opt-in
Because for some steps a cache hit is the wrong answer. Image generation is stochastic — if you press Run again wanting a different variation, silently handing you the previous image would be a footgun. So nothing is ever pinned unless you asked for it, and an unpinned step behaves exactly as it always did.
Where it pays off most
- Clean up image (background removal, upscale) — deterministic for a given photo, so reuse is pure saving. Pin it and a re-run of the batch cleans nothing twice.
- An expensive upstream step you are not currently editing, while you iterate on the steps after it.
- A shared prep step called from several flows via Run workflow.
Forcing a fresh result
Press Clear pinned outputs on the step. The next run calls the provider again and stores the new result. Use it when you deliberately want a new variation, or when an upstream change should have invalidated the cache but you would rather be certain.
The honest limits
- It can only ever under-charge, never double-charge — the reuse path books $0 and skips the call entirely.
- Agent steps cannot be pinned. An agent calls tools, and tools can have side effects; replaying a cached answer would skip work the outside world expects to happen.
- A pin is per workspace, so a colleague running the same flow on the same input benefits from it too.