Frequently asked questions

Do I need my own API keys?

Yes — FlowArray runs on your provider accounts (Google Gemini, Anthropic, OpenAI, OpenRouter, fal.ai, or any OpenAI-compatible endpoint). You are billed by them directly, at their prices. You can build and rehearse an entire flow before adding a key, using free test mode.

Does FlowArray charge me for the AI calls?

No. Every figure it shows is an estimate of what your provider will bill you — how cost is worked out. Your invoice is the source of truth.

Can someone on my team see my API keys?

No. Keys are encrypted at rest and masked everywhere; members and viewers cannot even open the API keys page, and the server refuses their requests to it however they ask. They run flows through a pool by name. See roles.

Are my keys ever sent to the AI, or logged?

No. Credentials are never put into a prompt, a log, an exported workflow file, a share, or a notification.

Does using several keys mean several times the speed?

Sometimes, and FlowArray refuses to pretend otherwise. Several providers meter per organisation or project, not per key — so two keys from one project share one limit. Mark them with the same Shared account name and FlowArray stops assuming capacity you do not have. Pools aggregate capacity you legitimately hold; they are not a way around a provider's rate limits.

How do I stop a mistake becoming an expensive mistake?

Set a daily spend limit on the workspace — it bounds every member, schedule and trigger in any rolling 24 hours. Then rehearse inputs in the Data tab for $0, and run one item before running two hundred.

What happens if a run half-fails?

Each item is tracked independently, so the rest of the batch completes. Retry failed re-runs only the failures, keeping what worked, with the cost shown first — see why a run failed.

What if the worker or the server restarts mid-run?

Execution state lives in the database, not in memory, so recovery reconciles from it. A paid, long-running job (video) is re-polled using the handle already held rather than resubmitted, so a crash cannot bill you twice. Anything it cannot recover is parked as needs review rather than retried silently.

Can a workflow run without me?

Yes — on a timetable (schedules), started by another system (triggers), or from your own code (the API). All three obey the same spend ceilings, and all three are visibly badged in the run list.

Can an automated caller run up a bill?

A trigger secret can start exactly one workflow and read nothing, with per-trigger caps (20 items a call, 10 runs an hour by default) and free duplicate deliveries. An API token is broader — it acts as you — so use a trigger for anything outside your control. Both are bounded by the workspace's daily limit.

Can Claude create and run flows for me?

Yes — connect FlowArray as a tool server. Flows created that way arrive as drafts with no keys attached, so they produce placeholder output until a person points them at real pools: chatting can never spend your money by itself.

How many items can one run handle?

A typed or imported batch takes up to 200 rows (spreadsheet import); a trigger or API call takes 20 items per call; a schedule captures up to 100 rows. Loops are capped at 5 tries and nested flows at 5 deep. The caps exist so an unattended mistake stays small.

Can I show a client the results without giving them access?

Yes — a read-only share: the flow, the outputs and each item's inputs, with cost hidden by default, an optional expiry, and one-click revoke. Add approvals if you want their verdict per item.

Can I get results out automatically?

Yes: the Send to a URL card POSTs each item's result to Slack, Zapier or your own backend; the Library can export after the fact; and run-outcome webhooks tell a system when a run finished.

Do I have to use the canvas?

No. Simple view is a plain form — a table of items, your steps, a Run button — and it can do everything the canvas can except draw branches and loops. The two views are interchangeable.

Is my work visible to other workspaces?

No. A workspace has its own keys, flows, runs, Library, members and ceiling, and nothing crosses between them — enforced in the engine, not just the interface. One caveat worth knowing: spend ceilings do not aggregate, so two workspaces at $50/day is $100/day against one bill.

What if a price in the usage screen looks wrong?

An admin can override it in Settings → Usage → Model prices; overrides win and apply within minutes. A model with no price shows $0.00 — that is a gap in the price list, not a free call.

Can I take my workflows with me?

Yes. Export any flow as JSON — without keys, pool bindings or webhook URLs — and import it elsewhere. Outputs are downloadable individually or in bulk from the results grid and the Library.