Pay for steps, not for servers.

No. Workflows are ordinary functions in your codebase. You wrap the calls you want to make durable in step.run and deploy as usual. Most teams move one flow in an afternoon.

Each step.run, sleep and waitForEvent is one step. Retries of a failed step are free, sleeping costs nothing, and steps restored from memo during a replay are never billed.

Step inputs and outputs are stored in the region you pick (us-east, eu-west or ap-south) and encrypted with per-workspace keys. Enterprise plans can run the whole runtime in your own cloud account.

A queue delivers messages. Orrindel keeps the state of the whole run: which steps finished, what they returned, what is waiting and why. That is what makes retries, waits of several days and replays possible without extra code.

Yes, and you decide how carefully. Mark a tool as needing approval and the run pauses for a person. Mark it idempotent and a retry can never run it twice.

Runs pause and resume where they stopped. Nothing is lost and nothing runs twice. Our status history and incident write-ups are public.