Introduction

A workflow is a list of steps you write once and run anywhere. Each step runs one component, an AI model call, an HTTP request, a file or data operation, and passes its result to the steps that follow. Wire the steps together and the engine runs them in order.

Automate everyday operations that run through APIs, files, and online services, with a model in the loop wherever something needs to be read, judged, or written. Because every step is static configuration, we can see exactly what ran, what each model was asked, the answer and what it cost, so there is a clear account of every decision an automation made on your behalf. Steps also nest, so a loop, a branch, or a whole other workflow is itself just one step, which keeps a large automation modular instead of sprawling.

The same definition reads clearly to people and to language models, and it runs unchanged from the command-line app, the REST API, or the web app. Under the hood the engine compiles each workflow into a static, inspectable plan, then executes it with conditions, loops, retries, and caching, so an automation stays predictable even with a model in the loop.