# Data flows A **flow** moves data from somewhere into somewhere useful. You point it at a source, review the shape SnoutData proposes, choose a destination, and run it. The flow is saved, so you can run it again, put it on a schedule, and see what every run did. Flows are how data gets *into* your databases. The rest of SnoutData is about working with data you already have; this is the way in. ## The shape of a flow Every flow is the same three parts: 1. **A source.** A file, a PDF, a web page, another database, or a model asked to generate records. See [sources](https://docs.snoutdata.com/flows/sources). 2. **Transforms.** Typed columns, renames, filters, deduplication and redaction, all optional. See [transforms](https://docs.snoutdata.com/flows/transforms). 3. **A destination.** A SQL table, a MongoDB collection, a file, a fine-tuning set or a vector index. See [destinations](https://docs.snoutdata.com/flows/destinations). ## Start one Drag a file onto the SnoutData window, or use **File, New Data Flow**. The wizard opens with whatever you dropped already parsed, typed and shaped, so the first thing you see is a proposal to review rather than an empty form. Nothing is read until you ask for it. You can drop several files, of different kinds, in as many goes as you like, and then read them all in one flow: the columns are unioned and every record carries a `_source` telling you which input it came from. ## Nothing happens until you say so Before a flow runs you see the columns it proposes, the reasons for them, the exact statements it will execute and where the data will go. A flow does not act on an assumption you were not shown. ## Where flows are saved Flows live on this computer, with your other SnoutData data. A flow is a saved definition, not a script you have to keep somewhere: you can reopen it, edit it, run it by hand or [schedule it](https://docs.snoutdata.com/flows/schedules-and-runs).