Write workflows as files — nodes, prompts, experiments, and the CLI that checks and runs them.
This is the file-based way of building an AQVEN workflow. A flow is a directory in your project — a
flow.yaml plus one YAML file per node — and prompts live in their own Markdown files, never inline
in the YAML. You edit these with a text editor and a terminal instead of a browser; it’s the same
flow definition Studio edits visually, just from the other side.
The sidebar groups this area into four parts. Node kinds is one how-to page per kind of step. Start with how to call a model and how to write a step in Python — the two you’ll use most. The other eight cover giving an agent a tool, pausing for a person, running steps in parallel, mapping a step over a collection, routing by a value, repeating a step with a limit, reusing a flow as a step, and narrowing a dynamic value to a type.
Prompts & data shape covers the three things every node’s inputs and outputs lean on: how to write a prompt (the Markdown files a model call reads from), how to constrain a field’s values (length, range, pattern, and choice limits a model’s answer has to obey), and how to handle a shape you don’t know in advance (for a field whose type isn’t fixed until run time).
Checks & experiments is how you find out whether a flow works reliably, not just on the case you tried. How to keep case media as files in the project puts the images, audio, video and documents of dataset cases in git next to the dataset. How to write an experiment puts one question in a file before any data. How to run a series answers it on many cases, from Studio, the terminal or an agent. How to read a series says what the verdict means and what to do next. How to write a custom evaluator adds a check no built-in can express.
CLI tooling is everything you run from a terminal instead of Studio. The check
command validates the whole tree before any of it reaches a teammate or a release. The
rest cover generating types and editor schemas, inspecting how a project’s pieces connect, running a
flow locally without a server, managing secrets, and checking your model providers are configured.
aqven series runs a series from the terminal and waits for its verdict: see
How to run a series.