Workflows
A single capability answers a single question. A workflow chains capabilities and decisions into a business process: score the claim, and if the risk is high, extract the details and park it for a human to approve. Workflows are built and run inside Pradra Studio, with every run recorded step by step.
What a workflow is made of
A workflow is a graph of steps:
| Step type | Does |
|---|---|
| Capability | Calls one of your deployed capabilities and passes its output onward. |
| Condition | Branches on a value ("risk > 0.8 → this path, otherwise → that path"). |
| Transform | Reshapes data between steps so outputs fit the next input. |
| Approval | Pauses the run until a person approves or rejects — with a time limit. |
| External | Hands off to an outside system. |
Validated before it can run
A workflow version is checked when you save it: the graph must be acyclic, and every step's inputs must be satisfiable from what earlier steps produce — contract-checked, like everything else. A version that passes is valid and runnable; one that fails is marked invalid with the reason, and cannot run until fixed. You find broken wiring at design time, not mid-run at 2 a.m.
Runs are durable and inspectable
- Each run executes step by step, recording every step's input, output, and status as it goes.
- A run waiting on an approval parks without burning resources and resumes when the decision arrives (or expires at the deadline you set).
- Failed steps are retried safely — a step that already completed is never double-executed.
- The run detail page shows the whole trace: what ran, what each step saw, where it branched, who approved what.
Working with workflows in the Studio
- Go to Workflows in the sidebar and create one.
- The editor is where you assemble steps and wire outputs to inputs; saving validates the version.
- Start a run with its initial input; the runs list on the workflow's detail page tracks each one live.
- Open a run to inspect the step-by-step trace — including any step's exact input and output.
- Editing the workflow creates a new version; in-flight runs finish on the version they started with.
Where workflows fit
Use a workflow when the process matters: multi-step decisions, human checkpoints, branching. For a single prediction, call the capability directly. For open-ended goals where the AI chooses its own steps, see Agents — workflows are deterministic by design; agents are not.
Next
- The non-deterministic sibling: Agents.