Natural-language authoring
Type the goal on the generate screen. You can do this in the desktop app, in the TUI's New tab, or headless with the CLI generate command. The DSL then appears for review.
Use case: Prototyping
Describe it. Review the graph. Press Play.
A local model turns a plain-language request into a Flow DSL graph, rendered on the canvas for review before execution. The translation model produces structured output only. It has no execution authority.
From idea to running flow
The fastest path from intent to execution
Review before run
Generated flows render in a review modal with a plan and static safety checks for destructive operations. Nothing merges onto the canvas until you accept.
Workspace-grounded
Agentic generation inspects your repo or workspace first. Generated nodes then carry real paths and real commands rather than hallucinated scaffolding.
Iterate by re-plan
When a run fails, the monitor proposes a corrective fix built from the failure context and the flow that actually ran. You can accept it, resume, or reject it.
Self-service for non-engineers
Business users type what they need and get the result file. For the routine cases, the queue to a specialist disappears.
Graduate to a template
A prototype that earns its keep becomes a saved template: versioned, shared, scheduled, and reviewed like code.
Case studies
Prompt to payoff
customer Prototyping
Just give me the assessment result
A business or technical-management user who owns an outcome
Type what you need; get the upload-ready file - no mainframe knowledge required.
Request-and-wait-days becomes prompt-to-result in minutes
customer IT & DevOps
The code-validate-run loop
A developer who just edited a JCL job
Stop typing four CLI commands and waiting between each - wire the loop once, run it on every edit.
Four commands and three waits become one click per edit
Your first flow is one sentence away
Generate it, review it, run it - then keep it.