Skip to main content

Generate with AI as code

The FluidGrids AI can draft a workflow from a description. With workflows as code, the draft is also available as TypeScript or Python, ready to drop into a project.

In the app​

  1. Open the AI assistant in Workflow mode and describe the automation, or ask it to change an existing workflow.
  2. The draft panel shows the proposal with Canvas, TypeScript and Python tabs.
  3. Open a code tab to read the workflow as graph-builder code, with its fidelity report. Download saves the files.
  4. Apply the draft to create the workflow as usual, or take the code into your project and push it from there.

From the CLI​

fluidgrids generate "When a lead arrives by webhook, enrich it and notify #sales if the company has 500+ employees" \
--as ts --out workflows/
  • --as ts or --as py chooses the language (the project language by default).
  • --out chooses the folder the files are written to (the project root by default).
  • --import also creates the workflow in your workspace as a draft, straight from the generated code, which proves the code turns back into the same workflow. --key and --name set its key and name.

Why the code is always valid​

The model never writes code directly. It produces a structured draft, FluidGrids builds a workflow definition from it and checks it like any other definition, and the code is then generated from that definition by the same exporter that Export as code uses. So the code always builds, and its fidelity report shows that it turns back into exactly the drafted workflow.

For the same reason, the code is always graph-builder (tier 1) code that describes a canvas workflow.

Credits​

Generating a draft uses AI credits exactly as it does without code. Turning the draft into code is free. When code output is not available to your workspace, asking for it is refused before the model runs, so no credits are spent.

Next steps​