Instead of assembling nodes one by one, describe the scenario and let the model write the first draft. Requires a configured AI provider, whether Anthropic, any OpenAI-compatible endpoint, or a local model.
Generating
-
Choose File → Generate Flow with AI… (or run Generate flow with AI from the command palette, ⌘⇧P).
-
Describe what you want. Specific beats vague, so name the steps, what to capture, what to assert:
Create a user CRUD flow: register, log in, capture the token, then create, fetch, update and delete a user, asserting the status at each step.
-
Click Generate or press ⌘⏎.
The result is a complete, schema-valid .aether file written into your workspace (with a
collision-safe name if one like it exists) and opened in the editor, a normal file from the
moment it lands. Nothing about it is special: run it, edit it, commit it.
Treat it as a draft
Generation gets the structure right: the chain of requests, captures wired between steps, assertions in place. You bring the specifics:
- Swap example URLs for your real ones, ideally
${{ variables.baseUrl }}and an environment. - Check the assertions actually assert what you care about.
- Run it and walk the failures with the usual debugging tools. The generated flow debugs exactly like a hand-written one, because it is one.
If generation fails
The error appears in the modal. The usual causes: no provider configured (Settings → AI provider), or a network problem reaching the endpoint. If you’re pointing at an OpenAI-compatible endpoint, Test connection in Settings tells you which of the base URL, key, or model id is at fault. Smaller local models fail generation more often than frontier ones; flows are schema-validated before they’re written, so a bad draft is rejected rather than saved.
Next: AI chat & ai-action nodes, or Connect an AI provider if generation is failing.