Skip to main content
The AI Composer turns natural-language prompts into working Data Feeds, Actions, and Studio apps. Better prompts get better results in fewer iterations and fewer AI Credits. This page is a reference for the prompting patterns that work.

The four-part prompt template

Most high-quality prompts have four parts. They don’t have to be in this order, but if you’re stuck, follow the structure:
A real example:
That’s ACTION = Add, SUBJECT = a donut chart of devices grouped by manufacturer, DETAIL = in the top-right; show top 6, rest into Other, CONSTRAINT = brand palette.

High-impact patterns

Specify position

Composer can’t read your mind about layout. Be explicit:

Specify exact data shape

Reference field names, types, and value examples where possible:

Specify failure modes

Tell Composer what to do when data is missing or off:

Reference existing elements via Inspect

For surgical edits, use the Inspect Component button in the Composer footer. Click the cursor icon, click the element in the live preview, then prompt:
Composer captures the element reference and applies the change to the specific instance.

Be explicit about state

If the change should persist:

Patterns by pillar

Data Feed prompts

Compose around: endpoint, auth, pagination, schema.
Pasting API documentation URLs directly into the prompt is gold — Composer reads them and scaffolds the request more accurately than working from memory.

Action prompts

Compose around: destination, inputs, payload shape, auth.

Studio app prompts

Compose around: purpose, structure, data, interactions.

What NOT to put in prompts

Incremental vs big-bang prompting

Big-bang — one long prompt building the whole app: ✅ Faster overall — fewer credits in total ❌ Harder to iterate when one piece is wrong (Composer might rewrite working code) Incremental — small prompts building one thing at a time: ✅ Easier to Restore checkpoint when a step goes wrong ✅ Each step is reviewable in the preview before the next ❌ Marginally more credits if you build a large app Recommendation: Big-bang the first scaffold (one prompt for the whole structure), then incremental for refinement.

When Composer doesn’t understand

Common signals + how to recover:

AI Credits frugality

Composer credits are tracked under Settings → Usage. The per-conversation counter is shown in the Composer header (“X credits used”). Strategies to lower credit burn:
  • Inspect Component for targeted edits instead of describing positions
  • Restore checkpoint before reattempting a bad path (don’t ask Composer to undo; restore)
  • Clear Chat if you’ve gone in circles and want a fresh slate (the app code stays as-is)
  • Big-bang the scaffold then incremental from there

Reusable prompt snippets

Branding

Polish

Distribution

See also

AI Composer

The Composer toolbar, prompts, Inspect, Restore checkpoint

Components

Every component you can reference by name in prompts

Planner

Goal-to-plan generation — produces ready-to-launch prompts

Build your first app

A worked example using these patterns