Skip to content
Open
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
15 changes: 15 additions & 0 deletions guides/data-apps.mdx
Original file line number Diff line number Diff line change
Expand Up @@ -44,8 +44,19 @@ The more context you give the agent, the better the result. You can attach sever
- **Charts** - pick existing saved charts to give the agent a working metric query it can reuse. The agent gets the structure of the query (the dimensions, metrics, filters, and table) so it knows exactly which fields to use.
- **Dashboards** - attach a dashboard to pull in every query on it at once.
- **Images** - upload screenshots, mockups, or design references. Images are one of the fastest ways to nail the visual style you want.
- **Screenshots of the running app** - the preview panel has a **Screenshot** button that captures the current state of the running app and attaches it as context to your next prompt. Use it when you want the agent to react to what you're looking at right now ("make this header smaller", "the spacing here is off") instead of describing it from scratch.
- **Sample data** - each chart and dashboard chip has an **Include sample data** toggle. When you turn it on, Lightdash runs the query and passes the first 10 rows to the agent so it can make better decisions about formatting, content, and copy. Sample data is opt-in per resource because rows can be sensitive - the agent will see whatever you send it.

### Choosing a model

Next to the send button, a model picker lets you choose which Claude model runs the generation:

- **Sonnet** (default) - balanced quality and speed; the right choice for most apps.
- **Opus** - highest quality, best for complex apps or when you want the agent to make more nuanced design decisions. Slower.
- **Haiku** - fastest, useful for small tweaks where you don't need the full reasoning of the larger models.

You can change the model on every prompt - including mid-iteration. The agent keeps its memory of what it already built, so switching models doesn't reset the app.

### Iterating on your app

Every prompt you send creates a new version of the app. The agent keeps its working state between versions, so follow-up prompts like "make the header smaller" or "swap the bar chart for a line chart" land as targeted changes instead of full rewrites.
Expand Down Expand Up @@ -172,6 +183,10 @@ To promote, you need:
- **Create** permission for data apps in the upstream project.
- The preview project must have an upstream project configured under **Settings → Data ops**.

<Info>
**Data apps in new previews.** When a preview project is created from an upstream production project, every data app in production is duplicated into the preview - so you can open, iterate on, and promote those apps from inside the preview. Iterating on a duplicated app and promoting it updates the original production app rather than creating a new one.
</Info>

### Promoting a dashboard with data app tiles

If you embed a data app in a [dashboard](/guides/how-to-promote-content#promote-dashboards) and promote the dashboard, every referenced app is promoted alongside it. Each app is created or updated in the upstream project using the same rules as standalone app promotion, and the tile is remapped to point at the upstream app so the production dashboard renders the production copy.
Expand Down
Loading