Data Apps and Custom Dashboards Come to Looker (and Wire)
Google released Looker Data Apps into preview in October 2026, a new feature that lets you build interactive, freeform pages on Looker’s semantic layer, authored using an AI Agent. Each app is published into Looker as a custom web application, leveraging the same security and access controls as your dashboards, looks and self-service explores.

In this blog I’ll explain What a Data App is and how it differs from a conventional Looker dashboard, how you build one using Google’s AI toolkit and how we use Wire, our plugin for Claude Code, to create Looker Data Apps as part of an end-to-end delivery workflow.
What is a Looker Data App?
Data Apps in Looker are a way to create custom, bespoke web applications that are hosted in your Looker instance and get their data, and access controls, via Looker’s semantic model.
At their most basic level, they’re a way to develop custom dashboards using an AI agent that gives you more freedom over the interface while keeping queries governed by LookML, but as custom web apps they can also be extended to write as well as read data and trigger actions in your business applications.
Looker Data Apps use JavaScript, HTML and CSS to create fully customised interfaces, including tabs, sidebars, filters and interactive controls. Google’s Lit runtime provides built-in components for displaying KPIs, charts and tables, all connected to live Looker data.
And rather than writing dashboard code manually, you describe the app you want to an AI coding agent such as Claude Code, Gemini CLI or Cursor. Google’s Data Apps toolkit helps the agent explore your LookML model, generate the application, preview and verify it, then publish it to Looker with your approval.
How are Looker Data Apps Hosted?
Publishing a data app creates an inline extension and a dashboard in Looker, then connects them as shown in the diagram below.

- The publisher saves the JavaScript in the inline extension’s code field using POST /data_app_extension.
- Its single element references the extension through extension_id: “db::<uuid>”.
- The dashboard uses the extension layout, displaying the element full screen with Looker’s header and the tile title hidden.
- The publisher reads the file as text, stamps an ID on every widget and saves its query using POST /queries. It stores the widget-to-query mapping in the element’s rich_content_json, enabling Explore from here for individual widgets.
Looker data apps rely on a runtime hosted separately to your Looker instance at www.gstatic.com, containing libraries such as data-app.js, Lit, Highcharts and AG Grid. Each published app stays pinned to the runtime version used when it was published, so that they don’t break as the runtime is updated over time.
Building a Data App using Google’s AI Toolkit
Looker Data Apps have three pre-requisities, the first two of which require an admin to enable and the third is something you can do as a developer once the first two are in-place:
- Your Looker instance has to be Google (rather than customer)-hosted and on 26.18 or later, and your Looker admin has to switch on Data Apps under Admin > General > Previews.
- Your user has API keys enabled (an admin setting on your user page) and the dashboard author permissions:
access_data,explore,see_user_dashboards,save_contentandsave_dashboards, plus Manage Access, Edit on the folder you will publish into. - You have created an API key (Client ID, Client Secret) under Admin > Users > your user > API Keys.
On the machine you’ll use to author the data app, you’ll need Python 3.9 or later, and macOS, Linux or Windows with Git Bash. Node is optional and the toolkit’s syntax check uses it when present.
To install the Data Apps toolkit, run the following command using Terminal:
bash <(curl -fsSL https://dl.google.com/looker-data-apps/install.sh)
Once the install script has run, the install directory and subdirectories look like the layout below, with the skill files under /skills/ containing the key instructions the toolkit uses to do the actuall app authoring.
~/looker-data-apps/
├── apps/ your apps, one folder each, each holding app.js
├── skills/
│ ├── data-app-creator/ the authoring skill, its tools, recipes and the component reference
│ ├── data-app-previewer/ start-preview.py and stop-preview.py
│ ├── data-app-publisher/ publish-data-app.py
│ ├── data-app-upgrader/ converts older HTML canvas dashboards
│ └── toolkit-updater/ checks for and applies toolkit updates
├── runtime/ data-app.js and data-app.css, the Lit component library
├── preview-server/ the local preview server and app gallery
├── vendor/ Highcharts, AG Grid, Lit and the Looker extension SDK
├── AGENTS.md routing for the agent: which skill to load for which task
└── data-apps-config.json your instance, API key and publish folder (keep it local)
Bootstrap also symlinks the skills into .claude/skills/ inside the install, so Claude Code picks them up when you open that folder. Reinstalling replaces everything except apps/ and the config files, and removes files an older version shipped.
Then, using Claude Code for example, open the toolkit install directory and describe what you want. The data-app-creator skill it contains requires you to specifiy four things in your prompt:
- the data (which model and explore),
- the design direction (a theme, a palette, a reference image if you have one),
- the components (KPIs, charts, filters, a table)
- and the story (who reads it and what for).
Build a web traffic app on the analytics model's web_sessions_fact explore for the marketing team, with:
- A KPI strip for sessions, unique users, page views, bounce rate and meetings booked,
- a weekly trend, sessions by channel and by device,
- the top landing pages as a table, and the top referrers.
- Dropdown filters for channel and site.
- Light theme, teal accent.
The skill then tells the agent to show something quickly and let you fine-tune it from there. A local-running preview server lets you preview and iterate on your design before you publish it for wider use on your Looker instance.

Then it’s a case of iterating on that design, fine-tuning the selection of columns and visualizations the data app contains, adding and removing tabs and then finally telling the toolkit to publish the app so that it runs, hosted in Looker, using a dashboard ID the same way as your user dashboards (though it’s not editable or copyable as regular Looker dashboards are).

Adding the Data App Delivery Workflow using Wire
Whilst Google’s toolkit handles the authoring part of building a Looker Data App, there’s obviously still work you need to do before you create the app in-order to ensure it’s actually what you’re looking-for and provides value to your users.
Wire, the agentic framework we use to deliver data platform projects using coding agents such as Claude Code, adds those steps into the data app development workflow from agreeing requirements and content to creating initial mockups, validating the app and reviewing its design through to publishing it for use in Looker.
If you’re interested in reading more about how this part of the process works, The Dashboard Extension tutorial in the Wire docs walks through these earlier requirements and mockup stages, along with installation and getting-started steps.
Then its just a case of telling Wire, through a Claude Code session, that you want a Looker data app created and to do-so using specs and definitions you created earlier using Wire, and it’ll take care of the end-to-end process, checking-in with you as questions come-up, like this:
> Build the Web Traffic data app for the marketing team on the Web Analytics explore: the approved mockup, three tabs, the filters in the spec.
Release 03-web-traffic-app (dashboard_extension), director: Mark.
Requirements approved 24 Oct, mockups approved 28 Oct: catalog 17 rows, spec
3 filters, tabs Overview / Pages / Engagement, presentation data_app.
Runnable now: data_app-generate.
Running /wire:data_app-generate 03-web-traffic-app
Toolkit 1.2.0 at ~/looker-data-apps (pinned 1.2.0), config present,
credentials OK: 18 models visible.
Explore analytics / web_sessions_fact: 177 dimensions, 22 measures.
Plan: 17 widgets from 17 catalog rows, 0 unmapped, 0 unresolved,
3 filters resolved.
Written: data_apps/web-traffic/app.js (17 widgets, tabs, drawer, two
dropdowns, metric switcher, top-N slider, search box).
Rules: 0 errors, 0 warnings.
Google's verification: syntax ok, sandbox ok, 19 queries dry-run on
Looker, 0 failed, 0 empty.
Preview: http://localhost:8765/preview-server/?app=apps/web-traffic/app.js&sandbox=true
status.md: data_app.generate = complete.
Open the preview and tell me what to change, or say validate.
You look at the initial preview and spot something that needs changing, before the app is ready for publishing.
> Top Referrers shows google.com at 128,000 sessions. The page total is 16,593.
That chart carried its own filter (referrer not null) inside its query, which
replaces the page's filters, so the 90-day window never reached it. Moved it
through the shared filter function. Re-verified: 19 queries, 0 failed.
google.com is now 727 for the window. Nothing else changed.
> Validate it.
Running /wire:data_app-validate 03-web-traffic-app
1 Target and toolkit PASS
2 Every catalog row produced a widget PASS 17 of 17
3 No unmapped chart types PASS
4 Every field resolves PASS
5 Model and explore as recorded PASS
6 Every spec filter present PASS 3 of 3
7 Wire's rules PASS 0 errors
8 Google's verify, stages 1 to 3 PASS 19 queries
9 Toolkit copy is the source PASS
10 No literal data PASS
11 Titles match the catalog PASS
12 Published copy is this source n/a not published yet
Status: PASS. Report: dev/data_app/web-traffic/validation.md
status.md: data_app.validate = pass.
Next is the stakeholder review. Who is reviewing, and where: the preview, or
shall I publish to a review folder first?
> The marketing lead reviewed it on the preview this morning and approved it.
Publish it to the Shared folder.
Running /wire:data_app-review 03-web-traffic-app
review = approved, reviewed_by: marketing lead, 1 Nov 2026.
Running /wire:data_app-publish 03-web-traffic-app
Target: rittman.eu.looker.com, folder 1 (Shared), title "Web Traffic".
First publish of this app.
Publish / Cancel?
> Publish.
Dashboard 419 created. App stored as its extension. 19 widget queries
registered, so "Explore from here" works on every widget but one: the
drawer's breakdown chart, whose measure depends on the card clicked.
https://rittman.eu.looker.com/dashboards/419
Dashboard id written into data_apps/web-traffic/app.js, committed on
feature/03-web-traffic-app, pull request opened for technical review.
status.md: data_app published, dashboard 419.
Done. Three things to know: a Data App cannot be embedded, printed or
scheduled; do not copy the dashboard in Looker, because copies share one
app; re-publishing from any machine updates 419, not a new dashboard.
Under the covers, we’ve added four new commands into Wire to support authoring Looker (and in-future, other types of) Data Apps, including a new “publish” step specific to data apps. Wire will take-care of running these commands for you, or you can type them into Claude Code directly as slash-commands.
/wire:data_app-generate <release>builds the app from the approved mockup chain and the explore, checks it, and starts the local preview/wire:data_app-validate <release>checks the app against the design and the semantic layer, and runs Google’s own verification; writes a PASS or FAIL report/wire:data_app-review <release>records the stakeholder’s decision/wire:data_app-publish <release>shows where the app will go, asks, publishes it, and records the dashboard it became
You can invoke these “slash” commands directly within a Claude Code session to build your Looker data app, or as in the example above you can just describe the app you want and point it to specs you created using Wire earlier-on.
Either way, the result at the end is a properly-specified and designed Looker Data App that’s gone through a governance and release process, aided and accelerated by an AI agent and our agentic delivery process.

Interested? Find Out More!
Rittman Analytics is a boutique data analytics consultancy that helps ambitious, digital-native businesses scale-up their approach to data, analytics and generative AI.
We’re authorised delivery partners for Google Cloud along with Oracle, Segment, Cube, Dagster, Preset, dbt Labs and Fivetran and are experts at helping you design an analytics solution that’s right for your organisation’s needs, use-cases and budget and working with you and your data team to successfully implement it.
If you’re looking for help with data and AI needs, or would just like to talk shop and share ideas on what’s going on in your organisation and the wider data analytics world, contact us now to organise a free, no-obligation call. We’d love to hear from you!