Skip to main content

​
Overview

When your agent analyzes data during a conversation, the charts, tables, and insights it produces are trapped in that chat thread. Apps give those outputs a permanent home — a live, interactive page inside Aster that anyone in your organization can open anytime. Ask your agent to build a pipeline dashboard, a traffic report, an audit summary, or a portfolio overview, and it creates an app at a stable URL you can bookmark, share with your team, and come back to whenever you need it.

​
Getting Started

Enable the tool

Add manage_apps to any agent that you want to be able to create apps.

Ask for what you want

Tell the agent what you’d like to see — a dashboard, a report, a comparison view.

Open the link

The agent returns a /apps/{id} URL. Click it to see your app.
That’s it. The agent handles the data gathering, layout, and design. You just describe what you want.

​
What Can Apps Do?

Apps are fully interactive views, not static images. They support:

Charts and graphs

Line charts, bar charts, pie charts, area charts, and more.

Data tables

Sortable, scrollable tables with formatted numbers and dates.

KPI tiles

Headline metrics with trend indicators and comparisons.

Tabs and navigation

Multi-section apps where you click between views.

Search and filtering

Agents can build search bars and filter controls.

Responsive layout

Apps look good on desktop and adapt to smaller screens.

Branded styling

Agents use the same design system as the rest of Aster.

In-app chat

Let viewers ask the agent questions right inside the app.

​
In-App Chat

An app can include a live chat panel so anyone viewing it can ask the agent a question without leaving the page — ideal for live demos, diligence-style Q&A, and quick follow-ups on what a dashboard is showing. Just ask the agent to add one:
“Add a chat panel to this app so people can ask questions about the portfolio.”
An Ask about this app panel appears in the layout. Viewers type a question and get a streamed answer — with the same tables, charts, tool results, and citations as the main chat.
  • It talks to the app’s own agent (or one you name), so it already knows the data and can use its tools to pull fresh detail on demand.
  • Each visit is a fresh conversation — the panel doesn’t carry history between page loads.
  • It’s for asking, not editing. Use it to explore what the app shows; to change the dashboard itself, ask the agent in a normal chat.
The chat panel is interactive and only works in the live app view — it won’t appear in a static export or PDF.

​
Viewing and Managing Apps

​
Finding Your Apps

Sidebar

Click Apps in the left navigation to see all apps in your organization.

Direct URL

Every app has a stable URL at /apps/{id} that you can bookmark or share.

Control Hub

The Apps tab in Control Hub shows the same listing with admin controls.
All members of your organization can view apps. Admins can delete apps from Control Hub.

​
Locking Apps

By default, any agent in your organization can update or delete any app. Once an app is finished — a board report, a portfolio dashboard, a shared source of truth — you may want to freeze it so it doesn’t get changed unexpectedly. That’s what locking does.
  • Lock it: Ask an agent to “lock this app.” While locked, only the app’s author (the person it was created for) or an organization admin can edit, delete, or unlock it.
  • Everyone else can still view it: Locking only restricts editing — the app stays open and visible to your whole organization.
  • Unlock anytime: Ask the agent to “unlock the app” to allow edits again. You can lock and unlock as often as you like.
  • Unlocked by default: New and existing apps start unlocked. Nothing changes until you choose to lock one.
If someone else’s agent tries to edit a locked app, it gets a clear message that the app is locked and stops — it won’t overwrite your work. Locked apps show a Locked badge in the Apps list and in Control Hub.
Scheduled tasks and other automated runs are treated as trusted admins, so an app’s own scheduled refresh will still update it even when locked. Locking protects against edits by other people, not against an app’s own automation.

​
Keeping Apps Up to Date

Apps show a snapshot of data from when they were last updated. To refresh:
  • Ask the agent: Tell the same agent to update the app with fresh data
  • Automate it: Pair the app with a Scheduled Task — create a task that tells the agent to refresh the app on a recurring schedule (daily, weekly, etc.)
When the agent updates an app, the URL stays the same — users just reload to see the latest version. An app keeps its data separate from its layout, which is what makes updates cheap and safe: refreshing usually means the agent adds or replaces rows in the underlying dataset while the layout code stays untouched. That’s why “add this week’s numbers” or “pull in the new deals” is a quick, low-risk update rather than a rebuild of the whole view.

​
Reliable scheduled refreshes: keep the build deterministic

If an app’s data is refreshed on a schedule, how the data is shaped should live as code inside the app — not as instructions in the scheduled task. This is the single most important practice for a data-backed app that refreshes itself, and it’s why a dashboard that “keeps breaking after the daily refresh” breaks. The reason: an app’s charts read specific fields by exact name. If the recurring refresh re-derives that shape from a long natural-language prompt each day, the agent will eventually emit one field slightly differently — a renamed column, a table that came back in the wrong layout — and that chart silently goes blank even though nothing crashed. Code emits the same shape every time; a prompt re-improvised daily does not. The reliable pattern:
  • The app carries its own build + validate scripts (bundled files, e.g. build_snapshot.py and validate_snapshot.py). The build script is a fixed transform — raw data in, the exact dashboard data out. The validator checks every field the app reads before anything is published.
  • The scheduled task stays dumb. It should say, in effect, “fetch the latest raw data, run the app’s build script, validate, render-check, and publish if it passes” — and contain no field names, calculations, or data rules itself. All of that lives in the app’s bundled code.
  • The agent orchestrates; it doesn’t compute. Each run it fetches the raw source data and hands it to the build script untouched. It only intervenes by hand for genuinely new situations (a new category appears, the source system changes) — and then it updates the app’s bundled script, so the fix is permanent.
If you find a recurring task prompt growing into pages of “make sure field X is named Y, the table must be in this layout, don’t drop section Z” — that’s the warning sign. Those rules belong in the app’s build/validate code, where they’re enforced deterministically, not in a task prompt where they drift. The agent can do this refactor for you: ask it to “move the snapshot-building logic out of the scheduled task and into a bundled build script the task just runs.”

​
Version History

Every update to an app is preserved as a version — nothing is overwritten. Agents can browse an app’s full version history and retrieve the complete source code of any past version. If an update goes wrong, ask the agent to look at the version history and restore an earlier version: “Show me the versions of the pipeline dashboard from before today and bring back yesterday’s.”

​
Use Cases

“Pull our Google Analytics data for the last 90 days and build a traffic dashboard with trends by channel, top pages, and device breakdown”The agent queries your connected GA4 properties, computes the metrics, and publishes a dashboard with interactive charts and tables.
“Build a pipeline overview from Salesforce showing deal count by stage, weekly trends, win rate, and our top 10 open opportunities”The agent pulls CRM data, calculates KPIs, and creates a persistent view your sales team can check daily.
“Show me our Search Console performance — clicks, impressions, CTR by query, and which pages are getting the most visibility”The agent combines GSC data across your properties and builds a multi-tab analytics app.
“Create a QuickBooks dashboard showing cash position, AR aging, and this quarter’s P&L trends”The agent queries your accounting data and builds a financial overview with formatted currency values and comparison periods.
“Research the top 5 competitors in our space and build a comparison dashboard with pricing, features, and recent news”The agent uses web search and scraping tools, organizes the findings, and presents them in a structured comparison view.

​
Advanced: Multi-File Apps

For more complex apps — ones with multiple sections, shared components, or generated data files — agents can use a project-based workflow:

Build files in a workspace

The agent builds files in a workspace using the code execution environment.

Describe the structure

It creates a small config file (aster-app.json) that describes the app structure.

Publish all at once

It publishes the whole project at once.
This supports:
  • Multiple component files — break a large app into organized, reusable pieces
  • Generated assets — include CSV exports, charts, PDFs, and images (up to 100MB each)
  • Data processing — run Python scripts to transform raw data before building the view
  • Iterative editing — load an existing app back into the workspace, make changes, and re-publish
You don’t need to manage any of this yourself — the agent handles the workflow. But if you’re curious or building agent prompts that target this workflow, see the Manage Apps tool reference for details.

​
Combining with Other Features

​
Tips

Be specific about what you want to see. “Build me a dashboard” is okay, but “Build a dashboard showing weekly revenue trends, top 10 customers by spend, and churn rate over the last 6 months” gets much better results.
  1. Iterate. If the first version isn’t quite right, tell the agent what to change: “Move the chart to the top” or “Add a table showing the raw data” or “Use a dark theme.”
  2. Bookmark your apps. The URL is permanent — save it, share it in Slack, add it to your daily routine.
  3. Set up auto-refresh. For dashboards you check regularly, create a scheduled task that updates the app so the data is always fresh when you open it.