Skip to main content

What it does

The Manage Apps tool lets agents create and manage persistent interactive pages inside Aster. When a user asks for a dashboard, report, or data visualization, the agent can build one that lives at a permanent URL — not just a one-time chat response. See Apps for the full user-facing feature overview.

When to use it

Add this tool to agents that:
  • Analyze data from integrations (GA4, Salesforce, Snowflake, etc.) and need a way to present results beyond chat
  • Build recurring reports that users want to bookmark and revisit
  • Create multi-section views with charts, tables, KPIs, and tabs
  • Need to update an existing view with fresh data

Parameters

Actions

create

Build and publish a new app in a single call. Required: name, source_tsx. Optional: description, snapshot_data.

update

Update an existing app’s content, data, or metadata. A new version is created internally — the URL stays the same. Required: app_id. Optional: name, description, source_tsx, snapshot_data.

list

Returns all apps in the organization.

get

Returns an app’s metadata and current source code — but not its snapshot data. To read or edit an app’s data (or faithfully rebuild it), use load_to_sandbox, which loads the full project (including data/snapshot.json) into the sandbox where execute_python can read it. Required: app_id.

delete

Removes an app. Required: app_id.

set_lock

Locks or unlocks an app. While an app is locked, only its author or an org admin can update, delete, or unlock it — any other agent that attempts update, delete, or set_lock gets an error and should tell the user the app is locked rather than retry. Apps are unlocked by default. Only the app’s author or an org admin can call set_lock. Required: app_id, locked (true to lock, false to unlock).

load_to_sandbox

Loads an existing app’s source, data, and assets into the code execution sandbox for editing. After loading, the agent can use execute_python to modify files, then call publish_from_sandbox to publish the changes. Pass a version_id to load a historical version instead of the current one — this hydrates that version’s source and its snapshot data and assets, which is the full-fidelity restore path: load the old version, then publish_from_sandbox to republish it as a new version. Required: app_id. Optional: app_dir (defaults to /home/user/aster-apps/app-{app_id}), version_id (defaults to the current version).

publish_from_sandbox

Reads an app project from the sandbox, validates it, compiles it, and publishes it. An aster-app.json manifest is optional — omit it and the entry is inferred (preferring src/App.tsx) with an empty snapshot. Required: app_dir. Optional: app_id (omit to create a new app, provide to update an existing one). The response classifies the publish from facts the server already holds, so the agent doesn’t have to judge its own change:
  • changed_code — whether the compiled bundle differs from the current version. false means a data-only publish, so the rendered layout cannot have changed and no render check is needed.
  • snapshot_diff — an advisory shape diff against the previous snapshot, flagging dropped, renamed, reshaped, or emptied keys. This is the most common cause of a tab going silently blank after a refresh.
Before creating an app or changing its code or layout, the agent loads the building-aster-apps skill — it covers project structure, the snapshot data contract, and a render check that catches blank or crashed tabs before publishing. A data-only refresh (rebuilding data/snapshot.json with the source untouched) skips the skill.

list_versions

Returns an app’s version history — every publish creates a version, and all of them are preserved. Each entry includes the version id, when it was published, whether it’s the current version, the bundle hash, source file list, and asset count. Returns the 50 most recent versions; older versions remain fetchable by id via get_version. Required: app_id.

get_version

Returns a specific historical version’s complete source code — the entry component plus the full source tree, but not its snapshot data. Use it to see how an app’s code changed between versions. To restore a historical version, use load_to_sandbox with the version_id instead — for data-driven apps, versions often differ only in snapshot data, so a faithful restore needs the data, not just the source. Required: app_id, version_id.

Available libraries

Apps can use these libraries in their source code: All Tailwind CSS classes are supported, including arbitrary values.
useAppData().data is always present — no loading/null guard needed. Still guard arrays before .map/.reduce (e.g. (data.rows || []).map(...)) for empty refreshes.
Every published app has a native Export button (top of the app) that produces a PDF or PNG — the current view, or, for tabbed apps, all tabs combined into one PDF. Don’t build your own PDF / print / download-image logic into an app — the platform handles it.For tabbed apps, build the tabs with the @aster/ui Tabs component (not hand-rolled useState switching) so the all-tabs export can capture each tab; custom tab bars only export the view that’s currently visible.
An app’s query string belongs entirely to the app, so an agent can link a user straight to a particular view — https://www.asteragents.com/apps/216?region=West&tab=2. Read it with useAppParams():
setParams merges into the URL, so a user who filters in-app gets a link they can share — and the Export button produces a PDF of that same filtered view. Agents can export a filtered view directly via the app_params parameter.
Use the hook, never window.location.search. A direct read works in the viewer but renders the app’s unfiltered default in PDF/PNG exports. Params filter within the published snapshot — they can’t fetch different data.

Sandbox project workflow

For complex apps with multiple files, generated assets, or iterative editing:
  1. Agent uses execute_python to build files in /home/user/aster-apps/{slug}/
  2. Agent writes an aster-app.json manifest:
  3. Agent calls manage_apps with action: "publish_from_sandbox"
To edit an existing app: call load_to_sandbox first, modify files with execute_python, then publish_from_sandbox.

Common use cases

Dashboard from live data

“Pull our GA4 traffic data and build a dashboard with trends by channel, top pages, and device split”

Updating an existing app

“Refresh the pipeline dashboard with this week’s numbers” The agent re-queries the data source and calls update with new snapshot data.

Complex multi-section app via sandbox

“Build a detailed analytics app with separate sections for overview, queries, pages, and opportunities — include downloadable CSV exports” The agent builds multiple component files and data assets in the sandbox, then publishes the whole project.