# Embedded agents: launch AI agents from buttons in your apps

> Apps can declare agents anyone using the app launches from its interface, each a full agent workflow with its own sandbox, history, and approvals.

Aug 12, 2026. Topics: Apps, Agents. https://www.generalinput.com/changelog/2026-08-12-agents-embedded-in-apps

Apps can now declare background agents that someone using the app launches from a button: a Research button on each row of a lead table, a Draft reply button on each support ticket. An app agent is not a reduced version of a workflow agent. It follows the same agent workflow spec, runs in its own sandbox, and gets the same run history, streaming, approval steps, and billing. Ask for one in the chat panel next to your app, the same way you'd ask for a new chart. Read the [launch post](/blog/agents-inside-your-apps).

### How it works

- **Same files as an agent workflow.** Each agent lives in the app bundle with its inputs (the button's variables), model and time budget, and a written procedure.
- **The interface is the trigger.** App agents have no triggers of their own, and their connections must be the app's own, so the Integrations tab shows which agent uses each.
- **Owner's access, viewer's identity.** Anyone the app is shared with can launch an agent; it runs through the owner's connections and knows who launched it.
- **Guardrails.** Every launch is confirmed, since runs spend credits, and an app can run at most 25 agents at once. Agents write results back to the app's data.
- **Tested by the builder.** The app-building agent runs its own agents and reads their traces before handing the app over.

### Also shipped

- App containers ship with more libraries preinstalled, from data tables to Excel, Word, and PDF generation.