Skip to main content

Agent Mini-Apps

A mini-app is a standing dashboard your agent builds once and you keep — a morning brief, a project status board, a finance overview, a home-control panel. Unlike an artifact that lives in one conversation, a mini-app is pinned to the agent itself: one tap away in every chat, on every visit.

Pinning one​

Just ask — "pin a morning-brief dashboard as a mini-app". The agent designs the page, pins it, and it appears behind the apps button next to the chat input — and greets you automatically when you open the agent.

With a single mini-app pinned, it becomes the agent's home page: open the agent and the dashboard fills the view — no tabs, just its board. Pin more and they organize into reorderable tabs, first tab on top. Above it all, the agent home also shows your live sessions — chats generating and background tasks running right now, one click from any of them.

On a personal agent your mini-apps are yours alone. On a shared agent the team can share one set — everyone sees the same board.

Kept fresh by your agent​

Mini-apps pair naturally with scheduled tasks: a 7 a.m. task rebuilds the morning brief, and any open tab refreshes live the moment it's updated. You can also just ask the agent to change something — same dashboard, new content, everywhere.

Buttons that do real work​

Mini-apps can carry action buttons — but only actions the agent declared and you approved. When an app declares actions, OtoDock shows you exactly what each button will do before any of them work:

  • Run a task — fire an existing task on the spot: refresh the data, kick off the render, run the daily check. Buttons can fill in declared parameters, too — a "Analyze this month" button that passes the month you're looking at.
  • Message the agent — send a pre-approved prompt into your current chat (or start one), optionally filled with values from the page: Analyze {{month}} becomes "Analyze July".
  • Call a tool directly — for instant controls: toggle the lights, refresh a number, flip a switch. The button runs one declared agent tool with pre-set values — no agent turn, no waiting for a reply — and the result comes straight back to the page. Anything the page fills in is checked against the exact form you approved.

Change the app's actions and the approval resets — buttons stay off until someone with the right access reviews them again. Approving a shared app's task buttons requires the access to run those tasks yourself, so a button never does anything its approver couldn't.

The same safety net as artifacts​

Mini-apps render in the same fully sandboxed frame as artifacts: no access to your session, no free-form calls into the platform — declared, user-approved actions only, delivered with clear provenance and rate limits.

tip

A great first mini-app: ask your agent for a daily brief with a "Refresh now" button wired to the task that builds it.

Next steps​

  • Artifacts → — the in-chat sibling: live, interactive UI inside a conversation.
  • Tasks → — the scheduled work that keeps mini-apps fresh and powers their buttons.
  • Agents → — personal and shared agents, and what that means for who sees an app.