Browser Actions

Application Runner

Coordinates multiple tools such as Auto Form Fill, Auto Upload, Smart Button Click, Page Understanding, and Detect Form Errors until the task completes or requires manual help.

What It Does

Application Runner is the highest-level browser coordinator. It keeps checking the current page, choosing the next action, and deciding whether the flow has advanced, stalled, blocked, or completed.

How It Works

  • Takes a goal and starts a multi-pass loop.
  • Uses Page Understanding before each action.
  • Uses Page Section Focus to inspect body content as supporting context.
  • Uses Detect Form Errors to understand visible validation failures and blocked states.
  • Sends the current state and error context to an LLM to choose the next action.
  • Typical actions are auto form fill, auto upload, smart button click, or stop.
  • Re-checks the page after each action and classifies the transition as advanced, progressed, blocked, completed, or unknown.
  • Continues until the goal is complete, confidence drops, max passes are reached, or manual correction is needed.

Best For

  • Complex multi-step browser automations
  • Application flows where the next action depends on changing page state
  • Situations where single-step tools need an orchestration layer

Behavior Notes

  • It is the highest-level coordinator rather than a single-purpose page action.
  • It relies heavily on repeated page-state inspection and LLM decision-making.
  • It intentionally stops when confidence drops too far rather than forcing a bad action.
  • It is designed to hand control back for manual fixes when the page becomes stalled or blocked.