跳转至

Task Overview

The Tasks page provides three main entries: Quick tasks, Scheduled tasks, and Notification-triggered tasks. A common workflow is to try a task as a quick task first, then turn it into a scheduled or notification-triggered task after it works.

Task execution flow

The following diagram shows the general flow from task start to finish. As a user, you do not need to memorize every internal state. The important idea is: AutoLXB prepares the device and task context, tries to reuse a task route for stable page navigation, then uses the vision model for dynamic UI decisions.

Task state machine

In plain words:

  1. Preparation: read device state, app state, input capability, and runtime context.
  2. Route check: if this task has a saved route, AutoLXB tries to use it first.
  3. Normal execution: if there is no route or the route cannot be used, AutoLXB analyzes the task, opens the target app, and navigates normally.
  4. Visual execution: when the current page must be understood or operated, the vision model observes the screen and chooses actions.
  5. Finish or fail: after completion, the result is recorded. If it fails, inspect Trace for the reason.

Which task type should I use?

Task type Best for Example
Quick task Temporary execution, trial runs, model testing "Open WeChat and send hello to File Transfer."
Scheduled task Run at a fixed time or repeatedly Open an app and check in every morning at 9:00.
Notification-triggered task Run after a matching notification arrives Reply after receiving a message from a specific chat.
  1. Try it as a quick task first: confirm that the model understands the task and the phone can click, swipe, and input normally.
  2. Automate it after it is stable: create a scheduled task or notification-triggered task.
  3. Save a route after a successful run: route reuse can reduce model calls and improve repeatability.
  4. Use Trace when something fails: Trace shows which phase failed and why.

Task routes in one sentence

A task route is a reusable local path generated from a successful task run. Later runs of the same task can replay the route first, and only fall back to visual execution when the route cannot finish the job.