Skip to main content
DocsRunManual Run

Execute Workflow Manually

Click Run in the editor, or run in bulk from the list—and read the live event stream on the right.

Manual execution is the most basic way for you to run your workflow - development and debugging, one-time business requirements, and regular and unscheduled reports all use this entrance. This article explains clearly "where to click, what each item in the dialog box is, and what to see after running."

Three Manual Execution Entrances

1. Click "Execute" in the Editor

Most commonly used. Suitable for you who are changing a workflow and want to try it out immediately.

2. Click "▶ Execute" on the Workflow List Card

Suitable for a workflow that is already stable and you don’t want to open the editor.

3. Batch Execution

Check multiple items at the top of the workflow list and click "Batch Execute". In the pop-up window:

  • Fill in the variables separately for each workflow (or use the default values directly)
  • Select the number of concurrency (1–10) to control how many lines are opened at the same time
  • Select failure strategy: whether one failure will interrupt other

Execution Dialog Breakdown

No matter which entry is entered, after clicking "Execute", a dialog box with the same structure will pop up:

Parameter (Variables) Area

Automatically display each parameter by workflow.variables:

  • Prefill with default value
  • Required fields are marked with a red asterisk
  • Hover with description can see the explanation
  • Enumerated types use drop-down menus
  • Array/object types edit JSON directly

Dry-Run Switch

It is highly recommended to check the box on your first run. See Quickstart above for details.

Trigger Source (Debug)

Default manual. If you want to simulate "timed trigger" or "Webhook trigger" for debugging, you can switch - the condition of some steps may depend on this.

Recovery Switch (Pro+)

When the workflow declares recovery.autoResumeOnRestart, there is an additional toggle in the dialog box "Whether automatic continuation is allowed for this execution". See details Auto-Resume.

Debug Switch (Pro+)

After checking, this execution will enter debug mode - you can set breakpoints, single-step execution, view variables, and modify variables. See details Breakpoint Debugging.

Behavior After Startup

After clicking "Start Execution":

  1. The front end immediately creates an execution record (status PENDING), and the page automatically jumps to the execution details.
  2. 平台调度器拉起执行（通常 < 200ms 内从 PENDING 变 RUNNING）。
  3. The page establishes an SSE connection, and each event is pushed to refresh the UI in real time.
  4. The "running step" in the canvas is highlighted; completed steps are highlighted in green, and failed steps are highlighted in red.

Live Event Stream on the Right

The right column of the execution details page is a "log + event" mixed stream. Important event types:

  • execution.started / execution.finished — start and end of execution
  • step.started / step.completed / step.failed / step.skipped — Step life cycle
  • step.log — stdout / stderr of code step
  • agent.message — An intermediate statement by Agent (most common in group_chat)
  • agent.tool_call — Agent calls a tool + parameters + return value
  • artifact.emitted — Produced a product (Markdown/Excel/Image/JSON)
  • breakpoint.hit — breakpoint hit
  • approval.requested / approval.granted / approval.denied

Each event has a timestamp, level (debug/info/warn/error), and attribution step. The filter bar allows you to filter by type/level.

Cancel Midway

The "Cancel" button in the upper right corner of the execution details page can actively stop a RUNNING execution:

  1. The status first changes to CANCELLING
  2. The currently running step receives an interrupt signal (the code step will receive SIGTERM, and the LLM step will actively abort)
  3. After all steps stop, the status changes to CANCELLED.

Some time-consuming operations (long network timeouts) may not respond immediately - the sandbox-level hard timeout will take care of it, and it will definitely stop in up to 5 minutes.

How to See the Results After Running

Execution Details Page

  • Gantt chart — The overall timeline and the time taken for each step
  • step list — Sequential step status overview
  • step details — Input/Output/Event Log/Tool Call
  • Variables panel — Snapshot of all available variables before this step is executed
  • product column — Generated file, click to preview or download

Token & Cost

The top summary bar displays: total time taken, total tokens (in / out), total cost (USD). Click to expand to see the individual consumption of each LLM step.

Common Operations

Run Again with the Same Parameters

The "Retry" button in the upper right corner of the execution details page - immediately initiates a new execution using the variables of this execution. It is convenient to compare the two results.

Run Again from a Certain Step in the Middle

Click "Retry from this step" next to a step - the results of the previously completed step will be retained and you can start over from this step. Suitable for "you don't want to spend money from scratch after fixing a certain step error".

Export Execution Report

"Export" at the top can export the metadata + step details + products of the entire execution as JSON. Suitable for reviewing later or sharing with colleagues.

When should manual execution be used and when should other triggers be used?

  • Manual — Development and debugging, one-time requirements, "I want it to run now"
  • Scheduled Runs — Fixed frequency automatic triggering
  • Webhook — External system event driven

Next