A run executes the flow on your engine, in a real browser. You watch it live on the canvas, in the logs and in the live preview.
Test run vs real run
| Test run | Real run | |
|---|---|---|
| Start | Test run (flask) in the run bar, ⌘Enter, node menu | Run (rocket) in the run bar, then confirm Run for real? |
| Effect nodes (⚡) | Not executed: first output dot, output { "dry": true }, log Dry run: <name> was not executed | Act for real: post, reply, send, buy |
| Data tables | Read only | Read and write |
| Called-flow cache | Read only | Read and write |
| Badge | Test run on the status card | none |
The editor runs your draft (it saves first). Triggers (Schedule, Webhook, Form) run the published version.
A Studio run has its own browsers (the flow's, not the campaign's), so it signs its accounts in again. If a campaign that is on uses the same account, a warning says so when the run starts.
Run part of a flow
| Action | Runs |
|---|---|
| Test run / Run | The whole flow from its trigger. |
| Node menu › Run this step | Only that node. Earlier outputs come from the last run (or pins). |
| Node menu › Run from here | From that node to the end. |
| Command palette › Run selected (test) | The selected nodes. |
| Retry from failed | From the node that failed, reusing what succeeded. |
When the draft changed since the run that failed, Retry from failed becomes a menu:
- With the current version — your edited draft.
- With the version it ran — the exact graph that run used (the engine keeps it with the run).
The run controls
In the editor's header: Test run and Run (with Retry from failed after a failed run). While a run is active they give way to its status (Running, with a flask icon for a test run), Pause / Resume and Stop (editors only). On narrow screens they show as icons. When the run waits for you, its question shows as a row under the header. Follow the running node (crosshair) sits in the canvas controls, bottom left.
The status card at the top shows the status, the current step, a progress bar (4 of 15 steps), loop progress (Each request: item 2 of 3), the run clock, and Follow the run. Panning the canvas by hand stops following; press Follow the run to resume.
When the run ends, a summary card shows 12 done · 1 failed · 2 skipped and the AI usage, e.g. AI: 5 calls · 5,431 tokens · $0.0021.
Statuses
| Status | Label | Meaning |
|---|---|---|
QUEUED | Queued | Waiting for its browser (one run per browser at a time) or for the container to start. |
RUNNING | Running | Working. |
PAUSED | Paused | You paused it. The current node goes back to pending and runs again on Resume. |
INTERVENTION | Needs you | Waiting for you: an Approval, an Intervention, or a step set to Ask a human. |
SUCCEEDED | Succeeded | Finished. |
FAILED | Failed | A step failed (and On error was Stop, or End the run and undo), the run hit its time limit, or the browser stopped. |
INCOMPLETE | Incomplete | A step failed with On error Save to resume later. The run waits to go on from that node: by itself after the node's wait, or from the Runs page (Resume, Discard). |
CANCELLED | Cancelled | You pressed Stop (its called flows stop too). |
The run windows
Logs, Preview, Output, Stacks and Runs are , in the bottom area by default. Starting a run (or opening one) shows Logs; the area's size is part of the layout this browser remembers.
Logs
Every log line has a time, level, node and message. Filter by Level (debug, info, warn, error), Node, Type (Steps, Navigation, AI, Intervention, System) and text search. Logs from a called flow show under its Run flow node, prefixed with the flow's name (Shopee Product Finder › Searched 20 products).
Output
Each node's output from this run, as Schema, Table or JSON.
Stacks
The run's stacks, one tab per node that keeps one (Collect posts, the AI's picks, the loop's items), with the same cards and controls as Up next on a campaign's .
Preview
The run's browsers, live. See .
Runs
Recent runs of this flow; open one to load it onto the canvas.
Waiting for you
When the run needs you, the status turns Needs you, the preview opens the browser it waits in (with control on), and a notification lands under the bell.
| Case | What you do |
|---|---|
| Intervention node, or a step on Ask a human | Do what the steps say in the live view (sign in, solve a captcha), then press Done: it checks first and tells you why when it isn't done (Continue anyway goes on regardless). The check also runs by itself every few seconds and goes on once it passes; turn off Auto check when it's done (in Studio and on a campaign's dialog) to change settings or switch accounts first, then press Done yourself. A step on Ask a human shows the failure with Retry, which runs it again from the start. Its simple fields (text, numbers, yes/no; never secrets or lists) show as inputs: change one (a search query, a URL) and Retry uses it for that step for the rest of the run. |
| Approval node | Read and edit Data, then Approve or Reject. |
| Approval with If no answer set | A countdown on the button (Approve · 0:12) decides by itself when it reaches zero, unless you answer first. |
Recordings
With Record runs on in (plans with run recordings: Pro 7 days, Team 30, Business 90), every browser step leaves a screenshot of its page when it ends. Each is a line in the log, 📷 Screenshot after : click it to open the picture. They're kept for the plan's days after the run ends, then the engine deletes them.
When the engine is busy
When every browser slot of the engine is taken, a run waits for one instead of failing (up to 10 minutes; the log says "The engine is busy"). Business plans and up wait first in line.
Time limits
- Each node has its own timeout (see ).
- The flow has a run time limit (); time spent waiting for you counts.
0means no limit, for endless flows.
AI cost
AI nodes log their token usage and cost. The run adds them up (ai.calls, ai.input, ai.output, ai.cached, ai.cost). The cost uses the prices you set on the AI key; a key without prices adds tokens only. A called flow's own run counts its calls; the copies relayed to the caller don't count twice.
FAQ
Why is my run stuck on Queued?
Its campaign (or the flow, in tests) is busy with another run (one main run at a time), or the engine is starting the container (up to 90 s). If the engine is at capacity, the run fails with The engine is at capacity.
Can I watch a run started by a schedule?
Yes. Open it from Runs (the flow's Runs window or the global Runs page).
Does Pause stop the browser?
No. Pause stops the flow at the current step; the browser and its pages stay as they are, and you can still use the live view.