Last month someone asked me this: they had five phones and wanted them to run the same workflow at the same time. Would a command land on the wrong phone?
A few years ago, device-farm control often put several devices on one session. You told the first phone to tap a button, and the others tapped too. That is where the worry comes from.
Each device has its own session
From version 1.2.0, each device has its own session. Before that, several devices shared one session and commands conflicted. Now five devices running the same workflow each run their own copy, and a command only goes to its own phone.
The version history puts it this way: multi-agent now has an independent session for each device, which avoids the conflicts that come from several devices using the same session at once.
Tasks do not cross from one phone to another. The phones still compete for three kinds of resource.
Bandwidth shows up first. If the same batch has to upload files, several phones pushing at once slow each other down. Every phone gets slower, and the total time is not necessarily shorter than running in batches.
Compute on the computer, and USB power, are the second kind. You barely notice it with a few devices. It starts to show past a dozen. With many USB ports and long cables, power runs short and phones drop offline. It looks like a random phone dropping, with no pattern. A hub with its own power supply fixes most of that.
The third kind is the account. The same account operated from several devices in a short window is itself a risk signal. The tool is not the source of that. Keep one account on one device. If you need several accounts, add devices. Do not have several phones take turns logging into the same account.
One workflow, or separate dispatches
If every phone does the same thing, such as each posting one item or each walking through a check, one workflow dispatched by group is enough. Set it up once and it covers the batch.
If each phone does something different, such as different accounts posting different content, or different shops running different flows, set up separate workflows and dispatch them separately. Force the differences into one template and the parameters keep growing. Later you cannot remember which parameter belongs to which phone.
While they run together, each phone’s network and performance differ. The same “wait 5 seconds” can be right on one phone and too short on another. Most runs pass, one or two fail now and then, and the phone that fails is not always the same one. Check the fixed waits in the flow first.
Replacing “wait N seconds” with a check against the screen is steadier. Use an image template and tell it to wait until this button appears, then tap. Each phone follows its own pace, so a difference in speed does not turn into a wrong step. The docs explain how to set an image template up.
From two phones up to the limit
Pick two phones and run the same workflow. When it finishes, look at the result on each phone and confirm the commands did not land on the other one.
After that, dispatch by group. A light task can go out in one push. A heavy task goes in batches of five to ten.
When the run finishes, check execution history. Then add phones step by step and watch two things: whether dropouts increase, and whether the total time is still getting shorter. When you reach a level where the total time stops falling, or dropouts rise clearly, that is the limit for this setup. There is no point adding more.
Ten phones taking turns on the same account look, to the platform, like one account active in ten places at once. Doing the same thing by hand carries the same risk. One device, one account. More accounts means more devices.
If you only have one or two phones now, this can wait. When you pass five devices and tapping them one by one gets too slow, try the order above.
Frequently asked questions
- Do tasks interfere when several phones run the same workflow?
- No crossing over. Since version 1.2.0 every device has its own independent session, which removed the conflict that used to occur when multiple devices shared one session. Five devices running the same workflow each execute their own, and commands do not land on the wrong device.
- So what does interfere?
- Three kinds of resource. Bandwidth, since simultaneous large uploads compete. Machine resources and USB power, since insufficient delivery with many ports causes dropouts. And accounts, since the same account being operated on several devices at once is a risk signal regardless of the tool used.
- One workflow across many devices, or separate workflows?
- It depends whether the tasks are identical. If the same action repeats on every device, such as each posting one item, one workflow dispatched by group is simplest. If each device does something different, configure separate workflows rather than forcing the differences into one template.
- Do inconsistent wait times between devices cause confusion?
- Not confusion, but they do desynchronise. Networks and performance differ, so the same wait duration has different effects on different devices. If a step must wait for a specific screen, use image-based detection rather than a fixed number of seconds.
- If one device fails, does it affect the others?
- No. Each device is an independent execution, so one failure does not interrupt the rest. The failed device gets its own record in Execution History with the error message, and you work from that.
- Is there a recommended concurrency level?
- There is no universal number. It depends on task weight and your hardware. Light tasks tolerate more. Heavy tasks work better in batches of five to ten. Start with five and watch total time and device stability, then increase until dropouts rise or total time stops improving. That is your ceiling.
- Is there a risk in running the same account across several devices?
- Yes, and this is the part that genuinely deserves attention. One account operating across several devices in a short window is a behaviour pattern platforms flag. Keep one account per device, and add devices when you need more accounts, rather than rotating several devices through the same login.
- How do I confirm there is no interference?
- Two ways. Set concurrency to two devices, run once, and check the actual result on each phone. Then go through Execution History line by line and look for any device reporting an error that belongs to another. Only scale up after that baseline holds.
Try two first
Get two devices stable, then add more
Software is completely free and runs on your own computer.