//Guide

Run one workflow on many phones at once

Once a script works on one phone, save it in Axilio as a workflow and run it on as many of your dedicated phones as you need, from one request. Each run gets its own phone session, recording and step-by-step trace.

Use it to finish a batch sooner by splitting it across phones, or to run the same job with different inputs side by side. You queue runs from the dashboard, the CLI, Python, Go or REST.

Updated October 6, 2026.

Workflows and runs

A workflow is the code. A run is one execution of it on one phone.

1.0The model

Workflow

Your Python, saved in Axilio in numbered revisions. Axilio runs it unchanged, on whichever phone each run gets.

Run

One execution of the workflow on one phone. Its run ID is also the ID of the phone session it creates.

Your phones

Pin each run to one of your dedicated phones. A run that can't start yet waits in the queue until a phone is free.

Save the workflow

Write the steps once, then save them as revision 1 of a named workflow.

2.0Step 1

Terminal

cat > workflow.py <<'PY'
# From the home screen: open Chrome, then search.
driver.locator(query="the Chrome app icon").tap()
driver.locator(query="the address bar").fill("axilio")
driver.key_press(Key.ENTER)
PY
 
axilio workflows create search-flow --platform android --code workflow.py

driver is ready for you

Axilio binds driver and Key to the phone each run gets, so the source never allocates a phone of its own.

Every save is a revision

Pull, edit and push the code as new revisions. Restoring an old one copies it forward, so history is never lost.

Recording and traces are on

Runs are recorded and traced by default. Turn either off for every run with --recording=false or --telemetry=false.

Queue it on many phones

One request, one entry per run. Here, three runs on three of your phones.

3.0Step 2

Python

from axilio import RunConfig
from axilio.platform import Client
 
client = Client()  # reads AXILIO_API_KEY
 
# One entry per run, each pinned to one of your dedicated phones.
phones = ["ph_123", "ph_456", "ph_789"]
result = client.runs.create(
    "wf_123",
    runs=[RunConfig(phone_id=phone, variables=[{}]) for phone in phones],
    start_timeout_seconds=300,  # how long a run may wait for its phone
)
print(result.run_ids)

What a run request takes, and how queued runs start
Compared on	Details
Runs per request	1 to 1,000. Each entry in runs is one run; the CLI's --count makes that many.
Which phone	phone_id pins a run to one dedicated phone. The CLI's --phone-id pins every run it creates to the same one.
Inputs	Each run carries its own variables. They're stored as plaintext, so never put passwords or API keys in them.
Waiting for a phone	start_timeout_seconds: 60 to 86,400 seconds, 300 by default. A run still queued after that is canceled.
How many at once	Concurrency and phone availability decide. The rest stay queued until a phone is free or their timeout passes.
When they start	Now, or at a time you book up to 30 days ahead: start_at in the API, or a start time in the dashboard.

Book runs for later

Add start_at to the same request and the runs start at that time instead of now. It works in the API and the dashboard today; the SDKs take it from their next release, shown here.

4.0Scheduling

Python

from datetime import datetime, time, timedelta, timezone
 
from axilio import RunConfig
from axilio.platform import Client
 
client = Client()  # reads AXILIO_API_KEY
 
# Tomorrow at 9:00 UTC, on the same three phones.
tomorrow = datetime.now(timezone.utc).date() + timedelta(days=1)
phones = ["ph_123", "ph_456", "ph_789"]
result = client.runs.create(
    "wf_123",
    runs=[RunConfig(phone_id=phone, variables=[{}]) for phone in phones],
    start_at=datetime.combine(tomorrow, time(9, 0), tzinfo=timezone.utc),
)
print(result.run_ids)  # listed as scheduled until then
 
# Plans changed? Cancel a booked run any time before it starts:
# client.runs.cancel(run_id)

Waits as scheduled

Booked runs list as scheduled and don't hold a phone until their time comes. Book up to 30 days ahead.

Starts on time

Within about 15 seconds of start_at, plus the usual wait for a phone. The queue timeout counts from then.

Cancel until it starts

runs.cancel stops a booked run before it starts. Each booking runs once; book again to run it again.

Follow the runs and read the results

Every run reports its status and leaves a trace and a recording behind.

5.0Step 3

Terminal

axilio runs list --workflow wf_123   # status of each run
axilio runs watch run_123            # follow one run's trace to the end
axilio runs stats wf_123             # total runs and success rate

Status and errors

Each run reports its status, whether it succeeded, its error message and its session ID.

Output and steps

A run's output and errors land in its trace: the Console shows what it printed, the Timeline each step.

The screen

Every run is a phone session: watch it live in the dashboard and replay its recording afterwards.

Success rate

runs stats gives a workflow's total runs and the share that succeeded.

Questions

Short answers to what people ask first.

6.0FAQ

Up to 1,000. Each entry in runs is one run, and each gets its own phone session.

As many as your phones and concurrency allow. The rest wait in the queue until a phone is free or their queue timeout passes.

Yes. Set start_at on the request, or pick a start time in the dashboard, up to 30 days ahead. The runs list as scheduled, can be canceled until then, and start within about 15 seconds of that time plus the wait for a phone. Each booking runs once. The Python and Go SDKs take start_at from their next release; until then, book from the API or the dashboard.

Yes. Each run entry carries its own variables. They're stored as plaintext, so never pass passwords or API keys.

Into its trace. The Console shows its output and errors on the same timeline as its steps; there's no separate logs field on the run. Watch and replay a session →

$15 a day, $75 a week or $180 a month per dedicated phone, with unlimited hours. Reading text with the default model is free; describing a target uses a vision model, billed per call. Pricing →

//Sources

//Keep reading

//GuideWatch and replay what your agent did //Use caseAI agents for app data and monitoring //PricingAxilio pricing

cat > workflow.py <<'PY'

# From the home screen: open Chrome, then search.

driver.locator(query="the Chrome app icon").tap()

driver.locator(query="the address bar").fill("axilio")

driver.key_press(Key.ENTER)

axilio workflows create search-flow --platform android --code workflow.py

from axilio import RunConfig

from axilio.platform import Client

client = Client() # reads AXILIO_API_KEY

# One entry per run, each pinned to one of your dedicated phones.

phones = ["ph_123", "ph_456", "ph_789"]

result = client.runs.create(

"wf_123",

runs=[RunConfig(phone_id=phone, variables=[{}]) for phone in phones],

start_timeout_seconds=300, # how long a run may wait for its phone

)

print(result.run_ids)

Details

Runs per request

1 to 1,000. Each entry in runs is one run; the CLI's --count makes that many.

Which phone

phone_id pins a run to one dedicated phone. The CLI's --phone-id pins every run it creates to the same one.

Inputs

Each run carries its own variables. They're stored as plaintext, so never put passwords or API keys in them.

Waiting for a phone

start_timeout_seconds: 60 to 86,400 seconds, 300 by default. A run still queued after that is canceled.

How many at once

Concurrency and phone availability decide. The rest stay queued until a phone is free or their timeout passes.

When they start

Now, or at a time you book up to 30 days ahead: start_at in the API, or a start time in the dashboard.

from datetime import datetime, time, timedelta, timezone

from axilio import RunConfig

from axilio.platform import Client

client = Client() # reads AXILIO_API_KEY

# Tomorrow at 9:00 UTC, on the same three phones.

tomorrow = datetime.now(timezone.utc).date() + timedelta(days=1)

phones = ["ph_123", "ph_456", "ph_789"]

result = client.runs.create(

"wf_123",

runs=[RunConfig(phone_id=phone, variables=[{}]) for phone in phones],

start_at=datetime.combine(tomorrow, time(9, 0), tzinfo=timezone.utc),

)

print(result.run_ids) # listed as scheduled until then

# Plans changed? Cancel a booked run any time before it starts:

# client.runs.cancel(run_id)

Run one workflow on many phones at once

Workflows and runs

Workflow

Run

Your phones

Save the workflow

driver is ready for you

Every save is a revision

Recording and traces are on

Queue it on many phones

Book runs for later

Waits as scheduled

Starts on time

Cancel until it starts

Follow the runs and read the results

Status and errors

Output and steps

The screen

Success rate

Questions

How many runs can one request start?

Do all the runs start at once?

Can I schedule runs for later?

Can each run get different input?

Where does a run's output go?

What does it cost?

//Sources

//Keep reading

Put your agent on a real phone

Run one workflow on many phones at once

Workflows and runs

Workflow

Run

Your phones

Save the workflow

driver is ready for you

Every save is a revision

Recording and traces are on

Queue it on many phones

Book runs for later

Waits as scheduled

Starts on time

Cancel until it starts

Follow the runs and read the results

Status and errors

Output and steps

The screen

Success rate

Questions

How many runs can one request start?

Do all the runs start at once?

Can I schedule runs for later?

Can each run get different input?

Where does a run's output go?

What does it cost?

//Sources

//Keep reading

Put your agent on a real phone