Skip to content

phongent Turns One Phone Into a Multi-Device Agent Console

An Apache-2.0 Kotlin project lets an iPhone or Android handset dispatch tasks to several Android devices and collect the results in one place.

Draftship Editorial3 min read

One phone, several Android agents

phongent multi-device agent control turns a single phone into a control panel for agents spread across several Android handsets. The GitHub project, mylea1/phongent, had 357 stars and 82 forks as of 2026-11-14, gathered in the 13 days since it was created on 2026-11-01. It is written mainly in Kotlin and licensed Apache-2.0.

The premise: write an instruction, pick which devices should carry it out, then watch progress, messages and live screens from one interface. iOS and Android can both act as the controller. Only Android can also run tasks on its own device, so an Android handset can dispatch work and execute it.

How setup and dispatch work

The readme lays out five steps. Install a build: an APK on Android, or an Xcode build you sign yourself on iOS. Configure the Android executor inside the app, including model settings and phone-operation settings, and grant the permissions it asks for, such as accessibility and notification reading. Both sides join the same local network and the Android service stays running; the first connection is approved on the Android device. Then set the controller's task model. Accounts are signed in on the individual devices rather than in one place.

Tasks go to every connected device, to a group, or to a single named device. You can long press a device and describe its job, or let the controller's task agent split the work. Groups can be created in bulk and collapsed in the agent view. Each device belongs to exactly one group, and anything ungrouped sits in a default group. Group membership is stored on the controller that created it and does not sync to other controllers.

Sample instructions from the readme include asking device A to open an app's home page and take a screenshot while B opens a browser to read the day's weather, or telling a content group to draft separate product descriptions while a review group checks each device's status.

Alongside dispatch there is a live mini window that mirrors the Android virtual screen and accepts taps, swipes and typed input, so you can take over by hand. It is on by default and can be switched off in settings. A message view sorts notifications by device, app and source, with replies sent through that device's agent, and central settings cover memory, skills, scheduled tasks and operation routes.

What it suggests for app builders

Little is packaged for reuse: the readme lists no public API or SDK. The value today is likely as a worked example of the pattern, one coordinator holding the task model, a pool of workers holding device permissions, and a status layer that merges results. That split is likely to matter to anyone adding agent features to a web app, because the hard parts repeat: routing a job to the right worker, showing partial progress, and keeping model keys away from whatever the user can read.

Build details are in the readme: JDK 17 and compileSdk 36 with ./gradlew :app:assembleDebug for Android, a minimum of Android 11, and iOS 17 with the MikaRemote scheme for the controller app.

Limits and open questions

The readme warns not to use the upstream project's APK as the executor, since it may be missing the group-control interfaces. It also says model keys, account credentials and signing material should stay out of source. Distribution appears to mean side-loading an APK and signing an iOS build yourself; the source says nothing about app store availability.

Several things stay unstated: which models the controller and executors use, how traffic is secured once devices share a network, the terms for bundled third-party assets beyond a note that each carries its own licence, and whether a controller can manage Android devices that do not run this project's executor. The reliance on accessibility and notification access is worth a careful look before anyone points this at a phone holding real accounts.

If you are building the app side

Draftship is an AI app builder where you describe an app and it writes, builds and hosts it, with a built-in database and sign-in. It is not connected to phongent. If you are sketching the app that would sit in front of an agent fleet, adding AI to your app covers the pieces on Draftship's side.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.