Turn spare phones into AI agents without the Linux bloat
It exists because installing an entire Linux distribution just to run an AI agent on a phone is storage overkill.

What it does
This project ports the OpenClaw AI-agent gateway to Android without the usual proot-distro overhead. Instead of installing a full Debian or Ubuntu chroot inside Termux—which costs 1–2 GB of storage and a proot performance tax—it installs only the GNU glibc dynamic linker (ld.so) and wraps official linux-arm64 Node.js binaries to run directly atop Android’s Bionic libc. A single install command handles path translation, temp-folder wiring, and systemd bypass, while a companion oa CLI manages updates and backups.
The interesting bit
The real trick is surgical minimalism: the project doesn’t port Linux, it just borrows the linker. By loading Node.js through ld.so and redirecting standard paths like /tmp and /bin/sh to their Termux equivalents, it gets standard Linux binaries running unmodified. If you prefer not to touch Termux at all, the standalone “Claw App” bundles a terminal emulator and WebView dashboard into one APK.
Key highlights
- Cuts storage overhead from roughly 1–2 GB to about 200 MB by skipping the full distro install.
- Runs at native speed instead of through a proot translation layer.
- Bundles a standalone Android app with a built-in terminal and WebView UI; no Termux required.
- Supplies an
oaCLI that updates components, createstar-based backups (working around Android’s hardlink restrictions), and restores them. - Targets low-power, always-on phone deployments.
Caveats
- Android’s battery optimizations and Phantom Process Killer will throttle or kill the gateway; the README warns that keeping the service alive requires changing multiple Developer Options settings.
- The Play Store build of Termux is explicitly discontinued and incompatible—you must install from F-Droid.
- OpenClaw’s native backup command fails on Android because hardlinks are blocked in app-private storage, so the project relies on its own
tar-based workaround.
Verdict
A neat fit for homelabbers who want to repurpose an old phone as a self-hosted AI node without buying a Pi. If you expect a fire-and-forget mobile app, look elsewhere—this is still terminal-first infrastructure with manual process babysitting.
Frequently asked
- What is AidanPark/openclaw-android?
- It exists because installing an entire Linux distribution just to run an AI agent on a phone is storage overkill.
- Is openclaw-android open source?
- Yes — AidanPark/openclaw-android is open source, released under the MIT license.
- What language is openclaw-android written in?
- AidanPark/openclaw-android is primarily written in Java.
- How popular is openclaw-android?
- AidanPark/openclaw-android has 1.7k stars on GitHub.
- Where can I find openclaw-android?
- AidanPark/openclaw-android is on GitHub at https://github.com/AidanPark/openclaw-android.