Three Unconventional Ways to Run OpenClaw on Edge Devices

While many solutions claim one‑minute OpenClaw deployment on computers or cloud servers, this article presents three unconventional edge‑side projects that run the AI agent framework on inexpensive hardware.

25‑Dollar Phone Running OpenClaw

Developer Marshall Richards demonstrated OpenClaw on a $25 Moto G 2025 phone equipped with 4 GB RAM and an 8‑core ARM64 CPU. The installation uses Termux and npm instead of the official bash script, requiring a manual 15‑30 minute compilation of llama.cpp. Because OpenClaw insists on using /tmp, which Termux cannot access, the problem is solved by adding an environment variable in .bashrc that redirects the temporary directory to Termux’s user space. Richards also created overlay_daemon.py so OpenClaw can display information on the phone screen, and leverages Termux:API and Termux:GUI to control hardware.

5‑Dollar Chip MimiClaw

The MimiClaw project pushes the limits further by running the core OpenClaw functionality on an ESP32‑S3 microcontroller. Implemented entirely in C, it requires no Linux or Node.js, only a thumb‑size board with 16 MB flash and 8 MB PSRAM (≈$10). Power consumption is about 0.5 W, allowing 24‑hour operation. The device receives Telegram messages over Wi‑Fi and runs a ReAct loop: Claude thinks, calls tools, reads memory, then replies. All data resides in local flash. MimiClaw uses a dual‑layer configuration system: compile‑time defaults are set in mimi_secrets.h, while runtime overrides are applied via a serial‑CLI. Additional utilities include web search, time sync, and an HTTP proxy for restricted networks.

BotDrop: Turning an Android Phone into an AI Host

BotDrop is an Android application built on Termux that requires no root access. It provides an Ubuntu‑like APT environment for installing Node.js‑based AI agents. Setup is performed through a Telegram bot, eliminating the need for external servers or cloud fees. Currently BotDrop supports OpenClaw and has been tested with Anthropic and Gemini models. The team plans to add OAuth for easy model switching. The app requires a phone with more than 4 GB RAM for smooth performance and is still at version v0.1, lacking the ability to execute complex tasks, but it offers a low‑cost entry point for users wanting to experiment with AI agent clusters.

Hardcore Manual Mode: Building BotDrop Yourself

For users who prefer full control, the same functionality can be reproduced manually with Termux. The steps are:

Initialize a Linux environment in Termux:

pkg update && pkg upgrade -y
pkg install nodejs git openssh tmux -y

Install clawdbot and adapt the npm path for Android permissions:

npm config set prefix ~/.npm-global
echo 'export PATH=$HOME/.npm-global/bin:$PATH' >> ~/.bashrc
source ~/.bashrc
npm i -g clawdbot

Apply three Android‑specific patches (clipboard, log directory, environment variables) using node -e one‑liners that modify the source files in the Termux npm directory.

Start the service inside a tmux session, ensuring the Termux battery policy is set to “unrestricted” and running termux-wake-lock to keep the CPU awake.

tmux new -s clawd
clawdbot gateway --port 18789

Connect remotely from a computer via SSH, eliminating the need to interact with the phone screen.

# On the phone, start sshd, then from PC:
ssh -p 8022 [email protected]

These commands produce a tiny OpenClaw host that is smaller than a Mac Mini.

Conclusion

The three projects illustrate new ways to deploy AI agents on the edge, prioritizing existing low‑cost hardware over high‑end servers. While they make AI more accessible in daily life, they also raise security concerns: agents have full control over the device and could leak data or act maliciously. Edge deployment remains a niche for enthusiasts due to limited compute, power constraints, and development complexity, but the long‑term vision is a cloud‑plus‑edge architecture where the cloud handles heavy tasks and edge nodes perform instant, privacy‑preserving operations.

Related Links

mimiclaw: https://github.com/memovai/mimiclaw

ClawPhone: https://github.com/marshallrichards/ClawPhone

botdrop-android: https://github.com/zhixianio/botdrop-android/

Termux App: https://github.com/termux/termux-app

Android APK: https://github.com/zhixianio/botdrop-android/releases/download/v0.2.1/botdrop-app_v0.2.1%2B0cf4115-apt-android-7-release_universal.apk