🤖 Droid AI Toolkit (Termux)

A high-performance, automated toolkit for running AI tools — OpenClaw, Gemini CLI, n8n, Ollama, Hermes, Nanobot, Pi, and Paperclip — natively on non-rooted Android devices. This toolkit bypasses kernel restrictions (renameat2), patches hardcoded system paths, and optimizes execution for mobile environments.

📱 Compatibility

OS: Android 9.0 and above.
Architecture: Tested on armv8l (32-bit) and aarch64 (64-bit) mobile CPUs.
Optimization: Automatically detects system RAM and recommends appropriate memory limits (512MB to 2048MB) for Node.js and n8n workloads.
Package Managers: Supports both npm (Standard) and pnpm (High Efficiency) for Node.js-based tools.
Process Management: Supports PM2 (Recommended) and termux-services (Native).

⚠️ Architecture Warning: Tools that depend on Rust-compiled Python extensions (Hermes, Nanobot) are not supported on armv8l/armv7l devices because upstream maturin rejects the architecture and pre-built wheels require glibc (not Android's bionic libc).

📋 Before You Start

Close other apps to free up RAM. The installer auto-detects available memory, but Android's Low Memory Killer may terminate Termux if other apps are running.
Connect to Wi-Fi. Large downloads include Paperclip (~2GB), n8n, and Ollama models. Mobile data plans may be consumed quickly.
Ensure free storage: Paperclip needs ~2GB free; other tools need ~200–500MB each.
Install Termux from F-Droid (not Play Store). The Play Store version is obsolete and lacks required packages.
Dependencies auto-installed: jq, whiptail, gum, curl, git, nodejs, and postgresql are installed automatically by the script if missing.

🚀 Quick Start

1. Environment Setup

Install Termux from F-Droid. Do not use the Play Store version as it is obsolete.

2. Run the Toolkit

Execute the following command to start the interactive toolkit:

curl -sSL https://raw.githubusercontent.com/niyazmft/droid-ai-toolkit/main/install.sh | bash

📝 What happens next: The script launches an interactive TUI menu (gum if available, otherwise whiptail). Pick one tool at a time — downloads happen in the terminal. It is safe to re-run the script at any time. Keep Termux open and avoid switching apps during installation, as Android's Low Memory Killer may silently terminate the process. Individual tool installs typically take 2–15 minutes depending on your device and network.

💡 Smart Repair (v1.5.0+): If a tool is already installed, the toolkit offers a [R] Repair mode. Use this to fix Android-specific patches in seconds without re-downloading the entire package.

3. Choose Your Tools

The toolkit uses a nested TUI menu. If gum (charm.sh) is installed, you get a premium, color-rich interface with visual separators; otherwise it gracefully falls back to whiptail:

Menu	Tools Available
🤖 AGENTS	OpenClaw, Hermes, Nanobot
⚙️ WORKFLOWS	n8n, Paperclip
🛠 UTILITIES	Gemini CLI, Pi Coding Agent, Ollama, GCP Bridge
🔧 SERVICES	PM2 Process Management, Native Background Services (sv)
🗑 UNINSTALL	Modular uninstall for any installed tool, or WIPE ALL

4. Onboard OpenClaw (If Installed)

Initialize your account and API providers:

openclaw onboard

Select QuickStart and choose an external provider (OpenRouter, OpenAI, etc.).

5. Background Service (Optimized)

To keep tools running even after you close Termux:

Run the toolkit and choose SERVICES → PM2 Process Management.
Select the service you want to start (OpenClaw, n8n, Ollama, Paperclip, etc.).
View logs with: pm2 logs

✨ Key Features

🎨 Modern UI Engine: v1.13.0+ uses gum (charm.sh) for premium, color-rich menus with visual separators and touch-friendly navigation; gracefully falls back to whiptail if unavailable.
⚡ Zero-Latency Navigation: In-memory caching for Bash config lookups eliminates ~1-second menu reload delays.
🛠 Smart Repair: Detects existing installations and provides a 2-second "Repair Only" path to re-apply patches without redundant downloads.
🩹 Zero-Config Patching: Automatically fixes the koffi native bridge and renameat2 kernel crashes for OpenClaw.
📂 Path Awareness: Aggressively redirects /bin/npm, /bin/node, and /tmp to Termux-compatible directories using $PREFIX.
🚀 PM2 Integration: Native support for starting, stopping, and monitoring OpenClaw, n8n, Ollama, Paperclip, Pi, and Gemini CLI via PM2 with optimized memory flags.
📦 pnpm Support: Integrated support for pnpm to speed up installations and save storage space.
🧠 Memory Guard: Automatically clears memory (PM2 kill) and increases Node.js heap limits (1.5GB+) to prevent crashes on low-RAM devices during updates.
🛡 Surgical Cleanup: The uninstaller offers Soft/Deep options and a Wipe Stack (Reset) function that preserves your system packages while cleaning the apps.
🧩 Gemini CLI Support: Dedicated installer with NDK environment optimizations and fs.promises.rename → copyFile+unlink patch to prevent Android ENOENT.
🦙 Ollama Support: One-click install via Termux native package (pkg install ollama).
⚡ Hermes Support: One-click install via official curl installer.
🤖 Nanobot Support: pip install with --no-build-isolation for pre-seeded dependencies.
🥧 Pi Coding Agent (Recommended): npm/pnpm global install with Termux-specific AGENTS.md context.
📎 Paperclip (EXPERIMENTAL): Delegates to paperclip_manual_install.sh which handles clone, patches, pnpm install, prebuilt tarball download, PostgreSQL bootstrap, and PM2 ecosystem file generation.

🤖 AI Agents

OpenClaw — AI Gateway

Multi-channel AI gateway with Telegram, Slack, and Discord support. Automatically patched for Android:

Koffi patch: renameat2 → rename to avoid kernel crashes.
Path redirection: /tmp/openclaw, /usr/bin/npm, /bin/node → Termux paths.
Plugin pruning: Disables 118 stock plugins on install to reduce memory footprint.


Install method	npm/pnpm global
Architecture	✅ All architectures supported
Memory	Varies by plugin load; ~512MB minimum
Critical warning	Never run `openclaw update` — use toolkit's [R] Repair or [U] Update

openclaw onboard          # Configure API keys
openclaw doctor --fix     # Repair schema issues

⚠️ NEVER run openclaw update — it overwrites Android patches. Use the toolkit's [R] Repair or [U] Update instead.

💡 Version Selection (v1.15.3+): The toolkit now supports installing a specific OpenClaw version (e.g. 2026.6.30) instead of always pulling @latest. Use [V] Install Specific Version from the OpenClaw menu.

💡 Zulip Plugin (v1.15.3+): A dedicated sub-menu [Z] Zulip Plugin is available under AGENTS → OpenClaw for installing, updating, or uninstalling the Zulip integration (clawhub:@niyazmft/openclaw-zulip). You can also pin to a specific plugin version (e.g. 2026.7.0).

Hermes — Nous Research Agent

AI agent by Nous Research, installed via the official curl installer.


Install method	Upstream curl installer + manual pip fallback
Architecture	❌ Not supported on `armv8l`/`armv7l` (maturin/jiter incompatibility)
Memory	~512MB RAM minimum
Build deps	python, clang, rust, make, pkg-config, libffi, openssl, binutils

hermes                # Start the agent

💡 Wheel Cache (v1.15.3+): After your first successful Hermes install, compiled Python wheels (cryptography, Pillow, pydantic-core, jiter, etc.) are automatically saved to ~/.hermes/wheel-cache. Future [R] Reinstall or [U] Update operations skip compilation entirely and reuse these cached wheels, dropping install time from ~90 minutes to under 2 minutes on the same device. On armv8l/armv7l, the toolkit will display a graceful error message and skip installation.

Nanobot — Python AI Agent

General-purpose Python AI agent with Anthropic Claude integration.


Install method	`pip3 install nanobot-ai`
Architecture	❌ Not supported on `armv8l`/`armv7l` (maturin/jiter incompatibility)
Memory	~512MB RAM minimum
Build deps	python, pip, setuptools, wheel

nanobot --help        # View available commands
nanobot               # Start the interactive agent

On armv8l/armv7l, the toolkit will display a graceful error message and skip installation.

⚙️ Workflows & Automation

n8n — Workflow Automation Server

Professional-grade workflow automation with an optional GCP bridge for secure public access.


Install method	npm/pnpm global (`n8n@latest`)
Architecture	✅ All architectures supported
Memory	Auto-capped to `min(2048, max(512, RAM * 0.75))`
Extras	Watchdog cron, autossh GCP tunnel, tmux session manager

n8n start             # Start manually
~/n8n_server/scripts/n8n-monitor.sh   # Watchdog restart

Access locally at http://localhost:5678.

⚠️ Warning: The installer forcibly kills running OpenClaw, n8n, and PM2 processes during setup to free memory. If you have active workflows or conversations, save your work before installing or updating n8n.

📄 License note: n8n is under the Sustainable Use License. See ATTRIBUTIONS.md for redistribution terms.

Paperclip — AI Orchestration Server (EXPERIMENTAL)

Open-source orchestration server for managing teams of AI agents.


Install method	Delegates to `paperclip_manual_install.sh`
Architecture	✅ All architectures supported (with caveats)
Memory	~2GB free RAM recommended; LMK-resilient install
Database	External PostgreSQL (`pkg install postgresql`)
Build	Prebuilt `dist/` tarball (primary) or local tsc build (fallback, ~65 min)

The standalone installer handles:

Cloning, patching (removes embedded-postgres, drops ui workspace).
pnpm install with LMK-kill detection and retry.
Symlink repair for tsc, tsx, esbuild.
Download of prebuilt dist/ and ui-dist tarballs from GitHub releases.
PostgreSQL bootstrap with stale-process cleanup.
Secret generation and PM2 ecosystem file creation.

cd ~/paperclip
export PAPERCLIP_HOME=~/paperclip
export DATABASE_URL=postgres://paperclip:paperclip@localhost:5432/paperclip
pnpm paperclipai onboard       # One-time setup
pnpm paperclipai configure     # Enable LAN access
pm2 start ecosystem.config.cjs # Start server

⚠️ Reinstalling Paperclip wipes config.json and secrets. The installer deletes ~/paperclip and reclones, which removes instances/default/config.json and config/paperclip.env. Your database (workflows, users, history) survives in PostgreSQL, but you must re-run pnpm paperclipai onboard after reinstall to regenerate config and secrets. Back up ~/paperclip/instances/default/ before reinstalling if you want to preserve settings. Requirements: ~2GB free RAM, 2GB+ storage, pnpm 9.15+, PostgreSQL running. UI is pre-built and downloaded as a tarball — never built on-device (Vite/esbuild requires ~4–6 GB transient RSS).

🛠 Developer Utilities

Gemini CLI — Google's Command-Line AI Assistant


Install method	npm/pnpm global (`@google/gemini-cli@latest`)
Architecture	✅ All architectures supported
Memory	~512MB RAM minimum
Build deps	`python`, `make`, `clang`, `pkg-config` (auto-installed if missing)
Patch	`fs.promises.rename` → `copyFile+unlink` to prevent Android `ENOENT`

gemini --help         # View available commands
gemini                # Start interactive session

💡 Smart Repair/Update: Re-running the toolkit for Gemini CLI offers [R] Repair (re-apply patches) or [U] Update (latest version).

Pi Coding Agent (Recommended)

The high-performance coding agent optimized for the Termux environment.


Install method	npm/pnpm global (`@earendil-works/pi-coding-agent@latest`)
Architecture	✅ All architectures supported
Memory	~512MB RAM minimum

pi --help             # View available commands
pi                    # Start the interactive agent

The toolkit automatically creates ~/.pi/agent/AGENTS.md with Termux-specific paths ($HOME, $PREFIX, termux-open-url) so the agent never hallucinates standard Linux paths. Legacy @mariozechner/pi-coding-agent installations are automatically detected and migrated.

⚠️ Warning: If you already have a command named pi on your system, the installer will remove it to avoid conflicts.

Ollama — Local LLM Runner

Run large language models locally. Installed via Termux's native package manager.


Install method	`pkg install ollama` (Termux native)
Architecture	✅ All architectures supported
Memory	~1GB+ RAM recommended for 7B models

ollama serve          # Start the server
ollama pull llama3    # Download a model
ollama run llama3     # Run a model

Use SERVICES → PM2 to keep Ollama running in the background. Downloaded models are stored in ~/.ollama and preserved during uninstall.

GCP Bridge Walkthrough (Optional)

To expose your n8n instance securely to the internet (https://yourdomain.com), follow this walkthrough:

Step 1: Prepare the GCP VM

Create Instance: In GCP Console, create an e2-micro VM (Debian/Ubuntu).
Static IP: Reserve a static external IP for this VM.
Firewall: Allow TCP 80 (HTTP), 443 (HTTPS), and 22 (SSH).

Step 2: Set up DNS

Point your domain (e.g., n8n.example.com) to the GCP VM's static IP.

Step 3: Configure Nginx (on GCP VM)

Install Nginx and Certbot: sudo apt install nginx certbot python3-certbot-nginx
Create a site config that proxies to localhost:5678.
Secure it with SSL: sudo certbot --nginx -d yourdomain.com

Step 4: Establish the Tunnel

Run the toolkit on your Android device and choose SERVICES → Configure GCP Bridge.
Follow the prompts to enter your VM IP and Domain.
Copy the generated SSH Public Key and paste it into the GCP VM's ~/.ssh/authorized_keys file.
The monitor script will now automatically maintain a secure autossh tunnel to the VM.

🗑 Uninstallation & Reset

Run the toolkit and select UNINSTALL to access the modular uninstallation menu. The order mirrors the on-device menu. Each option provides a detailed summary of the impact before you confirm:

Remove OpenClaw: Choice of Soft Uninstall (keeps memories/skills) or Deep Uninstall (full wipe). Automatically cleans up PM2 and background services.
Remove n8n: Surgically kills the GCP tunnel (port 5678) and removes the watchdog cron.
Remove Gemini CLI: Full removal of application binaries and configurations.
Remove Hermes: Runs the official uninstaller if available, otherwise removes directories manually.
Remove Ollama: Removes the package. Downloaded models in ~/.ollama are preserved.
Remove Pi: Full removal of global package and configuration.
Remove Paperclip: Stops the PM2 service and preserves the source code and PostgreSQL database.
Remove Nanobot: pip uninstall + directory cleanup.
Wipe Software Stack (Reset): Batch "Deep Uninstall" of all applications. Safe Reset: Cleans all toolkit-specific data but preserves system packages (Node.js, Git, Python, etc.) so your other Termux apps don't break.

📊 Quick Commands

Tool	Command
OpenClaw	`openclaw` · `openclaw doctor` · `sv restart openclaw`
Ollama	`ollama serve` · `ollama pull llama3` · `ollama run llama3`
Hermes	`hermes`
Nanobot	`nanobot`
Pi	`pi`
Paperclip	`pm2 start ~/paperclip/ecosystem.config.cjs`
n8n	`n8n start`
Gemini CLI	`gemini`
PM2	`pm2 status` · `pm2 logs` · `pm2 restart all` · `pm2 stop all`
Native Services	`sv up n8n` · `sv down n8n` · `sv status openclaw`
Wake Lock	`termux-wake-lock` (prevents Android from killing background processes)

🔄 Maintenance

🛡 Safe Updates & Smart Repair

⚠️ WARNING: Never use the built-in openclaw update command. It will overwrite the Android patches and break the application.

To update or repair safely, re-run install.sh, choose the tool's Install/Repair option, then pick the appropriate mode:

Mode	What it does	Time	Use when...
[R] Repair	Re-applies Android patches, fixes symlinks, restores configs	~2 seconds	The app broke after a Termux update or system change
[U] Update	Downloads the latest upstream version + re-applies patches	1–10 minutes	You want new features or bug fixes

💡 Hermes Wheel Cache: Compiled wheels are cached locally after the first install. Reinstalls on the same device use prebuilt wheels and skip Rust/C compilation. See the Hermes section below for details. 💡 Latest Version: This toolkit always installs the latest available version of each tool to ensure maximum feature compatibility and security.

🔋 Battery Optimization

To prevent Android from killing the background process:

termux-wake-lock

🛠 Troubleshooting

If an install fails or hangs

Re-run the script: curl -sSL https://raw.githubusercontent.com/niyazmft/droid-ai-toolkit/main/install.sh | bash — it is safe to run again.
Select the same tool from the menu and choose [R] Repair.
Check the log at ~/droid_ai_toolkit.log for the exact failure reason.
Free up resources: Close other apps, ensure Wi-Fi is stable, and verify you have enough free storage.

Common Issues

Telegram Plugin Not Available: This toolkit attempts to pre-fix this. If it persists, finish onboarding and run: openclaw channels add --channel telegram.
Homebrew Recommendations: Ignore them. Homebrew is not supported on Android. Use pkg install <package> for any missing dependencies.
Node.js Errors: Run the toolkit's Install/Repair option to reset environment locks and paths.
Ollama Not Found After Install: Restart Termux or run source ~/.bashrc to refresh your PATH.
Hermes/Nanobot Fail on armv8l: Expected — these tools require Rust compilation via maturin, which does not support the armv8l architecture. Use an aarch64 device instead.
Paperclip LMK Kill During Install: Expected on 3–4GB RAM devices. The installer detects the kill, verifies packages are present, and continues. If it fails entirely, ensure you have at least 2GB free RAM before starting.

🛠 Code Quality

This project implements a "Zero-Waste" and "Self-Healing" quality gate to maintain high standards for all contributions.

Tools Used

ESLint v10: Modern JavaScript and JSON linting via Flat Config.
Stylelint: Standardized CSS quality checks.
Markdownlint: Documentation consistency enforcement.
ShellCheck: Bash script static analysis.
Husky & lint-staged: Automated pre-commit hooks to auto-fix code.
Self-Healing: Custom Python scripts to safely refactor unused code.

Usage

Run the full quality audit locally:

pnpm run lint:all

📄 License

Distributed under the MIT License. See LICENSE for more information.

Name		Name	Last commit message	Last commit date
Latest commit History 68 Commits
.github/workflows		.github/workflows
.husky		.husky
.jules		.jules
assets		assets
scripts		scripts
.gitignore		.gitignore
.markdownlint.json		.markdownlint.json
.markdownlintignore		.markdownlintignore
.shellcheckrc		.shellcheckrc
.stylelintrc.json		.stylelintrc.json
AGENTS.md		AGENTS.md
ATTRIBUTIONS.md		ATTRIBUTIONS.md
LICENSE		LICENSE
README.md		README.md
build_paperclip_dist.sh		build_paperclip_dist.sh
eslint.config.mjs		eslint.config.mjs
install.sh		install.sh
package.json		package.json
paperclip_manual_install.sh		paperclip_manual_install.sh
pnpm-lock.yaml		pnpm-lock.yaml

Folders and files

Latest commit

History

Repository files navigation