PocketStrike-AI

🧠 AI Agent that gives Termux a Brain

PocketStrike AI is a highly optimized, fully featured super lightweight AI agent core designed to bridge high-level reasoning with low-level Android/Linux device APIs. It is built to turn Termux into an autonomous system operator without bogging down your mobile device.

It couples a gorgeous, responsive, glassmorphic chat interface with an advanced ReAct (Reasoning and Action) Function Calling Framework and native Model Context Protocol (MCP) support. This allows you to inspect your phone’s system parameters, run subnet-wide network sweeps, execute background crons, dump active UI layouts for device automation, run sandboxed Python scripts, search the web using RAG, and connect to remote tool servers over SSE (Server-Sent Events).

Additionally, it integrates a Telegram Bot backend with unified session tracking, allowing you to trigger any of these system tools, check background schedules, or query your AI models remotely from your Telegram app.

⚡ Why is it Super Lightweight?

• Zero Local AI Inference Overhead: Instead of running massive, hot-running local LLMs on your mobile CPU (which consumes 4GB+ RAM and drains battery in minutes), PocketStrike AI acts as an intelligent orchestrator. It runs a lightweight ReAct state engine locally and delegates heavy token processing to remote API endpoints or local Ollama servers.
• Minimal Memory Footprint: The background Flask server is highly optimized, consuming only 30MB - 50MB of RAM under active loads.
• Ultra-Fast Vanilla Frontend: The user interface is crafted using Vanilla HTML, CSS, and JS (no heavy layout engines like React or Tailwind), loading instantly even on older budget Android phones.
• Efficient Async I/O: High-speed utilities (like network sweeps and port scanners) run via custom parallel Python threads, completing sweeps in seconds with negligible CPU usage.

📋 System Requirements & Storage Footprint

🚀 Daily Quick Launch (One-Liner)

If you already completed the setup, open Termux and run this single command to launch your dashboard:

cd ~/PocketStrike-AI && bash launch.sh

📂 Project Structure

📂 PocketStrike-AI/
├── 📄 LICENSE                # Open-source MIT License terms
├── 📄 README.md              # Detailed documentation, guides, and tool specifications
├── 📄 install.sh             # Dependency installer (Python, Git, Termux-API, Nmap, Curl, etc.)
├── 📄 launch.sh              # Terminal-based visual launcher dashboard and status check
├── 📄 server.py              # Main Flask server, AI ReAct framework, and Telegram bot loop
├── 📄 setup.py              # CLI Setup Wizard for API keys and Telegram bot options
├── 📂 static/                # Static assets for the Web chat interface
│   ├── 📄 script.js          # Web event handling, EventSource streaming, and markdown parsing
│   └── 📄 style.css          # obsidian-dark / royal-blue responsive layout stylesheet
└── 📂 templates/             # HTML Templates
    └── 📄 index.html         # Glassmorphic, modern chat dashboard interface

✨ Features

• Always-On Background Voice Assistant ("Hey Strike"): Hands-free background voice listener daemon running in Termux that detects wake word "Hey Strike" while watching movies or using other apps, processes commands via the ReAct agent, and speaks answers back aloud via Android TTS & lockscreen notification cards. Also features Web Speech API Voice Mode in the Web UI.
• Unified Chat History Engine: The agent maintains a single unified mind across platforms. Messages sent via Telegram are instantly visible in the Web interface, and vice-versa, synchronizing context in real-time.
• 60-Message sliding window: Supports deep conversation tracking by passing up to the last 60 message states to the LLM API, while preserving complete logs on local storage.
• Self-Evolving Long-Term Memory: The AI dynamically updates agent/user.md, agent/memory.md, and agent/agent.md files via background conversation reflections. These files are re-injecting on every turn to adapt to your preferences.
• Active Threat Intrusion Sentinel: Runs a background daemon thread that checks /proc/net/arp and network metrics for active ARP Spoofing/MITM threats, automatically sending system lockscreen banners (via termux-api notification), text-to-speech warnings, and Telegram bot alerts.
• Rich Media Chat Previews: The Web UI automatically renders dynamic, cached-busted images (<img>) and video players (<video>) inside chat bubbles whenever workspace screenshots, photos, or recordings are generated or mentioned.
• Self-Dependency Doctor & Auto-Installer: Runs a full environment diagnostic (check_system_health) checking for required CLI utilities and Python modules, and can automatically execute non-interactive pkg install and pip install repairs if requested.
• Persistent Background Scheduler: A daemon thread checks for scheduled reminders or recurring cron intervals (e.g. "remind me to blink my eyes every 1 minute") and alerts you locally or via Telegram.
• Audio Beep & System Fallback: If the device lacks Termux:API, the scheduler utilizes the ASCII Bell code (\a) to beep/vibrate Termux natively, and dynamically redirects screen alerts to Telegram.
• Robust ADB/Shizuku execution: Features parameter-safe parsing and automatic standard ADB fallbacks if the Shizuku emulator binder (rish) becomes unauthorized or goes offline.
• Web & Network Security Auditors: Built-in scanners to detect active Wi-Fi Man-in-the-Middle (ARP Spoofing) attacks, audit VPN connection leaks, and evaluate SSL certificates and HTTP security headers.
• Stateful persistent Terminal Session: Maintains directory changes (cd) and environmental variables across multiple command runs, operating inside a persistent background shell.
• Subnet-Wide Network Sweeps: Scans class C subnets (1-254) in less than 3 seconds using 80 parallel workers, resolving device hostnames automatically.
• Parallel Port Scanner: Checks up to 100 ports concurrently on local hosts using thread pools, automatically identifying active service names (SSH, HTTP, Database, etc.).
• Deep RAG Web Search: Scrapes DuckDuckGo HTML and automatically fetches the actual main body text of the top 2 web pages in the background. It feeds this text directly to the AI, bypassing knowledge cutoff limitations.

🛠️ Installation & Setup on Termux

Follow these steps to configure your Termux server:

Step 1: Install Git in Termux

Run this command to install git

pkg install git

Step 2: Clone and Run the Installer

Launch Termux and run this one-line command to install all basic dependencies (Python, Git, Flask, Requests, Termux-API, Nmap, Dnsutils, Curl, Net-Tools, Iproute2, and Traceroute):

git clone https://github.com/AbuZar-Ansarii/PocketStrike-AI.git && cd PocketStrike-AI && sed -i 's/\r$//' install.sh && bash install.sh

Note: During installation, the script will request Android Storage Permissions (termux-setup-storage). Tap "Allow" on the system popup.

[!TIP] Getting "CANNOT LINK EXECUTABLE" Error? This is a common Termux issue caused by a corrupted/outdated Termux environment (e.g. if installed from Google Play instead of F-Droid). Fix your Termux package manager by running:

apt update && apt full-upgrade

If Termux is completely locked, uninstall your current version and download the official updated build from F-Droid or GitHub Releases.

Step 3: Configure Setup Wizard

Run the launcher:

bash launch.sh

1. Select option 1 to run the Setup Wizard.
2. Select your AI provider and fill in the details:
   • Google Gemini (gemini-1.5-flash, gemini-1.5-pro)
   • OpenAI (gpt-4o, gpt-4-turbo, etc.)
   • Anthropic Claude (claude-3-5-sonnet, etc.)
   • Ollama (Auto-configures to http://localhost:11434 for offline local models like Llama3, Phi3, Gemma, etc.)
   • OpenRouter & Custom APIs (compatible with OpenAI format)
3. Select whether to enable Telegram Bot integration (requires your Bot Token and Chat ID).

Step 4: Run the Server

Choose option 2 from the launcher. Open the Local URL in your phone's browser or the Network URL on your PC to start chatting!

🎙️ Always-On Voice Assistant ("Hey Strike")

PocketStrike AI features a background voice assistant that allows phone-wide, hands-free interaction:

• Background Wake-Word Listener: Simply say "Hey Strike", "Strike", or "Hey Pocket Strike" while watching movies, playing games, or using any app on your phone.
• Auto Tool Execution & Speech Response: