AI Doomsday Toolbox

An offline AI toolbox for Android that turns one phone, or a cluster of old phones, into a local AI workstation.

Run local LLMs, Whisper transcription, image generation, distributed inference, dataset creation, offline knowledge tools, and AI-powered utilities directly on Android. The project is built for people who care about privacy, edge AI, on-device AI, distributed compute, and squeezing useful work out of old phones instead of leaving them in a drawer.

Why It's Different

Most Android AI apps focus on one feature. AI Doomsday Toolbox goes further: it combines offline AI on Android with distributed inference, phone-to-phone model sharing, and workflows that can reuse old phones as a low-cost Android cluster or pocket edge-compute setup.

If you are searching for a local LLM on Android, image generation, AI agents to build your projects, offline AI assistant, mobile HPC experiment, phone cluster, Android distributed compute app, or a way to reuse old phones for edge AI workloads, this project is built in that direction.

Highlights

• Offline AI on Android with local LLM, Whisper, image generation, upscaling, and media tools
• Distributed inference features for coordinating multiple Android devices on the same network
• Built-in Ollama manager, llama native chat, and remote summary tools
• Dataset creator that turns text and PDFs into cleaned, rated instruction-answer pairs with Alpaca export
• Termux + proot tool environment with install helpers, SSH workflows, in-app webview access, and file management
• AI agent workspace with custom tools, custom agents, and project memory
• Tama virtual pet systems with adventures, farming, chat, and persistent memories
• Android share-intent support for PDFs, videos, images, and audio files

Features

Distributed Inference And Android Cluster Workflows

• Run distributed inference features across multiple Android devices
• Experiment with turning old phones into a low-cost phone cluster for local AI workloads
• Monitor worker/master flows from inside the app
• Share models and services over the local network for offline collaboration

Local AI Chat

• Chat with local LLMs on Android
• Support GGUF-based llama.cpp workflows
• Connect to llama.cpp-compatible servers and llama-server backends through a native in-app chat UI
• OpenAI-compatible local server mode on port 8080
• Multiple model support with switching from the app interface
• Optional LAN-visible server behavior through settings when needed
• Use Ollama-compatible workflows where they fit best for your setup
• Keep inference on-device or on your own local network instead of relying on cloud chat

Ollama Manager

• Add and manage Ollama servers from inside the app
• Pull, inspect, copy, delete, and organize models
• View and edit Modelfiles
• Create derived models without leaving the Android interface

Benchmarking

• Benchmark your device for LLM workloads
• Compare thread counts to find the best number of threads for a specific model
• Save benchmark results for later reference
• Use real llama benchmarking output instead of guessing performance

Dataset Creator

• Import .txt and PDF files
• Split source material into chunks for cleaner processing
• Clean chunks before question generation
• Generate five questions per chunk using neighboring chunk context for better continuity
• Generate answers, rate the pairs, and export the best entries
• Export in Alpaca JSON format
• Customize the prompts used for cleaning, question generation, answer generation, and review

Termux Tools And Proot Workflows

• Connect to a proot distro over SSH from inside the app
• Follow in-app setup help for enabling SSH inside the proot environment
• Install predefined tools such as Ollama, Open WebUI, Big-AGI, Oobabooga text-generation-webui, FastSDCPU, and experimental A1111 workflows
• Open compatible tools in an in-app webview
• Manage remote files with the built-in Termux file manager
• Optionally expose some services outside localhost when your workflow needs LAN access

Note: A1111 / AUTOMATIC1111 support is still experimental and actively being worked on, same with oobabooga since recent update of their repo.

AI Agent Workspace

• Run an AI coding/workflow agent environment powered by Termux and Ollama-compatible backends
• Create custom tools and custom agents
• Keep project-specific workspace memory and task context
• Build reusable automation flows around your own projects

PDF, Video, And Summary Tools

• Extract text from PDFs with OCR fallback when needed
• Summarize PDFs, videos, and transcription workflows
• Use Ollama and/or llama.cpp-compatible remote backends for summary generation
• Tune prompts, context limits, output length, and related summary parameters per workflow

Audio, Video, And Subtitle Tools

• Transcribe audio and video with Whisper
• Support multiple languages and model sizes from lightweight to larger accuracy-focused options
• Summarize video content after transcription
• Extract audio from videos with FFmpeg as part of summarization/transcription workflows
• Burn subtitles into video with styling controls such as font, color, and position
• Process media directly from Android share intents

Image Generation And Upscaling

• Generate images with Stable Diffusion workflows directly on Android
• Includes SD 1.5, SD 2.1, SDXL, and FLUX-oriented workflows
• Adjustable generation settings such as steps, CFG scale, dimensions, and tiling
• Upscale images and videos with RealESRGAN-based tools
• Multiple scale factors are available depending on the selected model
• Use FastSDCPU in Termux/proot workflows for additional image-generation setups
• Experiment with A1111-style web UI workflows through the Termux tools area

Offline Knowledge, Sharing, And Utilities

• Browse offline knowledge bases with Kiwix and ZIM file support
• Download ZIM files through the catalog or import them from internal storage
• Built-in Kiwix server support on port 8888
• Optional LAN access so other devices on your network can connect to shared content
• Share models and ZIM files over LAN with web UI and QR-based connection flows
• Export shared files to storage when needed
• Create and manage notes with Markdown support
• Automatic note creation for summaries and transcription workflows
• Use Android share intents to send PDFs, videos, images, and audio into the app’s processing flows

Tama Virtual Pet

• Raise a Tamagotchi-like pet inside the app
• Go on AI-generated adventures
• Work, farm, and interact across multiple gameplay systems
• Talk to your pet and build memories over time
• Discover different personalities and long-term companionship mechanics

Built With

Core AI And Media Projects

• llama.cpp
• whisper.cpp
• stable-diffusion.cpp
• FFmpeg
• Kiwix-tools
• Real-ESRGAN

Compatible And Integrated Tooling

• Ollama
• Open WebUI
• Big-AGI
• Oobabooga text-generation-webui
• FastSDCPU
• AUTOMATIC1111 / stable-diffusion-webui
• EasyDataset
• Termux

Android Stack

• Kotlin with Jetpack Compose for the UI
• Room for local persistence
• NanoHTTPD for embedded local servers
• ZXing for QR code generation
• ML Kit for OCR
• Apache PDFBox for PDF handling

Architecture

• Native binaries built for arm64-v8a
• Foreground services for long-running AI tasks
• Unified notification flows for background processing
• SAF-based file handling for imports and exports

Getting Started

Join The Google Play Beta

If you want the easiest install path, you can join the Google Play beta here:

AI Doomsday Toolbox on Google Play

The Google Play version uses an Android App Bundle, so the installation is usually smaller than downloading a universal package manually. Joining the beta also helps a lot by improving testing coverage, surfacing device-specific issues, and making it easier to validate updates before wider releases.

Requirements

• Android 8.0+ (API 26)
• arm64-v8a device
• More RAM and storage if you plan to run larger local models
• Additional devices on the same network if you want to experiment with distributed inference and phone-cluster workflows

Build From Source

git clone https://github.com/ManuXD32/AI-Doomsday-Toolbox.git
cd AI-Doomsday-Toolbox

./gradlew assembleDebug

For release bundles, the project expects Java 21 and the usual signing environment variables:

KEYSTORE_PATH=/absolute/path/to/your-release.keystore \
JAVA_HOME=/usr/lib/jvm/java-21-openjdk \
KEYSTORE_PASSWORD=your_keystore_password \
KEY_PASSWORD=your_key_password \
./gradlew :app:bundleRelease

Models And Content

Models and large offline content are managed separately inside the app. The project includes workflows for downloading, importing, and organizing models and ZIM content based on the feature you want to use.

Contributing

Contributions are welcome. If you want to improve a feature, fix a bug, or help with documentation, pull requests are appreciated.

Support

If this project helps you, you can support development here:

• Ko-fi
• PayPal

Disclaimer

This project is provided as-is, without any guarantee that it will be error-free, fit for a particular purpose, or safe for every workflow. You are responsible for how you use it, including any commands, model workflows, remote connections, generated output, or automation built on top of it.

The author is not responsible for misuse of the app, data loss, device issues, network exposure, third-party tool behavior, or damage caused by incorrect configuration, generated content, or user actions.

License

This project is licensed under the Apache License 2.0.

Author

ManuXD32 - GitHub

• Built with the help of codex and antigravity