Skip to content

Privacy-first Android AI assistant with on-device LLM chat, encrypted memory, and a modular plugin framework—built for offline use, with automation coming soon. No cloud required.

License

Notifications You must be signed in to change notification settings

Godzilla675/NeuroVerse

 
 

Repository files navigation

NeuroVerse

Offline Android AI chat with a modular plugin framework (automation coming soon).

NeuroVerse is a privacy‑first Android app that runs fully on‑device and showcases a plugin‑based architecture. The current public milestone is chat‑only via the first‑party AI Chat plugin. The command/automation engine exists in the codebase but is disabled and not user‑facing yet.


Try it now

First‑run: install the APK, grant permissions, open AI Chat and start chatting. No cloud account required.


Preview :: For Install Plugin :: Tool-Calling

NeuroVerse chat screenshot NeuroVerse Tool-Calling


Status

  • Current: Chat‑only experience using the built‑in AI Chat plugin.
  • Planned: Command JSON generation, action validation, and on‑device UI automation are in development and will ship in future releases.

What you can do today

  • Local AI chat with a clean Material 3 UI.
  • Load first‑party sample plugins to explore the plugin lifecycle and UI embedding.
  • Build the app from source and experiment with the plugin API while automation is WIP.

Official plugins

First‑party plugins live here: https://github.com/Siddhesh2377/Neuro-V-Sys-Plugins

Available modules:

  • ai-chat (current focus)
  • app-io (under active development)
  • demo-macro (prototype; not user‑facing yet)

Each module can be exported to a plugin.zip containing plugin.aar and Manifest.json, which NeuroVerse can import at runtime.

Plugin integration overview

  • Manifest: declare entryClass in src/main/Manifest.json (for example, com.mp.ai_chat.ChatScreenPlugin).
  • Packaging: produce plugin.zip with plugin.aar and Manifest.json.
  • Loading: import the zip from within NeuroVerse; the host validates and registers it.
  • UI: plugins may provide Jetpack Compose screens rendered within the app.

See the plugin repo README for the exportPluginZip task and packaging details.


Features (current vs planned)

Area Current (public) Planned (roadmap)
Chat AI Chat plugin with local UI and message history Tool‑aware chat, plugin suggestions
Models Local model support under evaluation; dev builds experiment with GGUF runtimes Curated on‑device models with auto‑download and versioning
Plugins Runtime loading of first‑party zips Community plugin gallery, ratings, templates
Automation Not available in public builds JSON command generation, validator, Accessibility‑based execution
Memory Basic chat history Encrypted NeuronTree with scoped recall and user controls

Architecture (high level)

Layer Responsibility
UI Jetpack Compose app shell and plugin screens
Plugin Host Discovers, validates, and mounts plugin entry points
Inference (chat) Local model adapter and chat loop
Command Engine Present but disabled in public builds
Storage Room‑backed data; encryption work in progress

Installation

Install APK

  1. Download from the latest release.
  2. Install on Android 14 or newer.
  3. Open AI Chat and start a conversation.

Build from source

  1. Clone the repository and open in Android Studio.
  2. Sync Gradle and build :app.
  3. Run on a physical device for best performance.

Developing plugins

  1. Start from the template in Neuro‑V‑Sys Plugins.
  2. Implement your entryClass and any Compose UI.
  3. Run the export task to produce plugin.zip.
  4. Import the zip into NeuroVerse from the Plugins screen.

Open a PR to add your plugin to the community list once automation support ships broadly.


Roadmap

  • Ship command generation and validator in a public build
  • Accessibility‑based action runner with safety guardrails
  • In‑app plugin gallery and submission flow
  • Basic task inspector and execution logs
  • Small, pre‑bundled model for first‑run offline chat

Feedback is welcome in Issues and Discussions.


Security and privacy

  • Chat runs locally; no cloud requirement for the core experience.
  • No analytics or telemetry in the core project.
  • Always review third‑party plugin code before installing.

Contributing

Contributions are welcome. Focus areas: plugin API, chat UX, model adapters, storage, and future automation.

  1. Fork the repository.
  2. Create a feature branch.
  3. Document changes.
  4. Open a pull request with a clear description and screenshots.

Good first issues are labeled accordingly.


License

NeuroVerse is licensed under the Apache License 2.0.

See LICENSE and NOTICE. Third‑party components remain under their original licenses.


Author

Siddhesh Sonar Android Developer · AI Enthusiast · Open‑Source Contributor GitHub: https://github.com/Siddhesh2377


Acknowledgements

  • llama.cpp and the GGUF ecosystem
  • SmolChat-Android
  • JetBrains
  • Android Open Source Project
  • GitHub

About

Privacy-first Android AI assistant with on-device LLM chat, encrypted memory, and a modular plugin framework—built for offline use, with automation coming soon. No cloud required.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Kotlin 93.9%
  • C++ 5.3%
  • Other 0.8%