ToolNeuron is built across multiple repositories. Here's how they fit together.
The main Android app. UI, chat engine, RAG pipeline, plugin system, AI memory, backup/restore — everything the user interacts with.
Core AI system libraries that power ToolNeuron's inference. Native modules for GGUF model loading, Stable Diffusion, TTS, and embedding generation. These get compiled into .aar libraries consumed by the main app.
Android-specific fork/integration of llama.cpp. Handles the JNI bridge between Kotlin and the native C++ inference engine. Optimized for ARM64 with device-aware thread management.
ToolNeuron (app)
|
+-- Ai-Systems-New (native AI libraries)
| |
| +-- llama.cpp-android (GGUF inference JNI bridge)
| +-- Stable Diffusion engine
| +-- TTS engine (ONNX Runtime)
| +-- Embedding engine
|
+-- ums (Unified Memory System, in-repo module)
+-- neuron-packet (encrypted RAG format, in-repo module)
+-- system_encryptor (native crypto, in-repo module)
Contributions are welcome across all three repos. If you're working on:
- UI, chat, plugins, RAG, memory — contribute to ToolNeuron
- Inference performance, model loading, native crashes — contribute to Ai-Systems-New or llama.cpp-android
See the main README for contribution guidelines.
Siddhesh Sonar — creator and primary maintainer of all three repositories.
Want to see your name here? Check the open issues on any of the repos above and submit a PR.