AIBrain is a native Windows desktop application for chatting with GGUF models already installed by Ollama and exploring real-time llama.cpp inference telemetry through a live connectome visualization. The display measures raw next-token logit statistics and generation timing; its graph is a visual layout, not a map of transformer neurons.
It is built with Python and PySide6/Qt. The interface is a real Windows window with a native OpenGL context—there is no React, Electron, browser, or WebView layer.
Requirements: Windows 10/11, Python 3.11+, and at least one locally installed Ollama GGUF model.
py cli\installer.py
.\.venv\Scripts\Activate.ps1
python cli\main.pyAIBrain only runs from its managed virtual environment. The installer creates and populates .venv, checks NVIDIA/CUDA
capability before choosing an inference wheel, and never installs packages into the system Python.
The GUI tools (main.py, diagnostic.py, and analysis.py) keep their attached console for compact runtime messages and
open a loader window before their first background validation completes. Logs are feature-scoped under logs/ (for
example, aibrain.main.log); a detailed crash.<feature>.log is created only after an uncaught exception.
The complete documentation is in docs/:
- Getting started
- User guide
- Models and inference
- Connectome and analysis
- JSON reference
- Native acceleration
- Architecture and development
- Configuration
- Troubleshooting
- Wiki publishing
The Publish documentation to Wiki GitHub Actions workflow publishes this folder to the repository wiki after one has
been enabled.
# Rebuild and verify the native DLL after editing its C source.
py cli\build_native.py --clean
# Build merged AIBrain plus independent main, diagnostic, and analysis folders.
.\.venv\Scripts\python.exe cli\build_dist.py
# Verify Python syntax from the managed environment.
python -m compileall -q cli src testsSee Native acceleration and Architecture and development for full build, verification, and contributor guidance.