CustomChar

CustomChar - Your customized AI characters

• Repository: https://github.com/nrl-ai/CustomChar.
• Built With: GGML, llama.cpp (LLM, with LLaMA-v2), whisper.cpp (speech recognition).

Your customized AI characters - personal assistants on any hardware! This project aims to be a framework to enable the creation of AI characters that can be used in games, simulations, virtual assistants, and other applications without depending on specific platforms, cloud services, or specialized hardware.

2. Updates

• 2023-08-06: Load character configuration from a file.
• 2023-07-30: Record a video with "record a video" and "stop recording".
• 2023-07-30: Added OpenCV + camera view.
• 2023-07-29: Added a plugin executor to open apps.
• 2023-07-25: Added UI with Dear ImGui.
• 2023-07-23: Documentation website is online at https://docs.customchar.com.
• 2023-07-22: Voice chat with your Llama-V2 model. TTS is supported on MacOS only.

3. Features

• Build AI characters from the knowledge database :
  • LLamaCpp for LLM. Chat with Llama-V2 on all platforms.
  • Knowledge indexing and searching (EmbedDB + SQLite).
  • Search knowledge DB to add into prompt.
• Talk to your character ??? :
  • STT with whisper.cpp. Now supports speech recognition on all platforms.
• Hear from your character ??? :
  • TTS with "say" on MacOS. Note: "say" is not available on other platforms (Windows, Linux)
  • TTS with bark.cpp. Expected to be available on all platforms.
• Plugin executor:
  • Open apps
  • Record video
• Perception engine to see the world through the camera ???"
  • Support video capturing
  • Object detection
• Add GUIs for the character interaction.
  • Dear ImgUI
  • Web-based GUI
• Add a GUI for the character creation.
• Create virtual characters that can be used in games, simulations, and other applications.
  • Iron Man (WIP)
  • Batman
  • Superman
• Add bindings to other languages:
  • Python
  • C# (Unity)
  • Javascript (On-browser)

4. Build

Clone the repository:

git clone https://github.com/nrl-ai/CustomChar --recursive
cd CustomChar

Install dependencies:

• CMake
• C++ 14 compiler

On macOS: Using Homebrew

brew install sdl2 glew glfw3
brew install opencv
brew install ffmpeg # For video recording

On Ubuntu:

sudo apt-get install libsdl2-dev libglew-dev libglfw3-dev
sudo apt-get install libopencv-dev
sudo apt-get install ffmpeg # For video recording

On Windows: Using vcpkg, Git Bash, and FFmpeg.

vcpkg install sdl2:x64-windows glew:x64-windows glfw3:x64-windows
vcpkg install opencv[contrib,nonfree,ffmpeg,ipp]:x64-windows --recurse

Build the CustomChar executable:

mkdir build
cd build
cmake ..
make

5. Usage

Download the models before running the program:

bash scripts/download-models.sh

Run the program:

./customchar

6. Build documentation

• Step 1: Install doxygen first.

• Step 2: Build the documentation:

doxygen Doxyfile.in

• Step 3: Deploy html documentation from docs/_build/html.

• Step 4: Our latest documentation is deployed at https://docs.customchar.com.

7. Contribution

We welcome all contributions to this project.

• For coding style, please follow the style of the existing code.
• Install clang-format for auto formatting the code.
• Install pre-commit for the auto-formatting hook or manually run the script scripts/format-code.sh to format the code.

pre-commit install # Install pre-commit hook
bash scripts/format-code.sh # Format code (C++ styling)

8. Star history

9. References

• Spinner: https://github.com/dalerank/imspinner.
• GGML: https://ggml.ai/.
• llama.cpp: https://github.com/ggerganov/llama.cpp.
• whisper.cpp: https://github.com/ggerganov/whisper.cpp.
• LLaMA-v2: https://ai.meta.com/llama/.