Inspired by Ollama, Apple MlX projects and frustrated by the dependencies from external applications like Bing, Chat-GPT etc, I wanted to have my own personal chatbot as a native MacOS application. Sure there are alternatives like streamlit, gradio (which are based, thereby needing a browser) or others like Ollamac, LMStudio, mindmac etc which are good but then restrictive in some means (either by license, or paid or not versatile). Also I wanted to enjoy both Ollama (based on llama.cpp) and Mlx models (which are suitable for image generation, audio generation etc and heck I own a mac with Apple silicon 👨🏻💻) through a single uniform interface.
All these lead to this project (PyOllaMx) and another sister project called PyOMlx.
I'm using these in my day to day workflow and I intend to keep develop these for my use and benifit.
If you find this valuable, feel free to use it and contribute to this project as well. Please ⭐️ this repo to show your support and make my day!
I'm planning on work on next items on this roadmap.md. Feel free to comment your thoughts (if any) and influence my work (if interested)
MacOS DMGs are available in Releases
PyOllaMx : ChatBot application capable of chatting with both Ollama and Apple MlX models. For this app to function, it needs both Ollama & PyOMlx macos app running. These 2 apps will serve their respective models on localhost for PyOllaMx to chat.
PyOMlx : A Macos App capable of discovering, loading & serving Apple MlX models downloaded from Apple MLX Community repo in hugging face 🤗
- Install Ollama Application & use Ollama CLI to download your desired models
ollama pull <model name>
ollama pull mistral
This command will download the Ollama models in a known location to PyOllaMx
- Install MlX Models from Hugging Face repo.
use hugging-face cli
pip install huggingface_hub hf_transfer
export HF_HUB_ENABLE_HF_TRANSFER=1
huggingface-cli download mlx-community/CodeLlama-7b-Python-4bit-MLX
This command will download the MlX models in a known location to PyOllaMx
- Now simply open the PyOllaMx and start chatting
- Dark mode support - Toggle between Dark & Light mode with a click of the icon
- Model settings menu - Brand new settings menu to set the model name and the temperature along with Ollama & MlX model toggle
- Streaming support - Streaming support for both chat & search tasks
- Brand New Status bar - Status bar that displays the selected mode name, model type & model temperature
- Web search enabled for Apple MlX models - Now you can use Apple MlX models to power the web search when choosing the search tab
Click the release version link above ☝🏻 to view demo gifs explaining the features.
- Web search capability (powered by DuckDuckGo search engine via https://github.com/deedy5/duckduckgo_search) a. Web search powered via basic RAG using prompt engineering. More advanced techniques are in pipeline b. Search response will cite clickable sources for easy follow-up / deep dive c. Beneath every search response, search keywords are also shown to verify the search scope d. Easy toggle between chat and search operations
- Clear / Erase history
- Automatic scroll on chat messages for better user experience
- Basic error & exception handling for searches
Limitations:
- Web search only enabled for Ollama models. Use dolphin-mistral:7b model for better results. MlX model support is planned for next release
- Search results aren't deterministic and vary vastly among the chosen models. So play with different models to find your optimum
- Sometimes search results are gibberish. It is due to the fact that search engine RAG is vanilla i.e done via basic prompt engineering without any library support. So re-trigger the same search prompt and see the response once again if the results aren't satisfactory.
Click the release version link above ☝🏻 to view demo gifs explaining the features.
- Auto discover Ollama & MlX models. Simply download the models as you do with respective tools and pyOllaMx would pull the models seamlessly
- Markdown support on chat messages for programming code
- Selectable Text
- Temperature control