-
🎛️ Controllable Memory: Take control of Loyal Elephie's memory! You decide which moments to save, and you can easily edit the context as needed. It is your second-brain for episodic memory. ✍️
-
🔍 Hybrid Search: Experience the powerful combination of Chroma and BM25 for efficient searches! It's also optimized for handling date-based queries. 📅
-
🔒 Secure Web Access: With a built-in login feature, only authorized users can access your AI companion, ensuring your conversations remain private and secure over the internet. 🛡️
-
🤖 Streamlined LLM Agent: Loyal Elephie uses XML syntax with no function-calling required. It is also optimized for cached generation and works smoothly with great local LLMs using Llama.cpp or ExllamaV2. 💬
-
📝 (Optional) Markdown Editor Integration: Connect with online Markdown editors to view the original referred document during chats and experience real-time LLM knowledge integration after editing your notes online. 🌐
Loyal Elephie supports both open and proprietary LLMs and embeddings serving as OpenAI compatible APIs.
Warning: This project was originally designed for Linux. Compatibility with Windows or macOS has not been fully tested. You are welcome to report any issues, but they may not be resolved immediately.
Meta-Llama-3-70B-Instruct.Q4_K_S.gguf is used when capturing the below screenshots
With SilverBulletMd, you can edit a note on the browser and then let Loyal Elephie rememeber it!
The UI is modified from https://github.com/mckaywrigley/chatbot-ui-lite, credits to the author Mckay Wrigley!
1. Clone Repo
git clone https://github.com/v2rockets/Loyal-Elephie.git2. Install Frontend Requirments
cd frontend
npm i3. Configure Login Users
frontend/users.json
[{
"username":"admin",
"password":"admin"
}]4. Install Backend Requirents
cd backend
pip install -r requirements.txt5. Configure Backend Settings
# backend/settings.py
NICK_NAME = 'Peter' # This is your nick name. Make sure to set it at the beginning and don't change so that LLM will not get confused.
CHAT_BASE_URL = 'https://api.openai.com/v1' # Modify to your OpenAI compatible API url
CHAT_API_KEY = 'your-api-key'
CHAT_MODEL_NAME = "gpt-3.5-turbo"6. Run App
frontend:
cd frontend
npm run build
npm run startbackend:
cd backend
python app.py- use "Save" button to save the current conversation into Loyal Elephie's memory.
- use "Reset" button to clear the current conversation (not affecting saving status, the same as refreshing page)
- click on the titles in "Reference" to navigate to the corresponding Markdown notes (but SilverBulletMd or another web Markdown editor has to be hosted and configured).
Some of the workable local LLMs tested:
- OpenHermes-2.5-Mistral-7B
- Phi-3-Medium-4k-instruct
- Mixtral-8x7B-Instruct-v0.1
- c4ai-command-r-v01
- Meta-Llama-3-70B-Instruct (Best so far)
For those who need hand-on local embedding API, an embedding server example is added to "external_example".