LoLLMs

One tool to rule them all!

A multi-user FastAPI backend and responsive HTML/Tailwind CSS frontend application designed to provide a chat interface powered by the lollms_client library. It features integrated Retrieval-Augmented Generation (RAG) using safe_store, MCPs, Personalities system, multimodal chat, user personalities, a friend system, direct messaging, and enhanced sharing capabilities.

Live Project: https://github.com/ParisNeo/lollms

Overview

This project aims to provide a self-hostable, user-friendly chat interface that can connect to various Large Language Models (LLMs) supported by lollms-client. It features multi-user support with basic authentication, persistent discussions, RAG functionality, multimodal chat (image input), customizable user personalities, a friend system with direct messaging, administrative user management, and sharing of datastores and personalities.

✨ Features

• Multi-User Support: Secure login via Token based authentication. Each user has their own isolated data.
• Persistent Discussions: Chat histories are saved per user (SQLite3 database) and can be revisited, renamed, starred, and deleted.
• LLM Integration: Uses lollms-client to interact with various LLM backends.
• Streaming Responses: AI responses are streamed for a real-time experience.
• Multimodal Chat: Upload images with text prompts for vision-capable models.
• User Personalities (System Prompts):
  • Create, edit, and delete custom personalities (system prompts with name, category, author, icon, etc.).
  • Select an active personality to guide LLM responses.
  • View and use public (system-provided) personalities.
  • Share (send a copy of) owned personalities with friends.
• Retrieval-Augmented Generation (RAG):
  • Multiple DataStores per user for organizing RAG documents.
  • Upload documents (.txt, .pdf, .docx, .html, etc.) to specific DataStores.
  • Toggle RAG usage per discussion, selecting a specific DataStore.
  • Manage DataStores (create, rename, delete) and their indexed documents.
  • Share DataStores with friends (read-only query access).
• Friend System & Direct Messaging (DM):
  • Send, accept, and reject friend requests.
  • View friends list and manage friendships (unfriend, block/unblock - block/unblock WIP).
  • Send and receive direct messages with accepted friends.
  • View conversation history with friends.
• Configurable Settings:
  • Users can set their default LLM model, RAG vectorizer, LLM generation parameters (temperature, top_k, etc.), RAG parameters, and personal profile information.
• Admin Panel:
  • Manage users (Add, List, Delete, Reset Password).
  • Initial superadmin user created from config.toml.
• Data Export/Import: Users can export/import their discussions, settings, and metadata.
• Responsive UI: Built with Tailwind CSS, including image previews, Markdown rendering with code highlighting (Highlight.js) and math (KaTeX).
• Dark/Light Theme & Internationalization (i18n).
• SQLite Backend: Central database for users, settings, personalities, friendships, DMs. SafeStore uses its own DBs per DataStore.

🏗️ Architecture

• Backend: FastAPI (Python)
• LLM Communication: lollms-client
• RAG/Vector Store: safe_store (SQLite based per DataStore)
• Main Database: SQLAlchemy with SQLite (users, personalities, friendships, DMs, etc.)
• Authentication: HTTP Basic Auth + Passlib
• Frontend: HTML, Tailwind CSS, Vanilla JavaScript
• Configuration: TOML (config.toml)
• Discussion Storage: YAML files per user

📸 Screenshots

(Add updated screenshots showing the new Friends & Messages modal, DM interface, personality editor, and sharing options)

Login Screen: ![Login Screen](placeholder_login.png)

Main Chat UI (with Personality Selector & RAG): ![Chat UI](placeholder_chat_main.png)

Friends & Messages Modal: ![Friends & Messages Modal](placeholder_friends_dm.png)

Personality Editor Modal: ![Personality Editor](placeholder_personality_editor.png)

Settings Modal (User Profile, LLM/RAG Params): ![Settings Modal](placeholder_settings_new.png)

DataStore Management & Sharing: ![DataStore Management](placeholder_datastore_manage.png)

Admin Panel: ![Admin Panel](placeholder_admin.png)

🚀 Getting Started

Prerequisites

• Python 3.8+
• Git, Pip
• An LLM backend accessible via lollms-client.

Installation

Manual installation

1. Clone:

   git clone https://github.com/ParisNeo/lollms_chat.git
   cd lollms_chat

2. Virtual Environment (Recommended):

   python -m venv venv
   # Windows: venv\Scripts\activate
   # macOS/Linux: source venv/bin/activate

3. Install Dependencies:

   pip install -r requirements.txt

4. Configure config.toml:
   • Copy config_example.toml to config.toml.
   • Crucially: Change password under [initial_admin_user].
   • Set binding_name and default_model_name in [lollms_client_defaults].
   • Review other settings.
5. Run:

   python main.py

Automatic installation

1. ** Windows ** execute install.bat
2. ** Linux/MazcOs ** execute install.sh

Usage

1. Access UI: http://localhost:9642
2. Login: Use initial admin or created user credentials.
3. Chat: Create/select discussions, use RAG, upload images.
4. Personalities: Manage and select active personalities via Settings.
5. Friends & DMs: Access via user menu to manage friends and send messages.
6. DataStores: Manage RAG DataStores and share them via user menu.
7. Settings: Configure user profile, LLM/RAG parameters, active personality.
8. Admin Panel: /admin for user management (admins only).

⚙️ Configuration (config.toml)

• [app_settings]: data_dir, database_url.
• [initial_admin_user]: username, password, optional first_name, email.
• [lollms_client_defaults]: binding_name, default_model_name, LLM parameters, put_thoughts_in_context.
• [safe_store_defaults]: RAG chunking, global_default_vectorizer.
• [server]: host, port.

📚 API Documentation

• Swagger UI: http://localhost:9642/docs
• ReDoc: http://localhost:9642/redoc

🤝 Contributing

Contributions are welcome! Please open an issue or submit a pull request.

📜 License

Apache License 2.0. See LICENSE.