Casibase is an open-source enterprise-grade AI knowledge base and management platform designed for building and deploying AI applications. It targets developers and organizations looking to integrate various large language models (LLMs) and embedding models into their workflows, offering a comprehensive solution for Retrieval-Augmented Generation (RAG) with a user-friendly interface and enterprise features like SSO.

How It Works

Casibase employs a dual-architecture approach with a JavaScript/React frontend for the user interface and a backend built with Golang (Beego) and Python (Flask), utilizing MySQL for data storage. This combination allows for a responsive UI and robust server-side logic. It supports a wide array of LLMs and embedding models from providers like OpenAI, Azure, Hugging Face, Google Gemini, and Claude, enabling flexible model selection and integration within a RAG framework.

Quick Start & Requirements

• Installation typically involves server setup. Detailed instructions are available at https://casibase.org/docs/basic/server-installation.
• Requires a MySQL database.
• Supports various LLMs and embedding models, requiring API keys or local model setup for specific providers.

Highlighted Details

• Supports over 30 LLMs and 10 embedding models from major providers.
• Includes an admin UI with user management and Single-Sign-On (SSO) capabilities.
• Offers a LangChain-like RAG framework for knowledge base creation.
• Provides online demos for both the chatbot and admin UI.

Maintenance & Community

• Community support is available via Discord: https://discord.gg/5rPsrAzK7S.
• Contributions are welcomed through issues and pull requests.

Licensing & Compatibility

• Licensed under the Apache-2.0 license.
• Permissive license suitable for commercial use and integration into closed-source applications.

Limitations & Caveats

The project is actively developed, and while it supports numerous models, specific model versions or configurations might require further testing or adjustments. The README does not detail specific hardware requirements for running local models.