BISHENG is an open-source LLM DevOps platform designed for enterprise-grade AI applications, offering a comprehensive suite of tools for GenAI workflows, RAG, agents, model management, and more. It targets businesses seeking to build and deploy complex AI solutions, providing an integrated framework that simplifies orchestration and enhances user control.

How It Works

BISHENG features a unique, independent workflow orchestration framework that supports complex logic like loops, parallelism, and conditional execution, visualized as a flowchart. This contrasts with other platforms that may rely on separate modules or bot invocations. A key differentiator is its "human-in-the-loop" capability, allowing user intervention during workflow execution, even in multi-turn conversations, which is not typically found in end-to-end execution systems.

Quick Start & Requirements

• Install: git clone https://github.com/dataelement/bisheng.git followed by cd bisheng/docker and bash docker-compose up -d.
• Prerequisites: Docker 19.03.9+, Docker Compose 1.25.1+.
• Recommended Hardware: 18 virtual cores, 48GB RAM. Minimum: 4 CPU cores, 8GB RAM.
• Default Components: Installs ES, Milvus, and Onlyoffice.
• Access: http://IP:3001 after startup.
• Docs: Self-hosting

Highlighted Details

• Supports hundreds of components and thousands of parameters for deep optimization in complex enterprise scenarios.
• Includes high-precision document parsing models for printed text, handwriting, rare characters, tables, and seals, deployable privately.
• Offers enterprise-grade features like RBAC, SSO/LDAP, high availability, monitoring, and security controls.
• Provides an open repository for sharing application cases and best practices.

Maintenance & Community

The project acknowledges contributions from langchain, langflow, unstructured, and LLaMA-Factory. Community discussion is encouraged.

Licensing & Compatibility

The README does not explicitly state the license. Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The project is described as "made by Chinese" and includes links to Chinese, English, and Japanese language versions of the README. Specific details on the license and commercial usage terms are not readily available.