Bisheng — Open LLM DevOps Platform for Enterprise AI

Introduction

Bisheng provides a full-stack platform for building, deploying, and managing LLM-powered enterprise applications. It combines a visual flow editor with built-in RAG, agent orchestration, model management, and evaluation tools so teams can go from prototype to production within a single system.

What Bisheng Does

• Offers a drag-and-drop workflow editor for building GenAI applications visually
• Provides a complete RAG pipeline with document parsing, chunking, and vector retrieval
• Supports multi-agent collaboration with human-in-the-loop intervention
• Manages multiple LLM providers and model versions from a unified dashboard
• Includes built-in evaluation, dataset management, and supervised fine-tuning tools

Architecture Overview

Bisheng's backend is built with Python (FastAPI) and communicates with a React-based frontend. The workflow engine executes DAGs defined in the visual editor, routing data between LLM calls, retrieval nodes, tool calls, and conditional logic. Vector storage uses Milvus or Elasticsearch, and relational metadata lives in PostgreSQL. Model serving integrates with vLLM, Ollama, or remote APIs. The platform supports multi-tenancy with role-based access control for enterprise deployments.

Self-Hosting & Configuration

• Deploy via Docker Compose with the included configuration files
• Configure LLM endpoints (OpenAI-compatible APIs, local models via vLLM/Ollama)
• Set up vector database (Milvus or Elasticsearch) for RAG document storage
• Configure PostgreSQL and Redis for metadata and caching
• Customize authentication, user roles, and workspace isolation via environment variables

Key Features

• Visual flow editor with support for loops, parallel execution, and batch processing
• Document review workflows with human approval gates
• Multi-model management: switch providers and compare outputs side by side
• Enterprise features: RBAC, audit logging, multi-tenancy, and SSO integration
• Chinese and English UI with strong support for Chinese-language LLM applications

Comparison with Similar Tools

• Dify — LLMOps platform with visual builder; Bisheng adds deeper enterprise features like multi-tenancy and fine-tuning
• Langflow — visual LangChain builder; Bisheng provides a broader platform with model management and evaluation
• Flowise — low-code LLM app builder; Bisheng targets enterprise use cases with RBAC and audit trails
• RAGFlow — focused on document RAG; Bisheng combines RAG with agent orchestration and workflow automation
• Haystack — Python framework for search and RAG; Bisheng adds a visual UI and enterprise management layer

FAQ

Q: What LLM providers does Bisheng support? A: Any OpenAI-compatible API, plus direct integration with vLLM, Ollama, and cloud providers like Azure OpenAI and Baidu Wenxin.

Q: Can Bisheng handle document processing? A: Yes. It includes parsers for PDF, Word, Excel, and other formats, with automatic chunking and vector indexing for RAG.

Q: Is Bisheng suitable for production enterprise use? A: Yes. It provides RBAC, multi-tenancy, audit logging, and high-availability deployment configurations.

Q: What license is Bisheng released under? A: Bisheng is released under the Apache 2.0 license.

Sources

• https://github.com/dataelement/bisheng
• https://bisheng.dataelem.com/