# BISHENG

> Open-source LLM application devops platform for enterprises, with workflow orchestration, RAG, agents, and document parsing.

BISHENG is an open-source LLM application devops platform from Dataelement, aimed at enterprise scenarios. It combines a visual workflow builder, a general-purpose agent, knowledge-base RAG, a document parsing model, and enterprise administration features. The repository is licensed under Apache 2.0 and is self-hosted with Docker Compose.

## What It Is

BISHENG is a platform for building and operating LLM-based applications inside organizations. The GitHub description lists GenAI workflow, RAG, agents, unified model management, evaluation, SFT, dataset management, enterprise system management, and observability. Typical use cases named by the project include document review, fixed-layout report generation, customer service assistance, meeting minutes generation, resume screening, call record analysis, and unstructured data governance.

## Workflow and Agent Capabilities

BISHENG Workflow is described as a single orchestration framework that supports loops, parallelism, batch processing, and conditional logic. It also supports human-in-the-loop intervention, so users can give feedback during execution, including in multi-turn conversations. Lingsight is a general-purpose agent that uses the Agent Guidance Language (AGL) framework to embed domain experts' preferences and business logic. The website says users can create SOPs that guide the agent through complex tasks, with visual drag-and-drop orchestration and preset task templates.

## Document Parsing and Enterprise Controls

The project includes ETL4LM, a document parsing model that converts scanned PDFs, images with text, and complex layouts into clean text for LLMs. The README says it covers printed and handwritten text, rare characters, tables, layout analysis, and seals, and can be deployed privately for free. Enterprise features include security review, RBAC, user group management, grouped traffic control, SSO/LDAP, vulnerability scanning and patching, high-availability deployment, monitoring, and statistics.

## Setup Path

The README lists a minimum of 4 virtual CPU cores, 16 GB RAM, Docker 19.03.9+, and Docker Compose 1.25.1+. The default install also brings up Elasticsearch, Milvus, and Onlyoffice. After startup, the web UI is available on port 3001, and the first registered user becomes system admin. The latest listed release is v2.6.0-fix2.

## Features
- Visual workflow orchestration with loops, parallelism, batch processing, and conditional logic
- Human-in-the-loop workflow intervention
- Lingsight general-purpose agent with SOP guidance
- High-precision document parsing (ETL4LM) with OCR, table, layout, and seal recognition
- RAG knowledge bases
- Unified model management
- Evaluation, SFT, and dataset management
- RBAC, user group management, and SSO/LDAP
- Security review, vulnerability scanning, monitoring, and statistics
- High-availability deployment

## Integrations
Elasticsearch, Milvus, Onlyoffice, Docker

## Platforms
WEB, LINUX, API

## Pricing
Open Source

## Version
v2.6.0-fix2

## Links
- Website: http://www.bisheng.ai
- Documentation: https://dataelem.feishu.cn/wiki/ZxW6wZyAJicX4WkG0NqcWsbynde
- Repository: https://github.com/dataelement/bisheng
- EveryDev.ai: https://www.everydev.ai/tools/bisheng