BISHENG is an open-source LLM DevOps platform, licensed under Apache-2.0, for building and operating next-generation enterprise AI applications, and it is aimed at organisations that need GenAI workflows, RAG, agents, and model management to run inside their own infrastructure.

What it is

BISHENG is an open LLM application devops platform, written primarily in Python, with an explicit focus on enterprise scenarios. It gathers GenAI workflow orchestration, RAG, agent construction, unified model management, evaluation, SFT, dataset management, enterprise-level system management, and observability into a single platform. The project is named after Bi Sheng, the inventor of movable type printing, and the README states the ambition that it should support widespread implementation of intelligent applications. The repository carries 11,973 stars, 1,968 forks, and 138 open issues, with the last push recorded on 2026-09-16. A hosted homepage is available at bisheng.ai.

The concrete problem it addresses is fragmentation in application orchestration. The README states that similar products rely on bot invocation, or on separate chatflow and workflow modules, to handle different classes of task, and that such products can only execute a workflow from start to finish without intervention. BISHENG instead presents one independent and comprehensive orchestration framework in which varied tasks execute together, and it adds human in the loop so users can intervene and give feedback mid-execution, including during multi-turn conversations. It also replaces dependence on third-party document parsing services by shipping its own high-precision parsing models, including printed text, handwritten text, rare character recognition, table recognition, layout analysis, and seal models, which the README states can be deployed privately for free.

Key capabilities

• Lingsight, a general-purpose agent built through the AGL (Agent Guidance Language) framework, which embeds domain expert preferences, experience, and business logic so the agent applies expert-level judgement to tasks.
• BISHENG Workflow, an independent orchestration framework that runs different task types inside one framework rather than splitting them across bot invocation or separate chatflow and workflow modules.
• Human in the loop, allowing users to intervene and provide feedback during workflow execution, including within multi-turn conversations.
• Loop, parallelism, batch processing, and conditional logic, with free combination of all logic components, expressed visually as a flowchart where drawing a loop forms a loop and aligning elements creates parallelism.
• Enterprise governance features: security review, RBAC, user group management, traffic control by group, SSO/LDAP, vulnerability scanning and patching, high availability deployment, monitoring, and statistics.
• High-precision document parsing models trained on high-quality data accumulated over the past five years, covering printed text, handwritten text, rare characters, tables, layout analysis, and seals.

Who uses it and how

• The README states BISHENG has been used by a large number of industry-leading organisations and Fortune 500 companies, indicating deployment at enterprise scale under internal governance rules.
• Back-office and compliance teams run document review, fixed-layout report generation, policy update comparison, and meeting minutes generation against parsed enterprise documents.
• Customer service functions use it for support ticket assistance, customer service assistance, and call record analysis.
• Recruiting teams apply it to resume screening, while data teams use it for unstructured data governance, knowledge mining, and data analysis, often alongside multi-agent collaboration.

Getting started

Install by cloning https://github.com/dataelement/bisheng.git, entering cd bisheng/docker, and running the Docker Compose stack. Prerequisites are Docker 19.03.9+, Docker Compose 1.25.1+, CPU of at least 4 virtual cores, and RAM of at least 16 GB, with 18 virtual cores and 48 GB recommended; Elasticsearch, Milvus, and Onlyoffice are also installed by default as third-party components.

How it compares

No list of paid products that BISHENG replaces was provided, and the README names no comparable tool by name, referring only to unnamed "similar products" that depend on bot invocation or separate chatflow and workflow modules. On the facts available, it stands alone in this registry rather than sitting in a named field of substitutes.

When to use it — and when not to

A self-hoster must operate the full Docker Compose stack plus Elasticsearch, Milvus, and Onlyoffice, and must meet a 16 GB RAM floor with 48 GB recommended, so it suits teams with container infrastructure and operational capacity rather than those wanting a managed service. Teams without that capacity, or wanting a hosted SaaS option, should not pick it. The install section of the provided README is truncated mid-command, so verifying the current deployment steps against the repository is advisable before committing.

git clone https://github.com/dataelement/bisheng.git
# Enter the installation directory
cd bisheng/docker

# If the system does not have the git command, you can download the BISHENG code as a zip file.
wget https://github.com/dataelement/bisheng/archive/refs/heads/main.zip
# Unzip and enter the installation directory
unzip main.zip && cd bisheng-main/docker

docker compose -f docker-compose.yml -p bisheng up -d