// the find
dataelement/bisheng
BISHENG is an open LLM devops platform for next generation Enterprise AI applications. Powerful and comprehensive features include: GenAI workflow, RAG, Agent, Unified model management, Evaluation, SFT, Dataset Management, Enterprise-level System Management, Observability and more.
Bisheng is a self-hosted LLM application platform built for enterprise use — workflow orchestration, RAG, agent building, document OCR/parsing, and model fine-tuning all under one roof. It's for teams that want an on-prem alternative to hand-wiring LangChain plus a vector DB plus a chatbot UI, especially ones with document-heavy workloads.
The OCR/document parsing stack (layout analysis, table recognition, seal detection) is trained on the vendor's own multi-year data set rather than being a thin wrapper around Tesseract or a generic VLM, and it works standalone. The workflow engine supports loops, parallelism, and mid-run human intervention in the same canvas — most chatflow builders like Flowise or Dify need separate modules for that. RBAC, SSO/LDAP, and group-based traffic control are first-class rather than bolted on after the fact, which shows in how much of the repo's own feature-tracking is dedicated to hardening multi-tenancy and permissions release over release.
Deployment is heavy: the default docker-compose brings up Elasticsearch, Milvus, and OnlyOffice alongside the app, with 18 vCPU / 48GB recommended — not something you spin up just to kick the tires. English documentation is thin; the real install and ops instructions sit behind a Feishu wiki link rather than in the repo, so anyone outside the Chinese-speaking user base hits a wall past the quick-start. Core abstractions like AGL (Agent Guidance Language) and the Lingsight agent are proprietary DSLs layered on top of LangChain rather than exposed LangChain primitives, so time invested in them doesn't transfer if you outgrow the platform.