Bisheng

also known as 毕昇, BISHENG

Type: low-code · Vendor: DataElem (数据项素) · Language: Python (visual) · License: Apache-2.0 · Status: active · Status in practice: emerging

Links: homepage docs repo

Provide an open, enterprise-grade LLMOps platform for building document-centric AI applications — GenAI workflows, RAG, Agents, evaluations and SFT — with high-precision document parsing and human-in-the-loop intervention baked into the workflow runtime.

Description. BISHENG (毕昇) is an open-source enterprise LLM application platform built by DataElem. The README states: 'BISHENG is an open LLM devops platform for next generation Enterprise AI applications.' Capabilities span 'GenAI workflow, RAG, Agent, Unified model management, Evaluation, SFT, Dataset Management, Enterprise-level System Management, Observability.' Its workflow engine is positioned as an 'independent and comprehensive application orchestration framework' that 'supports loops, parallelism, batch processing, conditional logic and free combination of all logic components,' and it ships with a high-precision document parsing stack covering print, handwriting, rare characters, tables, layout and seal detection. Human-in-the-loop is a documented first-class feature.

Agent loop shape. A workflow is a directed graph of typed nodes assembled in a visual editor and executed by Bisheng's orchestration runtime. Nodes cover model calls, retrieval over a knowledge base, Lingsight agent invocations (using the AGL — Agent Guidance Language — framework), code/condition logic, and human-in-the-loop interventions. The runtime supports loops, parallelism, batch processing and conditional logic. RAG sits inside the same workflow graph, fed by DataElem's high-precision document parsing models. Enterprise hardening (RBAC, SSO/LDAP, traffic control, security review) wraps every execution.

Primary use cases

enterprise document review and fixed-layout report generation
RAG over Chinese enterprise documents using high-precision parsing
multi-agent collaboration on document-centric tasks
policy / regulation update diff comparison
ticket-question-answering and unstructured data governance

flowchart TD TR[Trigger — user / API / batch] --> WF[Workflow graph] WF --> NODES{Typed nodes} NODES --> LLM[LLM / model node] NODES --> AGT[Lingsight agent — AGL] NODES --> KB[Knowledge Base — high-precision parsing] NODES --> CODE[Code / condition node<br/>loops + parallel + batch] NODES --> HITL{{Human-in-the-loop<br/>intervention}} AGT --> TOOLS[Components / tools] KB --> PARSE[Parse → chunk → embed → index → retrieve] HITL -.feedback.-> WF EVAL[Evaluation module] -.scores.-> WF SEC[(RBAC / SSO/LDAP / traffic control)] --> WF

Key concepts

Workflow → visual-workflow-graph (docs) — Visual orchestration graph supporting loops, parallelism, batch processing and conditional logic; the unit of execution.
Lingsight agent / AGL (docs) — Bisheng's general-purpose agent driven by the Agent Guidance Language framework that encodes domain expertise.
High-precision document parsing → agentic-rag (docs) — First-party models for print, handwriting, rare characters, tables, layout and seal detection feeding RAG.
Human-in-the-loop intervention → human-in-the-loop (docs) — Users can intervene and provide feedback during workflow execution including multi-turn conversations.
Evaluation (Eval) (docs) — Built-in evaluation module sitting alongside SFT and dataset management.
Enterprise security stack (docs) — RBAC, SSO/LDAP, traffic control, vulnerability scanning and security review on top of every workflow.

Bisheng

Neighbourhood

Alternatives & relatives

Listed as alternative by (3)

References

Provenance