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 – An Enterprise‑grade LLM Application Platform

What it is – Bisheng is an open‑source dev‑ops platform for building, deploying and managing LLM‑powered applications inside enterprises. It bundles a knowledge‑base, a multi‑step “task‑mode” agent, workflow orchestration, and a set of enterprise‑grade controls (RBAC, approvals, audit, SSO, etc.). The project is positioned as a self‑hosted alternative to commercial AI‑ops suites.

Core capabilities (as described in the README)

Feature	What you get	How it works
Task Mode – general‑purpose agent	Give a high‑level goal, attach files/knowledge bases, and the system produces a report, PPT, Excel or Word file.	The main agent creates a checklist, spawns parallel sub‑agents, and can pause for human input. Built‑in skill packs handle PPT/Excel/Word; custom skills can be added via local files or GitHub.
Knowledge Space	Team‑wide, hierarchical knowledge base with fine‑grained permissions.	Multi‑level folders, tags, list/card views, keyword+tag search. AI Q&A over a single document or the whole space, with citations that link back to source passages.
Channels	Automated ingestion of external news (WeChat, websites) into the knowledge base.	Topic‑based channels with sub‑channels; custom filter rules keep only relevant articles.
Bisheng Workflow	Visual flow editor that mixes chat‑turns and automated steps.	Supports loops, parallel branches, batch runs, and human‑in‑the‑loop pauses. No split between chatflow and workflow – they are a single orchestration engine.
Enterprise‑grade controls	User/department management, RBAC + ReBAC, approval flows, audit logs, dashboards, Open API, SSO/LDAP, HA deployment, content security, monitoring, etc.	Commercial edition adds dashboards and a desktop client (DSH Desktop) for model quota management.
High‑precision document parsing	OCR models for printed text, handwriting, rare characters, tables, layout, seals.	Models are trained on 5 years of data and can be deployed privately for free.

Quick start (from the README)

Prerequisites – 4 CPU cores, 16 GB RAM, Docker ≥ 19.03.9, Docker‑Compose ≥ 1.25.1 (recommended 18 cores / 48 GB). The Docker stack also pulls Elasticsearch, Milvus and OnlyOffice.
Installation – Clone the repo, cd bisheng/docker, then run:
```
docker compose -f docker-compose.yml -p bisheng up -d
```
Access – Open http://<host>:3001, register the first user (becomes admin).
Further docs – Self‑hosting guide is on the project’s Feishu wiki.

Ecosystem & acknowledgements

Bisheng re‑uses several well‑known open‑source projects:

LangChain – for LLM chain building.
Langflow – visual flow UI inspiration.
Unstructured – document parsing utilities.
LLaMA‑Factory – model fine‑tuning utilities.

Who might use this?

Enterprises that need an internal LLM portal with strict access control, audit trails and the ability to ingest proprietary documents.
Product teams building multi‑step AI assistants (e.g., report generation, data extraction) that require human‑in‑the‑loop checkpoints.
Developers looking for a self‑hosted alternative to SaaS AI‑ops platforms, with the ability to plug in custom skills or models.

Limitations (as implied by the README)

The platform is Docker‑centric and expects a fairly powerful host; not a lightweight library.
Some advanced features (dashboards, desktop client, certain enterprise integrations) are marked commercial edition – they may require a paid license.
The README points to a private Feishu wiki for detailed docs, so out‑of‑the‑box guidance may be limited for public users.

Bottom line – Bisheng is a genuine, full‑stack, open‑source LLM application platform aimed at enterprise deployments, offering task‑oriented agents, knowledge management, workflow orchestration, and robust admin controls.