Yuanjing Wanwu Agent Platform
An all-in-one enterprise-grade multi-agent development platform covering RAG, workflows, MCP, and general agents for fast AI delivery.
- Source repo
- UnicomAI/wanwu
- Stars
- ★ 2.4k
- Last updated
- 3d ago
- License
- Apache-2.0
- Primary language
- Go
- FA score
- 30/100 · Major gaps
At a glance
- How it runs
- Works with
- Universal · cross-platformOpenAI API
- Cost
- Free software; you pay for model usage
- Setup effort
- Medium · a few setup steps
- You'll need
- Typical use
- Customer service teams: Build high-accuracy business Q&A and ticket handling with RAG and agents.
- Not a fit if
- Lightweight users wanting zero-config setup without middleware and keys
- Solo developers who only need a simple chatbot, not multi-tenant workflows
- Source review
- 30/100 · Major gaps 5 safety controls not found
What does this agent do, and when should you use it?
Yuanjing Wanwu Agent Platform is China Unicom's enterprise-grade, multi-tenant agent development platform released under the Apache-2.0 license. It integrates model management, MCP marketplace, high-precision RAG, visual workflows, general agents, and skill orchestration, and provides BaaS APIs for seamless integration with enterprise systems. The platform can be deployed via Docker or from source, and supports Xinchuang (domestic) environments such as Kunpeng CPUs, openEuler, Kylin, TiDB, and OceanBase. Its core focus is providing a full-stack FDE toolchain to lower the barrier to AI project delivery.
The platform provides: unified model hub for hundreds of models (e.g., GPT, Claude, Llama) with vLLM/TGI inference backends; 100+ pre-integrated MCP servers; RAG knowledge base supporting 12 file formats and URL crawling, with OCR/MinerU parsing and GraphRAG enhancement; a low-code visual workflow editor; functionality to build general agents (WanwuBot) and vertical skills with zero-code invocation; and REST APIs for BaaS, enabling integration with OA/CRM/ERP. Deployment options include Docker Compose, source code compilation, and the UniClaw desktop client for high-privilege operations.
- Customer service teams: Build high-accuracy business Q&A and ticket handling with RAG and agents.
- Knowledge managers: Create enterprise knowledge bases with semantic search and intelligent summarization.
- Process automation teams: Use workflow engine for AI-assisted contract review and expense approval.
- Xinchuang-compliant enterprises: Deploy on Kunpeng CPUs with domestic OS and TiDB/OceanBase databases.
- FDE engineers: Leverage the full-stack toolchain to deliver AI projects on-site with deep integrations.
How do you install or deploy this agent?
Docker installation: 1) cp .env.example .env; 2) Edit .env setting WANWU_ARCH (amd64/arm64) and WANWU_EXTERNAL_IP; 3) Set WANWU_BFF_JWT_SIGNING_KEY; 4) cp .env.ontology.example .env.ontology; 5) Create network docker network create wanwu-net; 6) Start with docker compose --env-file .env --env-file .env.ontology --env-file .env.image.amd64 up -d (use arm64 flavor for ARM); 7) Access http://localhost:8081, default user admin, password Wanwu123456. For development, stop services with Makefile and compile bff-service.
How do you use this agent?
After login, import models via Model Management (e.g., Unicom Yuanjing LLM, OpenAI-compatible), create a knowledge base and upload documents, design workflows using the visual editor, and build agent applications. Detailed manuals are under configs/microservice/bff-service/static/manual.
What are this agent's strengths and limitations?
- Apache-2.0 license permits free extension, secondary development, and commercial use.
- Unique features like GraphRAG and Ontology Agent outperform competitors on MultiHop-RAG benchmark (F1 +14% vs Dify, +3.5% vs LightRAG).
- Docker one-click deployment and Xinchuang support lower operational barriers.
- 100+ built-in MCP servers and Skill Plaza provide out-of-the-box capabilities.
- Default credentials and keys must be changed, posing security risk if ignored.
- Newer workflow module depends on a separate wanwu-workflow project, adding complexity.
- GUI agent requires downloading a desktop client (via Baidu netdisk), which may be inconvenient.
- Steep learning curve due to extensive features; relies on documentation.
How does this agent compare with similar options?
Compared with Dify.AI, Wanwu offers direct OCR import, GraphRAG, Ontology Agent, and General Agent+Skills framework, and has a more friendly license (Dify is commercially restricted). Versus FastGPT, Wanwu supports model import and more advanced capabilities. Against RagFlow, Wanwu provides GraphRAG and permissive licensing. Compared to Coze open-source version, Wanwu supports local deployment and multi-tenancy.
Key facts side by side with the most closely related agents.
| Agent | Source review | Form / cost | Stars | Updated | Language | Full support on |
|---|---|---|---|---|---|---|
| Yuanjing Wanwu Agent Platform This agent | 30 · Major gaps | Self-hosted serviceFree + model costs | ★ 2.4k | 3d ago | Go | OpenAI API |
| FIM One Enterprise AI Agent Platform | 48 · Major gaps | Self-hosted serviceFreemium | ★ 1.6k | 2d ago | Python | OpenAI API · Claude API |
| EpicStaff Agent Workflow Platform | 57 · Major gaps | Self-hosted service | ★ 359 | 10d ago | Python | — |
| Hector Agent Runtime | 55 · Major gaps | CLIFree + model costs | ★ 60 | 5mo ago | Go | OpenAI API · Claude API |
How does FollowAgents rate this agent?
Why each dimension lost points
Evidence shows: README mentions multi-tenant isolation, fine-grained permission control, RSA encrypted passwords, JWT signing key configuration, but lacks implementation details or security audit. Deductions: no user confirmation mechanism, data flow transparency, external effects control, rollback mechanism evidence.
Evidence shows: README provides deployment steps and configuration instructions, but lacks error handling or recovery information. Deductions: no failure messages and error handling documentation.
Evidence shows: README describes multiple application scenarios and deployment methods, but lacks clear capability boundaries or trigger conditions. Deductions: no trigger precision and explicit capability boundary description.
Evidence shows: README has clear structure, provides installation instructions, license, and version upgrade guide, but lacks known limitations and detailed changelog. Deductions: no known limitations and detailed changelog.
Evidence shows: README emphasizes rich features and ease of use, but lacks performance benchmarks or cost analysis. Deductions: no specific evidence for output usability and cost-benefit.
Evidence shows: README includes feature comparisons and claims, but lacks verifiable test results or third-party validation. Deductions: no cross-source verification and separation of facts and inferences.
- Not found in source: confirmation before actingTurn on (or add) a confirmation step before it acts, and try it in a sandbox or test environment before real data.
- Not found in source: data-flow disclosureWatch which external services it contacts (proxy or firewall logs) and keep sensitive data out until you know where it goes.
- Not found in source: dependency securityPin versions and run a dependency audit (npm audit, pip-audit) before installing; prefer running it in a container.
- Not found in source: disclosed external effectsEstablish which external systems it writes to, sends to or changes, and verify with test accounts or repos before production.
- Not found in source: rollback or recovery pathBack up first, or work on a git branch or snapshot, so its changes can be undone.
- No user confirmation mechanism, potentially executing high-risk operations automatically.
- Insufficient data flow transparency, users cannot track data usage.
- No rollback mechanism, operations may not be recoverable after failure.
- Dependency security not assessed, potential vulnerability risks.
Recent releases
Latest entries from the project's GitHub Releases. Notes are quoted from upstream as written, not rewritten or tested by FollowAgents.
-
Release notes excerpt
✨v0.6.5 有哪些新功能？✨ 在此版本中，我们添加了大量新功能、提升了用户使用时的交互体验，并修复了用户反馈的一些问题，希望能为您带来更好的使用感受。 🚀 新功能 - 模型管理： - 新增腾讯文本生成、文本向量化模型 - 知识库： - 支持配置解析模板 - 角色智能体： - 原本体智能体升级为决策智能体 📈增强功能 🛠️ 错误修复 - 本体智能体相关： - 远程命令执行 RCE，callback 组无鉴权 + 沙箱执行任意命令 - 任意文件读取，callback 组无鉴权 + http.Get 无白名单 - 未授权接口调用，Handler 不调 `verifyOAuth - 用户名枚举，组无鉴权 + 响应差异 - SSRF放行私有地址View on GitHub ↗ -
Release notes excerpt
✨v0.6.4 有哪些新功能？✨ 在此版本中，我们添加了大量新功能、提升了用户使用时的交互体验，并修复了用户反馈的一些问题，希望能为您带来更好的使用感受。 > 💡 配置提示： 应用观测中的「调用明细」支持记录完整的请求体和响应体（Request Body / Response Body），该功能通过 .env 文件环境变量 WANWU_BFF_STATISTIC_RECORD_BODY 控制（默认 0 关闭，设为 1 开启）。 > > - 开启时： 模型调用、应用调用、API Key 调用的明细记录中将保存整体请求与响应的 JSON 内容。由于会占用大量存储空间，建议定期清理历史明细数据。 > - 关闭时： 不落库完整请求/响应体，仅记录精简的问答摘要（Question / Answer）。流式响应始终不记录 Response Body。 🚀 新功能 - 本体智能体： - 数字员工支持发布，可配置发布范围 - 支持在应用广场查看并使用已发布的数字员工 - 知识问答： - 支持历史记录保存、查看、删除 - 消息中心： - 新增消息中心，可提示应用上架、模型上下架、知识库上架 - 知识库： - 升级分段内容搜索 - OpenAPI： …View on GitHub ↗ -
Release notes excerpt
✨v0.6.2 有哪些新功能？✨ 在此版本中，我们添加了大量新功能、提升了用户使用时的交互体验，并修复了用户反馈的一些问题，希望能为您带来更好的使用感受。 🚀 新功能 - 渠道管理： - 新增通用智能体-数据分析 - 管理员中心： - 新增产品后台。系统管理员、组织管理员可查看用户创建应用详情。包括模型、资源库、应用开发模块的内容。 - 组织管理支持选择上级组织 - 用户管理支持批量删除/移除 - 知识库： - 支持展示文件大小 - 下线url文件解析 📈增强功能 - 通用智能体： - 优化沙箱启动速度 - 组织管理： - 默认折叠到二级组织 🛠️ 错误修复 - 智能体： - 智能体Openapi调用报错 - 智能体上传文件显示大小有误 - 接口报错后，重新生成未携带上一轮的文件信息 - 知识库： - 知识库与平台接口交互安全升级，修复无认证外部接口带来的安全风险漏洞 - 本体智能体： - 修复万悟中文账号登录访问数字员工报错问题 - 渠道管理： - 修复微信回复乱码问题 - 修复图片显示问题View on GitHub ↗
FAQ
Which model providers are supported?
Is GPU required for deployment?
Does it work in domestic (Xinchuang) environments?
How is data security handled?
.env, RSA encryption is used for login, and multi-tenancy isolates data, but users must change default credentials.