Max Knowledge Brain for Java — an enterprise-grade AI brain for Java teams, with zero friction
An out-of-the-box, model-agnostic RAG + LLM workflow engine built on Java 21 + Spring Boot 3 (Virtual Threads)
Intelligent customer service · Enterprise knowledge bases · Data analysis · Research & education
🚀 Quick Start · 🌐 Live Demo (demo / demo@123456) · 📄 Whitepaper · 🧪 Regression Report · 🗒️ Changelog · 💖 Support Us
--- ## Why Choose MaxKB4j? Enterprises face four common challenges when adopting LLM applications, and MaxKB4j answers each one: | Pain Point | MaxKB4j Answer | |:---------------------------------------------------------------------------------------------------------------------|:----------------------------------------------------------------------------------------------------------------------------------------| | **Complex integration** — mainstream platforms rely on Python / TS ecosystems, making adoption costly for Java teams | **Pure Java native** — built on the Spring Boot 3 stack; existing Java engineers can extend it directly with zero cross-language cost | | **Serious hallucinations** — generic LLMs know nothing about internal enterprise data | **Production-grade RAG** — document parsing → chunking → vectorization → hybrid retrieval → Reranker re-ranking, with traceable answers | | **Concurrency bottlenecks** — traditional architectures cannot sustain high-concurrency Q&A | **Virtual Threads + reactive architecture** — thousands of concurrent requests per node with lower resource usage | | **Single-function Q&A** — no way to orchestrate complex business processes | **Visual workflow + Multi-Agent** — 30+ node types covering complex business scenarios | ## ✨ Key Capabilities | Capability | Description | |:-----------------------------|:--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------| | 🔍 Knowledge Base Q&A | Upload PDF / Word / TXT / Markdown files or crawl web pages; automatic chunking → vectorization → storage → RAG pipeline, significantly reducing hallucinations; in-document images are fused into vectors for image-text hybrid retrieval | | 🧠 Advanced RAG / AgenticRAG | Vector, full-text, and hybrid multi-route retrieval + Reranker re-ranking; agents dynamically decide retrieval paths with intent recognition and conditional branches, supporting multi-hop Q&A | | ⚙️ Visual Workflow | Low-code orchestration with 30+ node types: conditional branches, loops, variable aggregation, NL2SQL, forms, HTTP requests, MCP, etc.; multi-turn and long-term memory | | 🤝 Multi-Agent Collaboration | Multiple role-specific agents (data analyst, code reviewer, customer service agent…) work in parallel or sequence; tasks are decomposed, dispatched, and aggregated automatically via a shared memory bus | | ⏰ Triggers | Cron scheduled tasks + Webhook event callbacks for unattended automation (daily report generation, CRM-lead-triggered persona analysis, etc.) | | 🌐 Model-Agnostic | Private models via Ollama / Xorbits Inference / LocalAI; public models: Qwen, DeepSeek, Doubao, Hunyuan, GLM, Kimi, GPT, Claude, Gemini, and more | | 🧩 Seamless Integration | RESTful API, iframe / Web SDK embedding, OpenAI-compatible chat API, and stream_http MCP agent integration — connect within 5 minutes | | 🎙️ Multimodal | **Multimodal fusion embedding** — text and images in documents are fused into unified vectors, enabling text-to-image and image-to-text retrieval (Aliyun Bailian / Volcengine Ark vision embedding models); ASR speech recognition, TTS speech synthesis, OCR image recognition, Stable Diffusion image generation | | 🔒 Security & Permissions | Fine-grained permissions (application / knowledge base / tool / model) based on Sa-Token; audit logs; groovy-sandbox for safe script execution | | 🌱 Ecosystem Extensions | Dozens of pre-built agent templates (customer service assistant, data analyst, code mentor…); plugin marketplace: MySQL / PostgreSQL / MongoDB connectors, Feishu / DingTalk / WeCom integrations, web search tools | ## 📊 How MaxKB4j Compares | Capability | **MaxKB4j** | Dify | MaxKB | FastGPT | RAGFlow | |:-------------------------------------|:---------------------------:|:-----------:|:------:|:----------:|:-------:| | Backend stack | **Java 21 + Spring Boot 3** | Python + TS | Python | TypeScript | Python | | Zero language-switch for Java teams | ✅ Native | ⚠️ | ⚠️ | ⚠️ | ⚠️ | | Virtual-Thread high concurrency | ✅ | ⚠️ | ⚠️ | ⚠️ | ⚠️ | | Visual workflow + Multi-Agent | ✅ | ✅ | ⚠️ | ⚠️ | ⚠️ | | Triggers (Cron / Webhook) | ✅ | ✅ | ⚠️ | ⚠️ | ⚠️ | | Multimodal (ASR / TTS / OCR) | ✅ | ⚠️ | ✅ | ⚠️ | ✅ | | Text + image fusion embedding | ✅ | ⚠️ | ⚠️ | ⚠️ | ⚠️ | | MCP protocol / OpenAI-compatible API | ✅ | ✅ | ✅ | ✅ | ✅ | > Note: this table is a capability-orientation comparison for selection reference. Evaluate with your own scenarios via > the live demo below. ## 🚀 Quick Start ### Requirements - Java 21+ - PostgreSQL 12+ (with pgvector extension enabled) - MongoDB 6.0+ (full-text search and file storage) ### Option 1: Docker Compose (recommended) ```bash docker-compose up -d ``` Then open `http://localhost:8080/admin/login` (default `admin` / `tarzan@123456`). The database is initialized automatically on first startup. ### Option 2: Docker single container ```bash docker run --name maxkb4j -d --restart always -p 8080:8080 \ -e SPRING_DATASOURCE_URL=jdbc:postgresql://localhost:5432/MaxKB4j \ -e SPRING_DATASOURCE_USERNAME=postgres \ -e SPRING_DATASOURCE_PASSWORD=123456 \ -e SPRING_DATA_MONGODB_URI=mongodb://admin:123456@localhost:27017/MaxKB4j?authSource=admin \ registry.cn-hangzhou.aliyuncs.com/tarzanx/maxkb4j ``` ### Option 3: One-click install script Interactive installers in `deploy/` handle environment checks, image pulling / building, and docker-compose orchestration (Linux / macOS / Windows): ```bash # Linux / macOS chmod +x deploy/install.sh ./deploy/install.sh # Windows deploy\install.bat ``` ### Option 4: Build from source ```bash mvn clean package -DskipTests java -jar maxkb4j-start/target/maxkb4j-start.jar ``` Profiles: `maxkb4j-start/src/main/resources/application-{dev,prod,test}.yml`, switch with `--spring.profiles.active=dev`. ### Deploy to cloud platforms Supports one-click deployment to [Sealos](https://blog.csdn.net/weixin_40986713/article/details/156026021) (overseas servers, no proxy needed, auto-scaling). > 🎉 **Deployed successfully?** > If MaxKB4j saved your team development time, consider [buying the author a coffee ☕](#support--sponsorship) — a coffee > unlocks the author's direct WeChat line and the core community group, where deployment questions get answered fast. ## 🌐 Live Demo - Live demo: http://43.143.235.194:8080/ (account `demo` / password `demo@123456`, normal-user permissions) - Local default admin: `admin` / `tarzan@123456`
## 🛠 Tech Stack & Project Structure
| Category | Tech |
|:---------------|:--------------------------------------------------------|
| Backend | Java 21, Spring Boot 3, Virtual Threads, Sa-Token |
| AI framework | LangChain4j 1.x, Docling document parsing, Volcengine Ark SDK |
| Storage | PostgreSQL 15 + pgvector, MongoDB 6.0+, Caffeine cache |
| Frontend | Element Plus,Vue 3, Node.js v20.16.0 |
| Script sandbox | groovy-sandbox |
```
MaxKB4j/
├── maxkb4j-common / maxkb4j-core # Utilities, core abstractions, domain models
├── maxkb4j-service/ # Business implementation: application / chat / knowledge / model / oss / system / tool / trigger / workflow
├── maxkb4j-service-api/ # Public contracts: -api modules (DTOs / VOs)
├── maxkb4j-start/ # Spring Boot entry point, config, packaging
└── deploy/ # One-click install scripts (install.sh / install.bat)
```
> Dependency direction: `start` → `service` → `service-api` → `core` → `common`. Public contracts live in `-api`
> modules; implementations live in `service`.
## 📄 Documentation & Resources
- 📄 [Business Value Whitepaper](docs/MaxKB4j-商业价值白皮书.md) — value and selection analysis for enterprise decision
makers
- 🧪 [Regression Test Report](docs/MaxKB4j-回归测试报告.md) — regression verification results of core features
- 📐 [Coding Conventions](docs/编码约定.md)
- 🗒️ [Changelog](CHANGELOG.md) — latest release: v2.9.0 (2026-06-17)
## 🤝 Contributing
| Item | Details |
|:--------------------|:----------------------------------------------------------------------------------------------------------|
| Feedback & requests | Report bugs, suggestions, or feature requests via [Gitee Issues](https://gitee.com/taisan/MaxKB4j/issues) |
| Contribution flow | Fork → create a branch → push to the DEV branch → open a Pull Request |
| Coding standards | Follow Alibaba Java coding conventions; include unit tests and update docs |
## 💖 Support & Sponsorship
> **MaxKB4j has no commercial backing and no funding** — it is built line by line by an independent developer in his
> spare time.
> Every contribution goes directly to real costs:
>
> - 🖥 **Cloud servers for the live demo** — running 24/7, free for everyone to try
> - 🤖 **Model token consumption** — every regression test costs real money
> - 🧪 **Bug fixes & new feature development** — caffeine for late-night coding
>
> Usage is reported regularly in the community group, fully transparent.
>
> ⚠️ The author's time is limited: **direct WeChat access and the core community group are reserved for sponsors** (from
> ¥10) — this protects both the author's focus and the signal quality of the group.
### Sponsorship Tiers
| Tier | Amount | Benefits | Best for |
|:---------------------:|:------:|:-------------------------------------------------------------------------------------------------------------------------------------------|:---------------------------------------------------------|
| ☕ Coffee | ¥10 | Direct WeChat line to the author + core community group + priority update notifications | Anyone who appreciates the project |
| 📚 Learning Member | ¥99 | All Coffee benefits + free access to the [Knowledge Planet](https://wx.zsxq.com/group/28882525858841) + priority answers inside the planet | Developers who want to master RAG / workflow in practice |
| 🏢 Enterprise Partner | ¥799 | All Learning Member benefits + frontend source code (one-time) + deployment / post-sales support | Teams going to production |
| 👑 Strategic Partner | ¥1399 | All Enterprise Partner benefits + 6-month frontend source upgrades + your logo on the sponsor wall | Long-term partners growing together |
> 💡 **Every tier includes the author's direct WeChat line + the core community group** — unlocked from the ¥10 Coffee
> tier.
### How to Sponsor (3 steps)
| Step 1 · Add the author on WeChat | Step 2 · Pay via Alipay | Step 2 · Pay via WeChat |
|---|---|---|
 |
 |
 |
| Note "MaxKB4j" | Step 3 · Send the payment screenshot to the author (with your nickname) — benefits activate immediately | |
