JeecgBoot 3.9.5: An AI Low-Code Platform That Generates Whole Systems, Not Just CRUD Scaffolds

What JeecgBoot Actually Solves

JeecgBoot targets a specific pain: Java enterprise projects that repeat the same CRUD, form, and report work across every module. The README claims it solves 70% to 90% of repetitive Java work, depending on which section you read. The platform bundles a backend (Spring Boot 4, Shiro, MyBatis-Plus, Spring Cloud Alibaba), a Vue 3 frontend, and a separate uniapp framework for mobile. The core idea is that you generate the boring parts, then hand-merge custom logic. It is for teams that already live in the Java ecosystem and need an internal tool for MIS, OA, ERP, CRM, or AI knowledge bases. It is not for greenfield projects that want to avoid Java entirely, nor for teams that demand pixel-level control over every generated file.

The Skills Mechanism: Natural Language to System

The headline feature in version 3.9.5 is AI Skills. The README describes a workflow: 'AI 生成 → 在线配置 → 代码生成 → 手工合并' (AI generation, online configuration, code generation, manual merge). In practice, you install Claude Code plus JEECG Skills from a one-click installer at https://help.jeecg.com/java/ai/skills/skill-install. The Skills repository lives at https://github.com/jeecgboot/skills. Once installed, you can say 'generate a whole system' and get code, forms, flows, reports, and dashboards. There are two output modes: zero-code mode builds a running system without writing code, and code-generation mode uses jeecg-codegen to produce full code plus table creation SQL. This is a meaningful difference from older generators that only produced CRUD scaffolds. The AI layer is not just a chat wrapper; it maps natural language to the platform's existing generators, which is why the output is runnable rather than a suggestion.

Getting It Running: Commands and Configuration

The README points to external docs for setup, but it gives the key facts. Default credentials are admin/123456. For local development, you follow the 'IDEA启动前后端(单体模式)' guide. For a quick start, Docker one-click startup is available for both monolith and microservices modes. The microservices path uses Spring Cloud Alibaba with Nacos, Gateway, Sentinel, and Skywalking. The README claims you can switch between monolith and microservices with a simple mechanism, but it does not show the exact config keys. You will need to consult https://help.jeecg.com/java/docker/quick for the monolith Docker command and https://help.jeecg.com/java/docker/quickcloud for the microservices variant. The platform also supports domestic Chinese databases (Dameng, Kingbase, TiDB) and middleware (TongWeb, TongRDS), which matters for Xinchuang compliance projects.

Zero-Code vs Low-Code: Two Modes, Same Platform

The dual-mode design is the platform's main differentiator. Zero-code mode lets business users build forms and reports through OnlineCoding, with 23 control types and drag-and-drop configuration. Low-code mode is for developers who need to write Java or SQL. The README claims the low-code generator supports single tables, tree lists, one-to-many, and one-to-one models, with four template styles for single tables and three for one-to-many. The practical effect is that a business analyst can prototype a module while a developer later regenerates it with custom code. However, the README also admits that the low-code mode still requires manual merge for complex features. That merge step is where projects often get messy. The platform tries to reduce it with online enhancement editors that support Java, SQL, and JavaScript, but that still means you are maintaining generated code that you did not write from scratch.

AI Application Platform: A Dify Clone Inside a Low-Code Tool

Beyond code generation, JeecgBoot includes a full AI application platform that the README explicitly compares to Dify. It offers AI chat, knowledge base Q&A, process orchestration, model management, vector database integration, and MCP plugin configuration. The RAG pipeline is built in, so you can create a knowledge base and query it through the same interface. The platform supports ChatGPT, DeepSeek, Ollama, Zhipu, and Qwen, with DeepSeek as the default in the latest version. This is a serious feature set, not a toy demo. But it also means the platform is trying to be two products at once: a low-code development platform and an AIGC application platform. That breadth is a strength if you want everything in one stack, but it also increases the learning curve and the risk that neither half is as deep as a dedicated tool.

The Flexibility Trade-Off: Generated Code Still Needs Merging

The README is honest about a key limitation: the 'half-intelligent manual merge' mode is required for complex functionality. The platform's philosophy is OnlineCoding for simple features and code generation plus manual merge for complex ones. That means you cannot rely on AI to produce a production-ready system without human review. The generated code will follow JeecgBoot's conventions, which may not match your team's style or architectural standards. Also, the README mentions that the platform supports microservices and monolith switching, but it does not document the exact migration steps. Teams that need fine-grained control over data access or security may find the generated code too opinionated. The README claims granular permissions (button-level, data-level, field-level), but you will need to verify that the generated code respects those permissions in your custom merges.

Alternatives: Dify for AI Apps, JHipster for Code Generation

If your primary need is the AI application platform, Dify is a direct alternative. The README itself compares JeecgBoot's AI module to Dify, so the comparison is fair. Dify focuses purely on LLM apps, RAG, and workflow orchestration, without the low-code CRUD layer. JeecgBoot bundles both, but Dify has a more mature ecosystem for AI-specific features like prompt management and model routing. On the code generation side, JHipster is a well-known alternative that generates Spring Boot and Angular/React/Vue applications from a domain model. JHipster does not offer AI natural language generation or zero-code forms; it is a developer tool that requires you to define entities in a DSL. JeecgBoot's advantage is the combined offering, but its disadvantage is that you adopt a larger, more opinionated framework. If you only need CRUD generation, JHipster gives you more control over the output. If you only need AI apps, Dify is more focused.

Maintenance, Licensing, and Upgrade Path

JeecgBoot is licensed under Apache-2.0, which permits commercial use, modification, and distribution with attribution. The repository is not archived and had a release on 2026-08-27 (v3.9.5), so maintenance appears active. The README mentions a starter project (jeecg-boot-starter) that encapsulates microservices startup, distributed locks, RabbitMQ, distributed transactions, and ShardingSphere. That means upgrade cost is non-trivial: you are tied to the platform's internal conventions, and upgrading the core may require updating your custom merges. The README also suggests you can delete demo modules to create a slim version, but that requires following a separate guide. There is no mention of a migration tool for existing non-JeecgBoot projects. For teams that adopt it, the main cost is learning the OnlineCoding conventions and keeping custom code aligned with generated changes. The AI Skills workflow adds a dependency on Claude Code and the Skills repository, so you need to track updates there as well.

Editorial conclusion

Adopt JeecgBoot if you run Java shops building MIS, OA, ERP, or CRM systems and want AI-assisted generation plus a full AI app platform (chat, RAG, MCP) in one codebase. Skip it if you need strict control over generated code or if your team avoids vendor-specific OnlineCoding conventions. Before adopting, verify that the Spring Boot 4 and Vue 3 versions match your infrastructure, test the microservices switch with your Nacos setup, and confirm that the AI Skills workflow (install via https://help.jeecg.com/java/ai/skills/skill-install) works with your chosen model, since the default is DeepSeek.

Official sources

1. Official documentation
2. Official README
3. Project repository
4. Release notes