不再沿着此方向调研

因为发现大模型定位是与用户交流，以及组织基于真实数据的分析和展示；erp注重真实数据的闭环；这两者的集合需要进行分层数据审查，以避免大模型引发数据混乱 https://mp.weixin.qq.com/s/91ZRC98zEYMu71sZJiB9-Q 目前新的方向开始转想map-scp （模型-代理-协议 + 安全-调用-协议）

MCP核心功能

flowchart TD
subgraph MoquiFramework
MCP[moqui-mcp组件]
end

    subgraph MCP内部结构
        direction TB
        MCP --> Tools[工具层]
        MCP --> Services[服务层]
        MCP --> Entities[实体层]
        MCP --> Scripts[脚本层]
    end

    Tools --> Ollama4J
    Tools --> ChromaClient
    Services --> AI_Service
    Services --> VectorDB_Service
    Entities --> KnowledgeBase
    Scripts --> InitData

Loading

消息处理流程

sequenceDiagram
    participant WeChat
    participant MCP
    participant Ollama
    participant MoquiERP
    
    WeChat->>MCP: 用户消息(XML)
    MCP->>Ollama: 意图识别请求
    Ollama-->>MCP: 意图分类结果
    alt ERP操作
        MCP->>MoquiERP: 调用ServiceFacade
        MoquiERP-->>MCP: 操作结果
    else AI查询
        MCP->>ChromaDB: 向量检索
        ChromaDB-->>MCP: 参考数据
        MCP->>Ollama: 生成回答
    end
    MCP->>WeChat: 最终响应
    

Loading

A moqui-wechat component

To install run (with moqui-framework):

$ ./gradlew getComponent -Pcomponent=moqui-wechat

To test run(use --tests ""*WechatServices" match the case)：

$ ./gradlew :runtime:component:moqui-wechat:test --tests "*WechatServices"

To build the moqui-wechat component and integrate a suitable AI tool, here are the steps and AI options you should consider:

To integrate Ollama with Llama 3.1 into your moqui-wechat component using the ollama4j plugin, here is a refined and clear description of the process:

This approach enables a private, secure, and scalable AI-powered WeChat interaction system within the Moqui ERP environment using Ollama with Llama 3.1 and the ollama4j plugin.

WeChat public account AI integration

pay attention to the model llama version's params difference. llama3.1 / llama3.2 need update the ollama jar for the new version of the model.

• call local ollama server with model llama3.2

curl -X POST http://localhost:11434/api/generate
-H "Content-Type: application/json"
-d '{ "model": "llama3.2", "prompt": "Hello, how are you?", "temperature": 0.7, "max_tokens": 100 }'

- [x] call the remote ollama server with model llama3.1

curl http://localhost:11434/api/generate -d '{ "model": "llama3.1", "prompt": "Why is the sky blue?" }' -H "Content-Type: application/json"

- [x] call remote ollama server from local

ssh -L 11434:localhost:11434 root@192.168.0.141
curl http://localhost:11434/api/generate -d '{ "model": "llama3.1", "prompt": "Why is the sky blue?" }' -H "Content-Type: application/json"

- [x] moqui-wechat call ollama by moqui-wechat

./gradlew :runtime:component:moqui-wechat:test --info

### Use RAG Flow to make my private domain data have the AI ability

![RAG  FLOW](rag_flow.jpg)

#### Step1:setup python env , chromadb and run as server mode
- require python3+ env: current use python 3.12.3
- python -m venv myenv
- active the env: source ~/myenv/bin/activate\n
- install the chromadb: pip install chromadb
- run chromadb as server mode: chroma run --path /db_path

//测试时允许重置数据库 ALLOW_RESET=TRUE chroma run --path /Users/demo/chromadb

#### Step2: run ollama and use EmbeddingFunction

// 配置 Ollama Embedding Function System.setProperty("OLLAMA_URL", "http://localhost:11434/api/embed"); EmbeddingFunction ef = new OllamaEmbeddingFunction(WithParam.baseAPI(System.getProperty("OLLAMA_URL")));

// 创建 HR 数据知识库 Collection Collection collection = client.createCollection("hr-knowledge", null, true, ef);

#### Step2: load the test data and run the gradle script
- Use HrDataIndexer to import test data into chromadb
- add dependence of the chromadb:chromadb-java-client.jar 
- use some test data

城市 姓名 手机号 上海 闵大 13661410000 上海 李二 15601720001 上海 戚三 15290860002 上海 李四 17721390003

### Step3: run the query script
- more info will get more correct

// 查询城市和姓名相关信息（例如：查询上海的李蜜的手机号） String query = "查询东莞常平镇的刘文博手机号"; Collection.QueryResponse qr = collection.query(Arrays.asList(query), 5, null, null, null);