Spring AI实现知识库搭建(实战篇)

框架：Ruoyi-Vue-Plus版本:5.3.1Spring-boot版本:3.4.4JDK:17spring-ai版本:1.0.0
需安装Ollama,且具备模型"nomic-embed-text"

资料已上传至技术群

ai_knowledge,ai_knowledge_document,ai_knowledge_segment,ai_model,ai_api_key

docker-compose --version

version: '3.5'services:  etcd:    container_name: milvus-etcd    image: quay.io/coreos/etcd:v3.5.18    environment:      - ETCD_AUTO_COMPACTION_MODE=revision      - ETCD_AUTO_COMPACTION_RETENTION=1000      - ETCD_QUOTA_BACKEND_BYTES=4294967296      - ETCD_SNAPSHOT_COUNT=50000    volumes:      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/etcd:/etcd    command: etcd -advertise-client-urls=http://etcd:2379 -listen-client-urls http://0.0.0.0:2379 --data-dir /etcd    healthcheck:      test: ["CMD", "etcdctl", "endpoint", "health"]      interval: 30s      timeout: 20s      retries: 3  minio:    container_name: milvus-minio    image: minio/minio:RELEASE.2025-04-22T22-12-26Z    environment:      MINIO_ACCESS_KEY: minioadmin      MINIO_SECRET_KEY: minioadmin    ports:      - "9000:9000"      - "9001:9001"    volumes:      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/minio:/minio_data    command: minio server /minio_data --console-address ":9001"    healthcheck:      test: ["CMD", "curl", "-f", "http://localhost:9000/minio/health/live"]      interval: 30s      timeout: 20s      retries: 3  standalone:    container_name: milvus-standalone    image: milvusdb/milvus:v2.5.13    command: ["milvus", "run", "standalone"]    security_opt:      - seccomp:unconfined    environment:      ETCD_ENDPOINTS: etcd:2379      MINIO_ADDRESS: minio:9000    volumes:      - ${DOCKER_VOLUME_DIRECTORY:-.}/volumes/milvus:/var/lib/milvus    healthcheck:      test: ["CMD", "curl", "-f", "http://localhost:9091/healthz"]      interval: 30s      start_period: 90s      timeout: 20s      retries: 3    ports:      - "19530:19530"      - "9091:9091"    depends_on:      - "etcd"      - "minio"  attu:    container_name: attu    image: zilliz/attu:v2.5.7    environment:      MILVUS_URL: milvus-standalone:19530    ports:      - "19500:3000"    depends_on:      - "standalone"networks:  default:    name: milvus

docker-compose up -d

docker images

docker ps

docker ps -a

docker restart etcd容器iddocker restart minio容器iddocker restart milvus容器iddocker restart attu容器id

# 浏览器访问miniolocalhost:9000

# 浏览器访问milvuslocalhost:19500

<tika.version>3.0.0</tika.version><spring-ai.version>1.0.0</spring-ai.version><commons-io.version>2.16.1</commons-io.version>

            <!-- Tika提取文件 必须 -->            <dependency>                <groupId>org.apache.tika</groupId>                <artifactId>tika-core</artifactId>                <version>${tika.version}</version>                <exclusions>                    <exclusion>                        <artifactId>commons-io</artifactId>                        <groupId>commons-io</groupId>                    </exclusion>                </exclusions>            </dependency>            <!-- 解析文档 必须 -->            <dependency>                <groupId>org.springframework.ai</groupId>                <artifactId>spring-ai-tika-document-reader</artifactId>                <version>${spring-ai.version}</version>            </dependency>            <!-- 解析文档 必须 -->            <dependency>                <groupId>commons-io</groupId>                <artifactId>commons-io</artifactId>                <version>${commons-io.version}</version>            </dependency>

spring:  application:    name: RuoYi-Vue-Plus  ai:    vectorstore:      milvus:        initialize-schema: true        database-name: default        collection-name: test        client:          host: milvus服务ip地址          port: 19530          username: root          password: milvus

# 字段1doc_id:文档id,也就是表ai_knowledge_segment中对应的字段vector_id,方便验证、查询

# 字段2embedding:向量维度,利用向量模型nomic-embed-text解析成向量,需要指定维度768,如果是1024或其它会报错！

# 字段3content:文档内容,利用tika解析文档

# 字段4metadata:元数据,需要存储业务数据,如知识库id、文档id

    /**     * 上传文档列表     *     * @param bo     * @return     */    @PostMapping("/createKnowledgeDocumentList")    public R<List<Long>> createKnowledgeDocumentList(@RequestBody AiKnowledgeDocumentListBo bo) {        return R.ok(aiKnowledgeDocumentService.createKnowledgeDocumentList(bo));    }

    /**     * 上传文档列表     *     * @param bo     * @return     */    List<Long> createKnowledgeDocumentList(AiKnowledgeDocumentListBo bo);

    /**     * 上传文档列表     *     * @param bo     * @return     */    @Override    @Transactional(rollbackFor = Exception.class)    public List<Long> createKnowledgeDocumentList(AiKnowledgeDocumentListBo bo) {        // 校验        aiKnowledgeService.validateKnowledgeExists(bo.getKnowledgeId());
        if (ObjectUtil.isEmpty(bo.getList())) {            throw new ServiceException("至少上传一个文件");        }
        // 批量读取文档内容        List<AiKnowledgeDocument> aiKnowledgeDocuments = prepareDocuments(bo);
        baseMapper.insertBatch(aiKnowledgeDocuments);        // 切片        processDocumentSegment(aiKnowledgeDocuments);
        return extractDocumentIds(aiKnowledgeDocuments);    }

VectorStore getOrCreateVectorStore(Class<? extends VectorStore> type, EmbeddingModel embeddingModel, Map<String, Class<?>> metadataFields);

    @Override    public VectorStore getOrCreateVectorStore(Long embedModelId, Map<String, Class<?>> metadataFields) {        // 获取模型信息        AiModelVo aiModelVo = validateModel(embedModelId);        AiApiKeyVo aiApiKeyVo = aiApiKeyService.validateApiKey(aiModelVo.getKeyId());        AiPlatformEnum aiPlatformEnum = AiPlatformEnum.validatePlatform(aiApiKeyVo.getPlatform());
        // 创建或获取嵌入模型        EmbeddingModel embeddingModel = modelFactoryService.getOrCreateEmbeddingModel(            aiPlatformEnum, aiApiKeyVo.getApiKey(), aiApiKeyVo.getUrl(), aiModelVo.getModel()        );
        return modelFactoryService.getOrCreateVectorStore(MilvusVectorStore.class, embeddingModel, metadataFields);    }