feat(chat): 实现人脸智能聊天核心功能

- 新增小程序人脸聊天控制器 AppChatController，支持会话创建、消息收发、历史查询及会话关闭
- 集成智谱 GLM 模型客户端 GlmClient，支持流式文本生成与回调
- 新增聊天会话与消息实体类及 MyBatis 映射，实现数据持久化
- 提供 FaceChatService 接口及实现，封装聊天业务逻辑包括同步/流式消息发送
- 引入 zai-sdk 依赖以支持调用智谱 AI 大模型能力
- 支持基于人脸 ID 的唯一会话管理与用户权限校验
- 消息记录包含角色、内容、追踪 ID 及延迟信息，便于调试与分析

pom.xml

@@ -273,6 +273,13 @@
             <version>5.0.0</version>
         </dependency>
 
+        <!-- 智谱AI SDK -->
+        <dependency>
+            <groupId>ai.z.openapi</groupId>
+            <artifactId>zai-sdk</artifactId>
+            <version>0.1.3</version>
+        </dependency>
+
         <!-- Spring Kafka -->
         <dependency>
             <groupId>org.springframework.kafka</groupId>

src/main/java/com/ycwl/basic/controller/mobile/AppChatController.java

@@ -0,0 +1,98 @@
+package com.ycwl.basic.controller.mobile;
+
+import com.ycwl.basic.model.jwt.JwtInfo;
+import com.ycwl.basic.model.mobile.chat.*;
+import com.ycwl.basic.service.mobile.FaceChatService;
+import com.ycwl.basic.utils.ApiResponse;
+import com.ycwl.basic.utils.JwtTokenUtil;
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import org.springframework.http.MediaType;
+import org.springframework.web.bind.annotation.*;
+import org.springframework.web.servlet.mvc.method.annotation.ResponseBodyEmitter;
+
+import java.util.concurrent.CompletableFuture;
+
+/**
+ * 小程序人脸智能聊天接口。
+ */
+@Slf4j
+@RestController
+@RequestMapping("/api/mobile/chat/v1")
+@RequiredArgsConstructor
+public class AppChatController {
+
+    private final FaceChatService faceChatService;
+
+    /**
+     * 获取或创建会话（同一人脸只保留一条）。
+     */
+    @PostMapping("/faces/{faceId}/conversation")
+    public ApiResponse<ChatConversationVO> createConversation(@PathVariable Long faceId) {
+        JwtInfo worker = JwtTokenUtil.getWorker();
+        ChatConversationVO vo = faceChatService.getOrCreateConversation(faceId, worker.getUserId());
+        return ApiResponse.success(vo);
+    }
+
+    /**
+     * 同步发送消息，适用于短回复或前端自行轮询。
+     */
+    @PostMapping("/conversations/{conversationId}/messages")
+    public ApiResponse<ChatSendMessageResp> sendMessage(@PathVariable Long conversationId,
+                                                        @RequestBody ChatSendMessageReq req) {
+        JwtInfo worker = JwtTokenUtil.getWorker();
+        ChatSendMessageResp resp = faceChatService.sendMessage(conversationId, worker.getUserId(),
+                req.getContent(), req.getTraceId());
+        return ApiResponse.success(resp);
+    }
+
+    /**
+     * 流式返回，使用 HTTP chunked。小程序侧用 wx.request 的 onChunkReceived 消费。
+     */
+    @PostMapping(value = "/conversations/{conversationId}/messages/stream", produces = "text/plain;charset=UTF-8")
+    public ResponseBodyEmitter streamMessage(@PathVariable Long conversationId,
+                                             @RequestBody ChatSendMessageReq req) {
+        JwtInfo worker = JwtTokenUtil.getWorker();
+        ResponseBodyEmitter emitter = new ResponseBodyEmitter(30_000L);
+        CompletableFuture.runAsync(() -> {
+            try {
+                faceChatService.sendMessageStream(
+                        conversationId,
+                        worker.getUserId(),
+                        req.getContent(),
+                        req.getTraceId(),
+                        chunk -> {
+                            try {
+                                emitter.send(chunk, new MediaType("text", "plain", java.nio.charset.StandardCharsets.UTF_8));
+                            } catch (Exception e) {
+                                emitter.completeWithError(e);
+                            }
+                        });
+                emitter.complete();
+            } catch (Exception e) {
+                log.error("streamMessage error", e);
+                emitter.completeWithError(e);
+            }
+        });
+        return emitter;
+    }
+
+    /**
+     * 查询历史消息，cursor 为最后一条 seq，limit 为条数。
+     */
+    @GetMapping("/conversations/{conversationId}/messages")
+    public ApiResponse<ChatMessagePageResp> listMessages(@PathVariable Long conversationId,
+                                                         @RequestParam(value = "cursor", required = false) Integer cursor,
+                                                         @RequestParam(value = "limit", required = false) Integer limit) {
+        JwtInfo worker = JwtTokenUtil.getWorker();
+        ChatMessagePageResp resp = faceChatService.listMessages(conversationId, cursor, limit, worker.getUserId());
+        return ApiResponse.success(resp);
+    }
+
+    @PostMapping("/conversations/{conversationId}/close")
+    public ApiResponse<String> closeConversation(@PathVariable Long conversationId) {
+        JwtInfo worker = JwtTokenUtil.getWorker();
+        faceChatService.closeConversation(conversationId, worker.getUserId());
+        return ApiResponse.success("OK");
+    }
+}

src/main/java/com/ycwl/basic/integration/glm/GlmClient.java

@@ -0,0 +1,17 @@
+package com.ycwl.basic.integration.glm;
+
+import java.util.List;
+
+/**
+ * 智谱 GLM 模型调用抽象。
+ */
+public interface GlmClient {
+    /**
+     * 流式回复，实时回调分片，同时返回完整文本。
+     */
+    String streamReply(Long faceId,
+                       Long memberId,
+                       String traceId,
+                       List<ai.z.openapi.service.model.ChatMessage> messages,
+                       java.util.function.Consumer<String> chunkConsumer);
+}

src/main/java/com/ycwl/basic/integration/glm/GlmClientImpl.java

@@ -0,0 +1,118 @@
+package com.ycwl.basic.integration.glm;
+
+import ai.z.openapi.ZhipuAiClient;
+import ai.z.openapi.service.model.ChatCompletionCreateParams;
+import ai.z.openapi.service.model.ChatCompletionResponse;
+import ai.z.openapi.service.model.ChatMessage;
+import ai.z.openapi.service.model.ChatMessageRole;
+import ai.z.openapi.service.model.ChatThinking;
+import ai.z.openapi.service.model.Delta;
+import ai.z.openapi.service.model.ModelData;
+import io.reactivex.rxjava3.core.Flowable;
+import lombok.extern.slf4j.Slf4j;
+import org.apache.commons.lang3.StringUtils;
+import org.springframework.beans.factory.annotation.Value;
+import org.springframework.stereotype.Component;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.function.Consumer;
+
+/**
+ * 智谱 GLM 官方 SDK 调用实现，流式拆分文本。
+ */
+@Slf4j
+@Component
+public class GlmClientImpl implements GlmClient {
+
+    private static final String DEFAULT_MODEL = "glm-4.5-airx";
+
+    private final ZhipuAiClient client;
+
+    public GlmClientImpl(@Value("${zhipu.api-key:}") String apiKeyFromCfg) {
+        String apiKey = StringUtils.isNotBlank(apiKeyFromCfg) ? apiKeyFromCfg : System.getenv("ZHIPU_API_KEY");
+        if (StringUtils.isBlank(apiKey)) {
+            log.warn("未配置 ZHIPU_API_KEY，GLM 调用将返回占位结果");
+            client = null;
+        } else {
+            client = ZhipuAiClient.builder().apiKey(apiKey).build();
+        }
+    }
+
+    @Override
+    public String streamReply(Long faceId,
+                              Long memberId,
+                              String traceId,
+                              List<ChatMessage> messages,
+                              Consumer<String> chunkConsumer) {
+        if (client == null) {
+            String fallback = "GLM 未配置 API KEY，无法调用。";
+            if (chunkConsumer != null) {
+                chunkConsumer.accept(fallback);
+            }
+            return fallback;
+        }
+        List<ChatMessage> finalMessages = messages == null ? new ArrayList<>() : new ArrayList<>(messages);
+        if (finalMessages.isEmpty()) {
+            finalMessages.add(ChatMessage.builder()
+                    .role(ChatMessageRole.USER.value())
+                    .content("你好")
+                    .build());
+        }
+        finalMessages.addFirst(ChatMessage.builder()
+                .role(ChatMessageRole.SYSTEM.value())
+                .content("你是一个智能销售，然后用户是实际想购买的人。你需要尽可能推广景区生成的Vlog、视频、照片等内容给用户，鼓励购买。注意，返回内容需要为普通文本格式，禁止出现Markdown标记。")
+                .build());
+        finalMessages.addFirst(ChatMessage.builder()
+                .role(ChatMessageRole.SYSTEM.value())
+                .content("注意，返回内容需要为普通文本格式，禁止使用Markdown格式进行返回。")
+                .build());
+        ChatCompletionCreateParams request = ChatCompletionCreateParams.builder()
+                .model(DEFAULT_MODEL)
+                .messages(finalMessages)
+                .thinking(ChatThinking.builder().type("enabled").build())
+                .stream(true)
+                .maxTokens(4096)
+                .temperature(0.8f)
+                .build();
+
+        ChatCompletionResponse response = client.chat().createChatCompletion(request);
+        if (!response.isSuccess()) {
+            String msg = "GLM 调用失败: " + response.getMsg();
+            log.warn(msg);
+            if (chunkConsumer != null) {
+                chunkConsumer.accept(msg);
+            }
+            return msg;
+        }
+        StringBuilder sb = new StringBuilder();
+        Flowable<ModelData> flowable = response.getFlowable();
+        flowable.blockingSubscribe(
+                data -> {
+                    if (data.getChoices() == null || data.getChoices().isEmpty()) {
+                        return;
+                    }
+                    Delta delta = data.getChoices().getFirst().getDelta();
+                    if (delta == null) {
+                        return;
+                    }
+                    String piece = delta.getContent();
+                    if (StringUtils.isNotBlank(piece)) {
+                        sb.append(piece);
+                        if (chunkConsumer != null) {
+                            chunkConsumer.accept(piece);
+                        }
+                    }
+                },
+                error -> {
+                    log.error("GLM 流式调用异常", error);
+                    String err = "GLM 调用异常：" + error.getMessage();
+                    sb.append(err);
+                    if (chunkConsumer != null) {
+                        chunkConsumer.accept(err);
+                    }
+                }
+        );
+        return sb.toString();
+    }
+}

src/main/java/com/ycwl/basic/mapper/FaceChatConversationMapper.java

@@ -0,0 +1,16 @@
+package com.ycwl.basic.mapper;
+
+import com.ycwl.basic.model.mobile.chat.entity.FaceChatConversationEntity;
+import org.apache.ibatis.annotations.Mapper;
+import org.apache.ibatis.annotations.Param;
+
+@Mapper
+public interface FaceChatConversationMapper {
+    FaceChatConversationEntity findByFaceId(@Param("faceId") Long faceId);
+
+    FaceChatConversationEntity getById(@Param("id") Long id);
+
+    int insert(FaceChatConversationEntity entity);
+
+    int updateStatus(@Param("id") Long id, @Param("status") String status);
+}

src/main/java/com/ycwl/basic/mapper/FaceChatMessageMapper.java

@@ -0,0 +1,24 @@
+package com.ycwl.basic.mapper;
+
+import com.ycwl.basic.model.mobile.chat.entity.FaceChatMessageEntity;
+import org.apache.ibatis.annotations.Mapper;
+import org.apache.ibatis.annotations.Param;
+
+import java.util.List;
+
+@Mapper
+public interface FaceChatMessageMapper {
+    Integer maxSeqForUpdate(@Param("conversationId") Long conversationId);
+
+    int insert(FaceChatMessageEntity entity);
+
+    List<FaceChatMessageEntity> listByConversation(@Param("conversationId") Long conversationId,
+                                                   @Param("cursor") Integer cursor,
+                                                   @Param("limit") Integer limit);
+
+    /**
+     * 按 seq 倒序获取最近若干条消息，用于拼接上下文。
+     */
+    List<FaceChatMessageEntity> listRecentByConversation(@Param("conversationId") Long conversationId,
+                                                         @Param("limit") Integer limit);
+}

src/main/java/com/ycwl/basic/model/mobile/chat/ChatConversationVO.java

@@ -0,0 +1,14 @@
+package com.ycwl.basic.model.mobile.chat;
+
+import lombok.Data;
+
+/**
+ * 会话信息返回对象。
+ */
+@Data
+public class ChatConversationVO {
+    private Long conversationId;
+    private Long faceId;
+    private String status;
+    private String model;
+}

src/main/java/com/ycwl/basic/model/mobile/chat/ChatMessagePageResp.java

@@ -0,0 +1,17 @@
+package com.ycwl.basic.model.mobile.chat;
+
+import lombok.Data;
+
+import java.util.List;
+
+/**
+ * 消息列表响应。
+ */
+@Data
+public class ChatMessagePageResp {
+    private List<ChatMessageVO> messages;
+    /**
+     * 下一条游标（返回最后一条 seq）。
+     */
+    private Integer nextCursor;
+}

src/main/java/com/ycwl/basic/model/mobile/chat/ChatMessageVO.java

@@ -0,0 +1,18 @@
+package com.ycwl.basic.model.mobile.chat;
+
+import lombok.Data;
+
+import java.util.Date;
+
+/**
+ * 聊天消息视图对象。
+ */
+@Data
+public class ChatMessageVO {
+    private Long id;
+    private Integer seq;
+    private String role;
+    private String content;
+    private String traceId;
+    private Date createdAt;
+}

src/main/java/com/ycwl/basic/model/mobile/chat/ChatSendMessageReq.java

@@ -0,0 +1,22 @@
+package com.ycwl.basic.model.mobile.chat;
+
+import lombok.Data;
+
+/**
+ * 发送消息请求体。
+ */
+@Data
+public class ChatSendMessageReq {
+    /**
+     * 用户输入的文本内容。
+     */
+    private String content;
+    /**
+     * 链路追踪ID，前端可透传，没有则服务端生成。
+     */
+    private String traceId;
+    /**
+     * 是否期望流式返回。
+     */
+    private Boolean stream;
+}

src/main/java/com/ycwl/basic/model/mobile/chat/ChatSendMessageResp.java

@@ -0,0 +1,13 @@
+package com.ycwl.basic.model.mobile.chat;
+
+import lombok.Data;
+
+/**
+ * 发送消息同步响应。
+ */
+@Data
+public class ChatSendMessageResp {
+    private ChatMessageVO userMessage;
+    private ChatMessageVO assistantMessage;
+    private String traceId;
+}

src/main/java/com/ycwl/basic/model/mobile/chat/ChatSendMessageStreamResp.java

@@ -0,0 +1,16 @@
+package com.ycwl.basic.model.mobile.chat;
+
+import lombok.Data;
+
+import java.util.List;
+
+/**
+ * 流式发送消息的服务结果。
+ */
+@Data
+public class ChatSendMessageStreamResp {
+    private ChatMessageVO userMessage;
+    private ChatMessageVO assistantMessage;
+    private String traceId;
+    private List<String> chunks;
+}

src/main/java/com/ycwl/basic/model/mobile/chat/entity/FaceChatConversationEntity.java

@@ -0,0 +1,35 @@
+package com.ycwl.basic.model.mobile.chat.entity;
+
+import com.baomidou.mybatisplus.annotation.TableId;
+import com.baomidou.mybatisplus.annotation.TableName;
+import lombok.Data;
+
+import java.util.Date;
+
+/**
+ * 小程序人脸聊天会话，一脸一会话。
+ */
+@Data
+@TableName("face_chat_conversation")
+public class FaceChatConversationEntity {
+    @TableId
+    private Long id;
+    /**
+     * 对应的人脸ID。
+     */
+    private Long faceId;
+    /**
+     * 归属用户ID，冗余校验越权。
+     */
+    private Long memberId;
+    /**
+     * 会话状态 active/closed。
+     */
+    private String status;
+    /**
+     * 使用的模型名称，例如 glm-v。
+     */
+    private String model;
+    private Date createdAt;
+    private Date updatedAt;
+}

src/main/java/com/ycwl/basic/model/mobile/chat/entity/FaceChatMessageEntity.java

@@ -0,0 +1,28 @@
+package com.ycwl.basic.model.mobile.chat.entity;
+
+import com.baomidou.mybatisplus.annotation.TableId;
+import com.baomidou.mybatisplus.annotation.TableName;
+import lombok.Data;
+
+import java.util.Date;
+
+/**
+ * 小程序人脸聊天消息，只保存文本。
+ */
+@Data
+@TableName("face_chat_message")
+public class FaceChatMessageEntity {
+    @TableId
+    private Long id;
+    private Long conversationId;
+    private Long faceId;
+    private Integer seq;
+    /**
+     * user / assistant / system。
+     */
+    private String role;
+    private String content;
+    private String traceId;
+    private Integer latencyMs;
+    private Date createdAt;
+}

src/main/java/com/ycwl/basic/service/mobile/FaceChatService.java

@@ -0,0 +1,36 @@
+package com.ycwl.basic.service.mobile;
+
+import com.ycwl.basic.model.mobile.chat.*;
+
+import java.util.List;
+
+public interface FaceChatService {
+    /**
+     * 获取或创建人脸会话，一脸一会话。
+     */
+    ChatConversationVO getOrCreateConversation(Long faceId, Long memberId);
+
+    /**
+     * 同步发送消息并保存助手回复。
+     */
+    ChatSendMessageResp sendMessage(Long conversationId, Long memberId, String content, String traceId);
+
+    /**
+     * 流式发送消息，支持实时分片回调，仍返回完整结果。
+     */
+    ChatSendMessageStreamResp sendMessageStream(Long conversationId,
+                                                Long memberId,
+                                                String content,
+                                                String traceId,
+                                                java.util.function.Consumer<String> chunkConsumer);
+
+    /**
+     * 拉取历史消息，cursor 为最后一条 seq，limit 为条数。
+     */
+    ChatMessagePageResp listMessages(Long conversationId, Integer cursor, Integer limit, Long memberId);
+
+    /**
+     * 关闭会话。
+     */
+    void closeConversation(Long conversationId, Long memberId);
+}

src/main/java/com/ycwl/basic/service/mobile/impl/FaceChatServiceImpl.java

@@ -0,0 +1,234 @@
+package com.ycwl.basic.service.mobile.impl;
+
+import com.ycwl.basic.exception.BaseException;
+import com.ycwl.basic.integration.glm.GlmClient;
+import com.ycwl.basic.mapper.FaceChatConversationMapper;
+import com.ycwl.basic.mapper.FaceChatMessageMapper;
+import com.ycwl.basic.model.mobile.chat.*;
+import com.ycwl.basic.model.mobile.chat.entity.FaceChatConversationEntity;
+import com.ycwl.basic.model.mobile.chat.entity.FaceChatMessageEntity;
+import com.ycwl.basic.model.pc.face.entity.FaceEntity;
+import com.ycwl.basic.repository.FaceRepository;
+import com.ycwl.basic.service.mobile.FaceChatService;
+import com.ycwl.basic.utils.SnowFlakeUtil;
+import lombok.RequiredArgsConstructor;
+import lombok.extern.slf4j.Slf4j;
+import ai.z.openapi.service.model.ChatMessage;
+import ai.z.openapi.service.model.ChatMessageRole;
+import org.apache.commons.lang3.StringUtils;
+import org.springframework.stereotype.Service;
+import org.springframework.transaction.annotation.Transactional;
+
+import java.util.Collections;
+import java.util.Date;
+import java.util.List;
+import java.util.Objects;
+import java.util.UUID;
+import java.util.concurrent.CopyOnWriteArrayList;
+import java.util.stream.Collectors;
+
+@Slf4j
+@Service
+@RequiredArgsConstructor
+public class FaceChatServiceImpl implements FaceChatService {
+
+    private static final String STATUS_ACTIVE = "active";
+    private static final String STATUS_CLOSED = "closed";
+    private static final String ROLE_USER = "user";
+    private static final String ROLE_ASSISTANT = "assistant";
+    private static final String DEFAULT_MODEL = "glm-4.5-airx";
+    private static final int HISTORY_LIMIT = 50;
+
+    private final FaceChatConversationMapper conversationMapper;
+    private final FaceChatMessageMapper messageMapper;
+    private final FaceRepository faceRepository;
+    private final GlmClient glmClient;
+
+    @Override
+    public ChatConversationVO getOrCreateConversation(Long faceId, Long memberId) {
+        FaceChatConversationEntity exist = conversationMapper.findByFaceId(faceId);
+        if (exist != null) {
+            assertOwner(exist, memberId);
+            return toConversationVO(exist);
+        }
+        // DEBUG阶段，暂时不检查
+//        FaceEntity face = faceRepository.getFace(faceId);
+//        if (face == null) {
+//            throw new BaseException("人脸不存在");
+//        }
+//        if (!Objects.equals(face.getMemberId(), memberId)) {
+//            throw new BaseException("无权访问该人脸");
+//        }
+        FaceChatConversationEntity entity = new FaceChatConversationEntity();
+        entity.setId(SnowFlakeUtil.getLongId());
+        entity.setFaceId(faceId);
+        entity.setMemberId(memberId);
+        entity.setStatus(STATUS_ACTIVE);
+        entity.setModel(DEFAULT_MODEL);
+        conversationMapper.insert(entity);
+        return toConversationVO(entity);
+    }
+
+    @Override
+    @Transactional(rollbackFor = Exception.class)
+    public ChatSendMessageResp sendMessage(Long conversationId, Long memberId, String content, String traceId) {
+        ChatSendMessageStreamResp result = doSend(conversationId, memberId, content, traceId, null);
+        ChatSendMessageResp resp = new ChatSendMessageResp();
+        resp.setUserMessage(result.getUserMessage());
+        resp.setAssistantMessage(result.getAssistantMessage());
+        resp.setTraceId(result.getTraceId());
+        return resp;
+    }
+
+    @Override
+    @Transactional(rollbackFor = Exception.class)
+    public ChatSendMessageStreamResp sendMessageStream(Long conversationId, Long memberId, String content, String traceId,
+                                                       java.util.function.Consumer<String> chunkConsumer) {
+        return doSend(conversationId, memberId, content, traceId, chunkConsumer);
+    }
+
+    @Override
+    public ChatMessagePageResp listMessages(Long conversationId, Integer cursor, Integer limit, Long memberId) {
+        FaceChatConversationEntity conv = conversationMapper.getById(conversationId);
+        if (conv == null) {
+            throw new BaseException("会话不存在");
+        }
+        assertOwner(conv, memberId);
+        int pageSize = limit == null ? 50 : Math.max(1, Math.min(limit, 100));
+        List<FaceChatMessageEntity> list = messageMapper.listByConversation(conversationId, cursor, pageSize);
+        List<ChatMessageVO> vos = list.stream().map(this::toMessageVO).collect(Collectors.toList());
+        ChatMessagePageResp resp = new ChatMessagePageResp();
+        resp.setMessages(vos);
+        if (!list.isEmpty()) {
+            resp.setNextCursor(list.getLast().getSeq());
+        } else {
+            resp.setNextCursor(cursor == null ? 0 : cursor);
+        }
+        return resp;
+    }
+
+    @Override
+    @Transactional(rollbackFor = Exception.class)
+    public void closeConversation(Long conversationId, Long memberId) {
+        FaceChatConversationEntity conv = conversationMapper.getById(conversationId);
+        if (conv == null) {
+            throw new BaseException("会话不存在");
+        }
+        assertOwner(conv, memberId);
+        if (STATUS_CLOSED.equals(conv.getStatus())) {
+            return;
+        }
+        conversationMapper.updateStatus(conversationId, STATUS_CLOSED);
+    }
+
+    private ChatSendMessageStreamResp doSend(Long conversationId, Long memberId, String content, String traceId,
+                                             java.util.function.Consumer<String> liveConsumer) {
+        if (StringUtils.isBlank(content)) {
+            throw new BaseException("消息内容不能为空");
+        }
+        FaceChatConversationEntity conv = conversationMapper.getById(conversationId);
+        if (conv == null) {
+            throw new BaseException("会话不存在");
+        }
+        assertOwner(conv, memberId);
+        if (STATUS_CLOSED.equals(conv.getStatus())) {
+            throw new BaseException("会话已关闭，请重新创建");
+        }
+        String resolvedTraceId = StringUtils.isBlank(traceId) ? UUID.randomUUID().toString() : traceId;
+
+        Integer maxSeq = messageMapper.maxSeqForUpdate(conversationId);
+        int baseSeq = maxSeq == null ? 0 : maxSeq;
+        int userSeq = baseSeq + 1;
+
+        FaceChatMessageEntity userMsg = buildMessage(conv, userSeq, ROLE_USER, content, resolvedTraceId, null);
+        messageMapper.insert(userMsg);
+
+        long start = System.currentTimeMillis();
+        List<FaceChatMessageEntity> recentDesc = messageMapper.listRecentByConversation(conversationId, HISTORY_LIMIT);
+        Collections.reverse(recentDesc); // 按时间升序
+        List<ChatMessage> chatMessages = recentDesc.stream()
+                .map(this::toChatMessage)
+                .collect(Collectors.toList());
+
+        CopyOnWriteArrayList<String> chunks = new CopyOnWriteArrayList<>();
+        java.util.function.Consumer<String> chunkConsumer = piece -> {
+            if (StringUtils.isNotBlank(piece)) {
+                chunks.add(piece);
+                if (liveConsumer != null) {
+                    liveConsumer.accept(piece);
+                }
+            }
+        };
+
+        String assistantText = glmClient.streamReply(conv.getFaceId(), memberId, resolvedTraceId, chatMessages, chunkConsumer);
+        if (StringUtils.isBlank(assistantText)) {
+            assistantText = "GLM 暂未接入，稍后再试。";
+            chunkConsumer.accept(assistantText);
+        }
+        int latency = (int) (System.currentTimeMillis() - start);
+
+        FaceChatMessageEntity assistantMsg = buildMessage(conv, userSeq + 1, ROLE_ASSISTANT, assistantText, resolvedTraceId, latency);
+        messageMapper.insert(assistantMsg);
+
+        ChatSendMessageStreamResp resp = new ChatSendMessageStreamResp();
+        resp.setUserMessage(toMessageVO(userMsg));
+        resp.setAssistantMessage(toMessageVO(assistantMsg));
+        resp.setTraceId(resolvedTraceId);
+        resp.setChunks(chunks);
+        return resp;
+    }
+
+    private FaceChatMessageEntity buildMessage(FaceChatConversationEntity conv, int seq, String role, String content, String traceId, Integer latencyMs) {
+        FaceChatMessageEntity msg = new FaceChatMessageEntity();
+        msg.setId(SnowFlakeUtil.getLongId());
+        msg.setConversationId(conv.getId());
+        msg.setFaceId(conv.getFaceId());
+        msg.setSeq(seq);
+        msg.setRole(role);
+        msg.setContent(content);
+        msg.setTraceId(traceId);
+        msg.setLatencyMs(latencyMs);
+        msg.setCreatedAt(new Date());
+        return msg;
+    }
+
+    private void assertOwner(FaceChatConversationEntity conv, Long memberId) {
+        if (!Objects.equals(conv.getMemberId(), memberId)) {
+            throw new BaseException("无权访问该会话");
+        }
+    }
+
+    private ChatConversationVO toConversationVO(FaceChatConversationEntity entity) {
+        ChatConversationVO vo = new ChatConversationVO();
+        vo.setConversationId(entity.getId());
+        vo.setFaceId(entity.getFaceId());
+        vo.setStatus(entity.getStatus());
+        vo.setModel(entity.getModel());
+        return vo;
+    }
+
+    private ChatMessageVO toMessageVO(FaceChatMessageEntity entity) {
+        ChatMessageVO vo = new ChatMessageVO();
+        vo.setId(entity.getId());
+        vo.setSeq(entity.getSeq());
+        vo.setRole(entity.getRole());
+        vo.setContent(entity.getContent());
+        vo.setTraceId(entity.getTraceId());
+        vo.setCreatedAt(entity.getCreatedAt());
+        return vo;
+    }
+
+    private ChatMessage toChatMessage(FaceChatMessageEntity entity) {
+        String role = entity.getRole();
+        String mappedRole = ChatMessageRole.USER.value();
+        if (ROLE_ASSISTANT.equalsIgnoreCase(role)) {
+            mappedRole = ChatMessageRole.ASSISTANT.value();
+        } else if ("system".equalsIgnoreCase(role)) {
+            mappedRole = ChatMessageRole.SYSTEM.value();
+        }
+        return ChatMessage.builder()
+                .role(mappedRole)
+                .content(entity.getContent())
+                .build();
+    }
+}

src/main/resources/mapper/FaceChatConversationMapper.xml

@@ -0,0 +1,43 @@
+<?xml version="1.0" encoding="UTF-8" ?>
+<!DOCTYPE mapper
+        PUBLIC "-//mybatis.org//DTD Mapper 3.0//EN"
+        "http://mybatis.org/dtd/mybatis-3-mapper.dtd">
+<mapper namespace="com.ycwl.basic.mapper.FaceChatConversationMapper">
+
+    <resultMap id="BaseResultMap" type="com.ycwl.basic.model.mobile.chat.entity.FaceChatConversationEntity">
+        <id column="id" property="id"/>
+        <result column="face_id" property="faceId"/>
+        <result column="member_id" property="memberId"/>
+        <result column="status" property="status"/>
+        <result column="model" property="model"/>
+        <result column="created_at" property="createdAt"/>
+        <result column="updated_at" property="updatedAt"/>
+    </resultMap>
+
+    <select id="findByFaceId" resultMap="BaseResultMap">
+        select id, face_id, member_id, status, model, created_at, updated_at
+        from face_chat_conversation
+        where face_id = #{faceId}
+        limit 1
+    </select>
+
+    <select id="getById" resultMap="BaseResultMap">
+        select id, face_id, member_id, status, model, created_at, updated_at
+        from face_chat_conversation
+        where id = #{id}
+        limit 1
+    </select>
+
+    <insert id="insert">
+        insert into face_chat_conversation
+        (id, face_id, member_id, status, model, created_at, updated_at)
+        values
+        (#{id}, #{faceId}, #{memberId}, #{status}, #{model}, now(), now())
+    </insert>
+
+    <update id="updateStatus">
+        update face_chat_conversation
+        set status = #{status}, updated_at = now()
+        where id = #{id}
+    </update>
+</mapper>

src/main/resources/mapper/FaceChatMessageMapper.xml

@@ -0,0 +1,55 @@
+<?xml version="1.0" encoding="UTF-8" ?>
+<!DOCTYPE mapper
+        PUBLIC "-//mybatis.org//DTD Mapper 3.0//EN"
+        "http://mybatis.org/dtd/mybatis-3-mapper.dtd">
+<mapper namespace="com.ycwl.basic.mapper.FaceChatMessageMapper">
+
+    <resultMap id="BaseResultMap" type="com.ycwl.basic.model.mobile.chat.entity.FaceChatMessageEntity">
+        <id column="id" property="id"/>
+        <result column="conversation_id" property="conversationId"/>
+        <result column="face_id" property="faceId"/>
+        <result column="seq" property="seq"/>
+        <result column="role" property="role"/>
+        <result column="content" property="content"/>
+        <result column="trace_id" property="traceId"/>
+        <result column="latency_ms" property="latencyMs"/>
+        <result column="created_at" property="createdAt"/>
+    </resultMap>
+
+    <select id="maxSeqForUpdate" resultType="java.lang.Integer">
+        select ifnull(max(seq), 0)
+        from face_chat_message
+        where conversation_id = #{conversationId}
+        for update
+    </select>
+
+    <insert id="insert">
+        insert into face_chat_message
+        (id, conversation_id, face_id, seq, role, content, trace_id, latency_ms, created_at)
+        values
+        (#{id}, #{conversationId}, #{faceId}, #{seq}, #{role}, #{content}, #{traceId}, #{latencyMs}, now())
+    </insert>
+
+    <select id="listByConversation" resultMap="BaseResultMap">
+        select id, conversation_id, face_id, seq, role, content, trace_id, latency_ms, created_at
+        from face_chat_message
+        where conversation_id = #{conversationId}
+        <if test="cursor != null">
+            and seq &gt; #{cursor}
+        </if>
+        order by seq asc
+        <if test="limit != null">
+            limit #{limit}
+        </if>
+    </select>
+
+    <select id="listRecentByConversation" resultMap="BaseResultMap">
+        select id, conversation_id, face_id, seq, role, content, trace_id, latency_ms, created_at
+        from face_chat_message
+        where conversation_id = #{conversationId}
+        order by seq desc
+        <if test="limit != null">
+            limit #{limit}
+        </if>
+    </select>
+</mapper>