feat(aihr): expose model usage telemetry

backend/ruoyi-modules/ruoyi-aihr/src/main/java/org/dromara/aihr/domain/AihrModelDto.java

@@ -70,6 +70,14 @@ public final class AihrModelDto {
     public record ChatRequest(String prompt, String model, String systemPrompt) {
     }
 
+    public record ModelUsage(
+        Integer promptTokens,
+        Integer completionTokens,
+        Integer totalTokens,
+        long latencyMs
+    ) {
+    }
+
     public record ChatResponse(
         boolean configured,
         String providerCode,
@@ -77,7 +85,8 @@ public final class AihrModelDto {
         String answer,
         String mode,
         String error,
-        List<String> hints
+        List<String> hints,
+        ModelUsage usage
     ) {
     }
 }

backend/ruoyi-modules/ruoyi-aihr/src/main/java/org/dromara/aihr/service/AihrModelSeedService.java

@@ -14,6 +14,7 @@ import org.dromara.aihr.domain.AihrModelDto.ConfigResponse;
 import org.dromara.aihr.domain.AihrModelDto.ProviderRequest;
 import org.dromara.aihr.domain.AihrModelDto.ProviderResponse;
 import org.dromara.aihr.domain.AihrModelDto.ProviderStatusRequest;
+import org.dromara.aihr.domain.AihrModelDto.ModelUsage;
 import org.dromara.common.satoken.utils.LoginHelper;
 import org.springframework.beans.factory.annotation.Value;
 import org.springframework.dao.DataAccessException;
@@ -200,7 +201,8 @@ public class AihrModelSeedService {
                 "AI 调用已被试点成本闸门关闭，当前返回文本兜底：先安抚业主情绪，确认事实与责任人，再承诺首次反馈时间。",
                 "cost-guard-fallback",
                 null,
-                List.of("需要恢复真实调用时设置 AIHR_AI_RUNTIME_ENABLED=true 且 AIHR_AI_CHAT_ENABLED=true。")
+                List.of("需要恢复真实调用时设置 AIHR_AI_RUNTIME_ENABLED=true 且 AIHR_AI_CHAT_ENABLED=true。"),
+                null
             );
         }
 
@@ -215,13 +217,14 @@ public class AihrModelSeedService {
                 List.of(
                     "在 aihr_model_provider / aihr_model_config 配置 api_host、api_key、model_name 后可切换到真实 OpenAI-compatible 调用。",
                     "接口不会返回 api_key。"
-                )
+                ),
+                null
             );
         }
 
         try {
-            String content = callOpenAiCompatible(runtime, modelName, prompt, request == null ? null : request.systemPrompt(), 0.2);
-            return new ChatResponse(true, runtime.providerCode(), modelName, content, "openai-compatible", null, List.of());
+            ModelCallResult call = callOpenAiCompatible(runtime, modelName, prompt, request == null ? null : request.systemPrompt(), 0.2);
+            return new ChatResponse(true, runtime.providerCode(), modelName, call.content(), "openai-compatible", null, List.of(), call.usage());
         } catch (Exception e) {
             return new ChatResponse(
                 true,
@@ -230,7 +233,8 @@ public class AihrModelSeedService {
                 "真实模型调用失败，已回退内置默认兜底：先确认事实、同步时限、再生成工单闭环。",
                 "openai-compatible-failed",
                 "模型调用失败，已回退内置兜底；请检查模型供应商配置。",
-                List.of("检查 aihr_model_provider.api_host/api_key 和 aihr_model_config.model_name 是否与供应商一致。")
+                List.of("检查 aihr_model_provider.api_host/api_key 和 aihr_model_config.model_name 是否与供应商一致。"),
+                null
             );
         }
     }
@@ -284,7 +288,7 @@ public class AihrModelSeedService {
             return Optional.empty();
         }
         try {
-            return Optional.of(callOpenAiCompatible(runtime, runtime.modelName(), userPrompt, systemPrompt, temperature));
+            return Optional.of(callOpenAiCompatible(runtime, runtime.modelName(), userPrompt, systemPrompt, temperature).content());
         } catch (Exception e) {
             log.warn("aihr llm tryChat failed, caller falls back to seed（处理错误已隐藏）");
             return Optional.empty();
@@ -376,7 +380,8 @@ public class AihrModelSeedService {
         }
     }
 
-    private String callOpenAiCompatible(RuntimeConfig runtime, String modelName, String prompt, String systemPrompt, double temperature) throws Exception {
+    private ModelCallResult callOpenAiCompatible(RuntimeConfig runtime, String modelName, String prompt, String systemPrompt, double temperature) throws Exception {
+        long startedAt = System.nanoTime();
         ObjectNode body = objectMapper.createObjectNode();
         body.put("model", modelName);
         body.put("temperature", temperature);
@@ -422,7 +427,21 @@ public class AihrModelSeedService {
         if (isBlank(content)) {
             throw new IllegalStateException("LLM response missing message.content");
         }
-        return content;
+        JsonNode usage = root.path("usage");
+        ModelUsage modelUsage = new ModelUsage(
+            integerUsage(usage, "prompt_tokens"),
+            integerUsage(usage, "completion_tokens"),
+            integerUsage(usage, "total_tokens"),
+            Duration.ofNanos(System.nanoTime() - startedAt).toMillis()
+        );
+        return new ModelCallResult(content, modelUsage);
+    }
+
+    private static Integer integerUsage(JsonNode usage, String field) {
+        if (usage == null || usage.isMissingNode() || !usage.has(field) || !usage.get(field).canConvertToInt()) {
+            return null;
+        }
+        return usage.get(field).intValue();
     }
 
     private RuntimeConfig runtimeConfig(String requestedModel) {
@@ -593,6 +612,9 @@ public class AihrModelSeedService {
     private record RuntimeConfig(String providerCode, String modelName, String baseUrl, String apiKey, boolean configured, String source) {
     }
 
+    private record ModelCallResult(String content, ModelUsage usage) {
+    }
+
     private record ProviderData(
         String providerName,
         String providerCode,

backend/ruoyi-modules/ruoyi-aihr/src/test/java/org/dromara/aihr/service/AihrSensitiveTextTest.java

@@ -48,6 +48,9 @@ class AihrSensitiveTextTest {
         String speechControllerSource = Files.readString(controllerSource("AihrSpeechController.java"));
 
         assertTrue(modelSource.contains("AihrSensitiveText.forModel(prompt)"));
+        assertTrue(modelSource.contains("prompt_tokens"));
+        assertTrue(modelSource.contains("completion_tokens"));
+        assertTrue(modelSource.contains("latencyMs"));
         assertTrue(modelSource.contains("外部响应体已隐藏"));
         assertTrue(modelSource.contains("模型调用失败，已回退内置兜底；请检查模型供应商配置。"));
         assertTrue(sopSource.contains("AihrSensitiveText.forModel(renderPrompt(prompt.template(), queryText, context))"));

docs/BRD_IMPLEMENTATION_AUDIT.md

@@ -197,3 +197,4 @@
 - 2026-07-14 BRD 4.4 每日三题结果留痕补强：每日三题提交返回的分数此前只存在响应内，刷新任务列表后无法恢复数值分数；现新增 `aihr_practice_assignment.score`，提交、查询、员工学习页均保留“得分 + 反馈”。当前得分仍来自可解释关键词规则兜底，不把它宣称为语义级 AI 评分；后续若启用模型评估，需单独补评分模式、成本闸门和人工抽检证据。
 - 2026-07-14 BRD 4.4 每日一练评分能力复核：当前 `scoreDailyDrill` 使用参考答案关键词命中和长度规则即时反馈，属于可解释的规则兜底，不等同于 BRD 所称的语义级“AI 秒评”；正式升级前需确认模型、成本闸门、评分维度和人工抽检样本，当前不把规则分数宣称为 AI 评测结果。
 - 2026-07-14 BRD G6 成本计量复核：模型调用已有手动运行断路器，但当前没有 `llm_call_log` 或等价的租户级 token、延迟、费用持久化；由于供应商计费字段、单价和月度阈值尚未确认，本轮不新增价格硬编码或生产数据表，先把该项保留为生产前闸门。
+- 2026-07-14 TechSpec 3.8 调用反馈补强：模型测试接口成功调用时现在返回供应商 `prompt_tokens/completion_tokens/total_tokens` 和本次 `latencyMs`；未配置、成本断路器或调用失败仍返回空 usage，不伪造 token 或费用。该反馈可作为后续 `llm_call_log` 的输入，但不等于已完成月度成本计量和自动阈值降级。

frontend/src/api/aihr/model.ts

@@ -62,6 +62,12 @@ export type ModelChatResponse = {
   mode: string;
   error?: string;
   hints: string[];
+  usage?: {
+    promptTokens?: number;
+    completionTokens?: number;
+    totalTokens?: number;
+    latencyMs: number;
+  } | null;
 };
 
 export type VectorIndexStatus = {