Show retrieved knowledge chunks in audit logs

ai_knowledge_base_v2/apps/admin-web/src/App.vue

@@ -1463,6 +1463,21 @@ function buildChatQuery() {
             </el-tab-pane>
             <el-tab-pane label="AI 请求" name="aiLogs">
               <el-table :data="aiLogs" stripe>
+                <el-table-column type="expand" width="48">
+                  <template #default="{ row }">
+                    <section v-if="row.retrievedChunks?.length" class="retrieval-chunks">
+                      <article v-for="chunk in row.retrievedChunks" :key="`${row.id}-${chunk.index}`" class="retrieval-chunk">
+                        <div class="chunk-head">
+                          <strong>#{{ chunk.index }} {{ chunk.knowledgeName || '未知知识库' }}</strong>
+                          <span>{{ chunk.title || '未命名片段' }}</span>
+                        </div>
+                        <pre>{{ chunk.content }}</pre>
+                        <a v-if="chunk.sourceUrl" :href="chunk.sourceUrl" target="_blank" rel="noreferrer">查看来源</a>
+                      </article>
+                    </section>
+                    <el-empty v-else description="本条请求未保存召回片段" :image-size="64" />
+                  </template>
+                </el-table-column>
                 <el-table-column prop="sessionId" label="会话ID" width="90" />
                 <el-table-column prop="modelName" label="模型" width="160" />
                 <el-table-column prop="knowledgeIds" label="知识库" width="120" />
@@ -1528,6 +1543,17 @@ function buildChatQuery() {
               <span>耗时：{{ log.costMs || '-' }}ms</span>
               <span>时间：{{ log.createdAt }}</span>
             </div>
+            <section v-if="log.retrievedChunks?.length" class="retrieval-chunks">
+              <article v-for="chunk in log.retrievedChunks" :key="`${log.id}-${chunk.index}`" class="retrieval-chunk">
+                <div class="chunk-head">
+                  <strong>#{{ chunk.index }} {{ chunk.knowledgeName || '未知知识库' }}</strong>
+                  <span>{{ chunk.title || '未命名片段' }}</span>
+                </div>
+                <pre>{{ chunk.content }}</pre>
+                <a v-if="chunk.sourceUrl" :href="chunk.sourceUrl" target="_blank" rel="noreferrer">查看来源</a>
+              </article>
+            </section>
+            <el-empty v-else description="本条请求未保存召回片段" :image-size="64" />
             <pre class="prompt-preview">{{ log.prompt || '无 Prompt 记录' }}</pre>
             <p v-if="log.errorMessage" class="error-text">{{ log.errorMessage }}</p>
           </el-collapse-item>

ai_knowledge_base_v2/apps/admin-web/src/styles.css

@@ -712,6 +712,58 @@ textarea {
   font-size: 13px;
 }
 
+.retrieval-chunks {
+  display: grid;
+  gap: 12px;
+  margin: 12px 0;
+}
+
+.retrieval-chunk {
+  padding: 12px;
+  border: 1px solid #dce8e4;
+  border-radius: 6px;
+  background: #fbfdfc;
+}
+
+.chunk-head {
+  display: flex;
+  flex-wrap: wrap;
+  gap: 8px 12px;
+  align-items: baseline;
+  margin-bottom: 8px;
+  color: #203832;
+}
+
+.chunk-head strong {
+  color: #0f735d;
+}
+
+.chunk-head span {
+  color: #60746d;
+  font-size: 13px;
+}
+
+.retrieval-chunk pre {
+  max-height: 260px;
+  margin: 0;
+  padding: 10px;
+  overflow: auto;
+  border-radius: 6px;
+  background: #f4f7f6;
+  color: #263832;
+  white-space: pre-wrap;
+  word-break: break-word;
+  font-family: "SFMono-Regular", Consolas, "PingFang SC", monospace;
+  line-height: 1.65;
+}
+
+.retrieval-chunk a {
+  display: inline-flex;
+  margin-top: 8px;
+  color: #0f735d;
+  text-decoration: none;
+}
+
 .error-text {
   color: #b42318;
 }

ai_knowledge_base_v2/apps/admin-web/src/types/api.ts

@@ -133,6 +133,7 @@ export interface AiLogRecord {
   modelName?: string | null;
   knowledgeIds?: string | null;
   retrieveCount: number;
+  retrievedChunks: RetrievedKnowledgeChunk[];
   inputToken?: number | null;
   outputToken?: number | null;
   totalToken?: number | null;
@@ -143,6 +144,15 @@ export interface AiLogRecord {
   createdAt: string;
 }
 
+export interface RetrievedKnowledgeChunk {
+  index: number;
+  knowledgeId?: number | null;
+  knowledgeName: string;
+  title: string;
+  content: string;
+  sourceUrl?: string | null;
+}
+
 export interface ChatDetail {
   session: ChatRecord;
   messages: ChatMessageRecord[];

ai_knowledge_base_v2/apps/backend/alembic/versions/0003_ai_log_retrieved_chunks.py

@@ -0,0 +1,17 @@
+from __future__ import annotations
+
+from alembic import op
+import sqlalchemy as sa
+
+revision = "0003_ai_log_retrieved_chunks"
+down_revision = "0002_expand_model_config"
+branch_labels = None
+depends_on = None
+
+
+def upgrade() -> None:
+    op.add_column("sys_ai_request_log", sa.Column("retrieved_chunks", sa.Text(), nullable=True))
+
+
+def downgrade() -> None:
+    op.drop_column("sys_ai_request_log", "retrieved_chunks")

ai_knowledge_base_v2/apps/backend/app/api/admin_records.py

@@ -1,8 +1,10 @@
 from __future__ import annotations
 
 import csv
+import json
 from datetime import datetime
 from io import StringIO
+from typing import Any
 
 from fastapi import APIRouter, Depends, Query
 from fastapi.responses import Response
@@ -236,6 +238,7 @@ def _ai_log_dict(log: AiRequestLog, *, include_prompt: bool = False) -> dict:
         "modelName": log.model_name,
         "knowledgeIds": log.knowledge_ids,
         "retrieveCount": log.retrieve_count,
+        "retrievedChunks": _parse_retrieved_chunks(log.retrieved_chunks),
         "inputToken": log.input_token,
         "outputToken": log.output_token,
         "totalToken": log.total_token,
@@ -247,3 +250,29 @@ def _ai_log_dict(log: AiRequestLog, *, include_prompt: bool = False) -> dict:
     if include_prompt:
         data["prompt"] = log.prompt
     return data
+
+
+def _parse_retrieved_chunks(raw: str | None) -> list[dict[str, Any]]:
+    if not raw:
+        return []
+    try:
+        chunks = json.loads(raw)
+    except json.JSONDecodeError:
+        return []
+    if not isinstance(chunks, list):
+        return []
+    normalized: list[dict[str, Any]] = []
+    for index, item in enumerate(chunks, start=1):
+        if not isinstance(item, dict):
+            continue
+        normalized.append(
+            {
+                "index": item.get("index") or index,
+                "knowledgeId": item.get("knowledgeId"),
+                "knowledgeName": item.get("knowledgeName") or "",
+                "title": item.get("title") or "",
+                "content": item.get("content") or "",
+                "sourceUrl": item.get("sourceUrl") or None,
+            }
+        )
+    return normalized

ai_knowledge_base_v2/apps/backend/app/models/logs.py

@@ -19,6 +19,7 @@ class AiRequestLog(Base):
     prompt: Mapped[str | None] = mapped_column(Text, nullable=True)
     knowledge_ids: Mapped[str | None] = mapped_column(String(500), nullable=True)
     retrieve_count: Mapped[int] = mapped_column(Integer, default=0, nullable=False)
+    retrieved_chunks: Mapped[str | None] = mapped_column(Text, nullable=True)
     input_token: Mapped[int | None] = mapped_column(Integer, nullable=True)
     output_token: Mapped[int | None] = mapped_column(Integer, nullable=True)
     total_token: Mapped[int | None] = mapped_column(Integer, nullable=True)

ai_knowledge_base_v2/apps/backend/app/services/ai_request_log_service.py

@@ -1,8 +1,12 @@
 from __future__ import annotations
 
+import json
+from collections.abc import Sequence
+
 from sqlalchemy.orm import Session
 
 from app.models.logs import AiRequestLog
+from app.services.rag_service import RetrievedChunk
 
 
 class AiRequestLogService:
@@ -20,6 +24,7 @@ class AiRequestLogService:
         input_token: int,
         output_token: int,
         cost_ms: int,
+        retrieved_chunks: Sequence[RetrievedChunk] | None = None,
     ) -> None:
         db.add(
             AiRequestLog(
@@ -30,6 +35,7 @@ class AiRequestLogService:
                 prompt=prompt,
                 knowledge_ids=knowledge_ids,
                 retrieve_count=retrieve_count,
+                retrieved_chunks=_dump_retrieved_chunks(retrieved_chunks),
                 input_token=input_token,
                 output_token=output_token,
                 total_token=input_token + output_token,
@@ -51,6 +57,7 @@ class AiRequestLogService:
         retrieve_count: int,
         cost_ms: int,
         error_message: str,
+        retrieved_chunks: Sequence[RetrievedChunk] | None = None,
     ) -> None:
         db.add(
             AiRequestLog(
@@ -61,8 +68,26 @@ class AiRequestLogService:
                 prompt=prompt,
                 knowledge_ids=knowledge_ids,
                 retrieve_count=retrieve_count,
+                retrieved_chunks=_dump_retrieved_chunks(retrieved_chunks),
                 cost_ms=cost_ms,
                 status="FAILED",
                 error_message=error_message,
             )
         )
+
+
+def _dump_retrieved_chunks(chunks: Sequence[RetrievedChunk] | None) -> str | None:
+    if not chunks:
+        return None
+    payload = [
+        {
+            "index": index,
+            "knowledgeId": chunk.knowledge_id,
+            "knowledgeName": chunk.knowledge_name,
+            "title": chunk.title,
+            "content": chunk.content,
+            "sourceUrl": chunk.source_url,
+        }
+        for index, chunk in enumerate(chunks, start=1)
+    ]
+    return json.dumps(payload, ensure_ascii=False)

ai_knowledge_base_v2/apps/backend/app/services/chat_service.py

@@ -87,6 +87,7 @@ class ChatService:
         db.flush()
 
         started_at = perf_counter()
+        rag_result = None
         try:
             # 获取历史消息（不含刚插入的 user_message，它还没 flush id）
             history = list(
@@ -121,11 +122,12 @@ class ChatService:
                 message_id=user_message.id,
                 user_id=user.id,
                 model_name=None,
-                prompt=normalized_question,
+                prompt=rag_result.prompt if rag_result is not None else normalized_question,
-                knowledge_ids=None,
+                knowledge_ids=rag_result.knowledge_ids if rag_result is not None else None,
-                retrieve_count=0,
+                retrieve_count=len(rag_result.chunks) if rag_result is not None else 0,
                 cost_ms=cost_ms,
                 error_message=str(exc),
+                retrieved_chunks=rag_result.chunks if rag_result is not None else None,
             )
             db.commit()
             raise HTTPException(
@@ -168,6 +170,7 @@ class ChatService:
             input_token=completion.input_token,
             output_token=completion.output_token,
             cost_ms=cost_ms,
+            retrieved_chunks=rag_result.chunks,
         )
         db.commit()
         return completion.answer

ai_knowledge_base_v2/apps/backend/app/services/chat_stream_service.py

@@ -80,6 +80,7 @@ class ChatStreamService:
                 retrieve_count=len(rag_result.chunks) if rag_result is not None else 0,
                 cost_ms=cost_ms,
                 error_message=str(exc),
+                retrieved_chunks=rag_result.chunks if rag_result is not None else None,
             )
             db.commit()
             raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail=str(exc)) from exc
@@ -98,6 +99,7 @@ class ChatStreamService:
                 retrieve_count=len(rag_result.chunks) if rag_result is not None else 0,
                 cost_ms=cost_ms,
                 error_message="模型未返回有效内容",
+                retrieved_chunks=rag_result.chunks if rag_result is not None else None,
             )
             db.commit()
             raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="模型未返回有效内容")
@@ -137,6 +139,7 @@ class ChatStreamService:
             input_token=model_response.input_token if model_response is not None else 0,
             output_token=_rough_token_count(answer),
             cost_ms=cost_ms,
+            retrieved_chunks=rag_result.chunks if rag_result is not None else None,
         )
         db.commit()
 
@@ -205,6 +208,7 @@ class ChatStreamService:
                 retrieve_count=len(rag_result.chunks) if rag_result is not None else 0,
                 cost_ms=cost_ms,
                 error_message=str(exc),
+                retrieved_chunks=rag_result.chunks if rag_result is not None else None,
             )
             db.commit()
             raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail=str(exc)) from exc
@@ -223,6 +227,7 @@ class ChatStreamService:
                 retrieve_count=len(rag_result.chunks) if rag_result is not None else 0,
                 cost_ms=cost_ms,
                 error_message="模型未返回有效内容",
+                retrieved_chunks=rag_result.chunks if rag_result is not None else None,
             )
             db.commit()
             raise HTTPException(status_code=status.HTTP_502_BAD_GATEWAY, detail="模型未返回有效内容")
@@ -295,6 +300,7 @@ def _write_success(
         input_token=model_response.input_token if model_response is not None else 0,
         output_token=_rough_token_count(answer),
         cost_ms=cost_ms,
+        retrieved_chunks=rag_result.chunks if rag_result is not None else None,
     )
     db.commit()