diff --git a/backend/package/yuxi/knowledge/base.py b/backend/package/yuxi/knowledge/base.py
index 0fd7d8da..278c8e8c 100644
--- a/backend/package/yuxi/knowledge/base.py
+++ b/backend/package/yuxi/knowledge/base.py
@@ -7,6 +7,7 @@ from yuxi.knowledge.chunking.ragflow_like.presets import (
     ensure_chunk_defaults_in_additional_params,
     resolve_chunk_processing_params,
 )
+from yuxi.knowledge.utils import sanitize_processing_params
 from yuxi.utils import logger
 from yuxi.utils.datetime_utils import coerce_any_to_utc_datetime, utc_isoformat
 
@@ -667,7 +668,6 @@ class KnowledgeBase(ABC):
                         "type": file_info.get("file_type", ""),
                         "status": file_info.get("status", "done"),
                         "created_at": created_at,
-                        "processing_params": file_info.get("processing_params", None),
                         "is_folder": file_info.get("is_folder", False),
                         "parent_id": file_info.get("parent_id", None),
                     }
@@ -1057,9 +1057,11 @@ class KnowledgeBase(ABC):
                     "content_hash": record.content_hash,
                     "size": record.file_size,
                     "content_type": record.content_type,
-                    "processing_params": resolve_chunk_processing_params(
-                        kb_additional_params=kb_additional_params,
-                        file_processing_params=record.processing_params,
+                    "processing_params": sanitize_processing_params(
+                        resolve_chunk_processing_params(
+                            kb_additional_params=kb_additional_params,
+                            file_processing_params=record.processing_params,
+                        )
                     ),
                     "is_folder": record.is_folder,
                     "error": record.error_message,
@@ -1153,7 +1155,7 @@ class KnowledgeBase(ABC):
                     "content_hash": meta.get("content_hash"),
                     "file_size": meta.get("size"),
                     "content_type": meta.get("content_type"),
-                    "processing_params": meta.get("processing_params"),
+                    "processing_params": sanitize_processing_params(meta.get("processing_params")),
                     "is_folder": meta.get("is_folder", False),
                     "error_message": meta.get("error"),
                     "created_by": str(meta.get("created_by")) if meta.get("created_by") else None,
@@ -1207,7 +1209,7 @@ class KnowledgeBase(ABC):
                 "content_hash": meta.get("content_hash"),
                 "file_size": meta.get("size"),
                 "content_type": meta.get("content_type"),
-                "processing_params": meta.get("processing_params"),
+                "processing_params": sanitize_processing_params(meta.get("processing_params")),
                 "is_folder": meta.get("is_folder", False),
                 "error_message": meta.get("error"),
                 "created_by": str(meta.get("created_by")) if meta.get("created_by") else None,
diff --git a/backend/package/yuxi/knowledge/utils/__init__.py b/backend/package/yuxi/knowledge/utils/__init__.py
index 551ab5db..1a983f9e 100644
--- a/backend/package/yuxi/knowledge/utils/__init__.py
+++ b/backend/package/yuxi/knowledge/utils/__init__.py
@@ -10,6 +10,7 @@ from .kb_utils import (
     get_embedding_config,
     merge_processing_params,
     prepare_item_metadata,
+    sanitize_processing_params,
     split_text_into_chunks,
     validate_file_path,
 )
@@ -18,6 +19,7 @@ __all__ = [
     "calculate_content_hash",
     "get_embedding_config",
     "prepare_item_metadata",
+    "sanitize_processing_params",
     "split_text_into_chunks",
     "merge_processing_params",
     "validate_file_path",
diff --git a/backend/package/yuxi/knowledge/utils/kb_utils.py b/backend/package/yuxi/knowledge/utils/kb_utils.py
index 20040ed1..3641c65e 100644
--- a/backend/package/yuxi/knowledge/utils/kb_utils.py
+++ b/backend/package/yuxi/knowledge/utils/kb_utils.py
@@ -91,6 +91,17 @@ def _unescape_separator(separator: str | None) -> str | None:
     return separator
 
 
+def sanitize_processing_params(params: dict | None) -> dict | None:
+    """移除批次级临时字段，避免写入单文件元数据。"""
+    if not params:
+        return None
+
+    sanitized = params.copy()
+    sanitized.pop("_preprocessed_map", None)
+    sanitized.pop("content_hashes", None)
+    return sanitized
+
+
 def split_text_into_chunks(text: str, file_id: str, filename: str, params: dict = {}) -> list[dict]:
     """
     将文本分割成块，使用 LangChain 的 MarkdownTextSplitter 进行智能分割
@@ -222,9 +233,7 @@ async def prepare_item_metadata(item: str, content_type: str, db_id: str, params
         }
 
         if params:
-            # 移除内部参数以免污染 metadata
-            safe_params = params.copy()
-            safe_params.pop("_preprocessed_map", None)
+            safe_params = sanitize_processing_params(params) or {}
             # 覆盖 content_type 为 file，确保后续解析走文件流程（MinIO 下载 -> HTML 解析）
             # 而不是再次尝试作为 URL 抓取
             safe_params["content_type"] = "file"
@@ -314,7 +323,7 @@ async def prepare_item_metadata(item: str, content_type: str, db_id: str, params
 
     # 保存处理参数到元数据
     if params:
-        metadata["processing_params"] = params.copy()
+        metadata["processing_params"] = sanitize_processing_params(params)
 
     return metadata
 
diff --git a/backend/test/test_ragflow_like_chunking.py b/backend/test/test_ragflow_like_chunking.py
index 88f7d684..3e900c63 100644
--- a/backend/test/test_ragflow_like_chunking.py
+++ b/backend/test/test_ragflow_like_chunking.py
@@ -13,6 +13,7 @@ from yuxi.knowledge.chunking.ragflow_like.presets import (
     map_to_internal_parser_id,
     resolve_chunk_processing_params,
 )
+from yuxi.knowledge.utils.kb_utils import sanitize_processing_params
 
 
 def test_general_maps_to_naive() -> None:
@@ -218,3 +219,15 @@ def test_laws_markdown_articles_should_not_collapse_into_chapter_chunk() -> None
     # 条级切分时，第一条与第二条不应被合并到同一块。
     assert all("第二条" not in chunk for chunk in first_article_chunks)
     assert max(count_tokens(ck["content"]) for ck in chunks) <= 120
+
+
+def test_sanitize_processing_params_should_drop_batch_only_fields() -> None:
+    sanitized = sanitize_processing_params(
+        {
+            "chunk_preset_id": "general",
+            "content_hashes": {"a.md": "hash-a"},
+            "_preprocessed_map": {"a.md": {"path": "/tmp/a.md"}},
+        }
+    )
+
+    assert sanitized == {"chunk_preset_id": "general"}
diff --git a/docs/develop-guides/roadmap.md b/docs/develop-guides/roadmap.md
index 17ea45c2..844fc671 100644
--- a/docs/develop-guides/roadmap.md
+++ b/docs/develop-guides/roadmap.md
@@ -72,6 +72,7 @@
 - 修复沙盒后端接入回归：补齐 composite backend 的 `sandbox_backend` 参数、限制 `/api/sandbox/prepare` 仅允许访问当前用户线程、确保 `release()` 之后的 `destroy()` 会真正停止热池容器，并恢复 docker-compose 的完整模式默认值
 - 重构沙盒为 deer-flow 风格的 AIO provider：切换为 thread-local sandbox、统一 `/home/yuxi/user-data/{workspace,uploads,outputs}` 固定路径、移除公开 `/api/sandbox/*` 生命周期接口，并补充 lite 模式下的 provider 生命周期、filesystem API 与 sandbox 复用/隔离 E2E 验证
 - 调整聊天附件存储链路：线程附件改为直接落盘到 `saves/threads/<thread_id>/user-data/uploads`，解析成功后额外生成 `uploads/attachments/*.md`，不再依赖 MinIO 或显式上传到 sandbox
+- 修复知识库文件列表包体异常膨胀：上传阶段不再把批次级 `content_hashes` 写入每个文件的 `processing_params`，并从数据库详情列表接口中移除该字段，改为按需读取单文件详情
 
 ## v0.4
 
diff --git a/web/src/components/FileTable.vue b/web/src/components/FileTable.vue
index 38100b48..c1e7c22a 100644
--- a/web/src/components/FileTable.vue
+++ b/web/src/components/FileTable.vue
@@ -1157,23 +1157,32 @@ const handleParseFile = async (record) => {
   await store.parseFiles([record.file_id])
 }
 
+const defaultIndexParams = {
+  chunk_size: 1000,
+  chunk_overlap: 200,
+  qa_separator: '',
+  chunk_preset_id: ''
+}
+
+const loadRecordProcessingParams = async (record) => {
+  if (record?.processing_params) {
+    return record.processing_params
+  }
+
+  const detail = await documentApi.getDocumentInfo(store.databaseId, record.file_id)
+  return detail?.processing_params || null
+}
+
 const handleIndexFile = async (record) => {
   closePopover(record.file_id)
-  // 打开参数配置弹窗
   currentIndexFileIds.value = [record.file_id]
   isBatchIndexOperation.value = false
   indexConfigModalTitle.value = '入库参数配置'
 
-  if (record?.processing_params) {
-    Object.assign(indexParams.value, record.processing_params)
-  } else {
-    // Reset to defaults if no existing params
-    Object.assign(indexParams.value, {
-      chunk_size: 1000,
-      chunk_overlap: 200,
-      qa_separator: '',
-      chunk_preset_id: ''
-    })
+  Object.assign(indexParams.value, defaultIndexParams)
+  const processingParams = await loadRecordProcessingParams(record)
+  if (processingParams) {
+    Object.assign(indexParams.value, processingParams)
   }
 
   indexConfigModalVisible.value = true
@@ -1185,11 +1194,12 @@ const handleReindexFile = async (record) => {
   isBatchIndexOperation.value = false
   indexConfigModalTitle.value = '重新入库参数配置'
 
-  if (record?.processing_params) {
-    Object.assign(indexParams.value, record.processing_params)
+  Object.assign(indexParams.value, defaultIndexParams)
+  const processingParams = await loadRecordProcessingParams(record)
+  if (processingParams) {
+    Object.assign(indexParams.value, processingParams)
   }
 
-  // 显示参数配置模态框
   indexConfigModalVisible.value = true
 }
 
@@ -1230,12 +1240,7 @@ const handleIndexConfigCancel = () => {
   currentIndexFileIds.value = []
   isBatchIndexOperation.value = false
   // 重置参数为默认值
-  Object.assign(indexParams.value, {
-    chunk_size: 1000,
-    chunk_overlap: 200,
-    qa_separator: '',
-    chunk_preset_id: ''
-  })
+  Object.assign(indexParams.value, defaultIndexParams)
 }
 
 // 导入工具函数