14 Commits

Author SHA1 Message Date
kuaifan
f8eaab5a50 fix(search): 评审修复——重建自愈加探针表与锁内复检,向量模型变更自动重建
- 重建前用探针表验证 ai 端点,探不通不动真表不写 marker(下次自愈重试)
- 锁内复检 marker,消除并发进程重复重建窗口
- 检测 embedding 模型变更(vector:model 对比查询侧缓存)自动触发重建
- Ihttp 返回体 code 非 0/200 时视为失败(401/500 不再被当成功)
- 查询向量缓存键掺入模型指纹,换模型后不再命中旧维度缓存
- 批量写入按行数+字节双预算分块、块内延迟渲染 SQL,降低峰值内存
- user 同步加脏检查,字段未变不重复触发向量重算
- 日志截断超长 SQL;文档与注释同步 KB_INGEST_TOKEN 退役、截断长度对齐
2026-07-15 14:17:01 +00:00
kuaifan
746337c2f6 feat(search): 语义搜索迁移至 Manticore Auto Embeddings,向量走 ai 插件免费模型
- embedding 统一改走 ai 插件 /embeddings(免费向量模型,1024 维),删除第三方 key 分支
- 5 张向量表启用引擎 Auto Embeddings(MODEL_NAME/FROM/API_URL),写入/更新时引擎自动生成向量
- 表定义使用派生密钥 sha256(APP_KEY:embeddings),避免主密钥进入引擎元数据
- 写入改为 REPLACE(单行原子)与多行批量 REPLACE(整语句一次向量化调用,失败回退逐行)
- 结构指纹 vector:schema 标记:版本/模型/端点/APP_KEY 变化时自动清空重建并重置同步指针
- 退役 PHP 向量管道:删除 generate-vectors 命令、双指针、withVector 分支与批量向量生成方法
- 根治缺陷:全文重同步不再抹掉向量;内容编辑后向量由引擎自动重算,不再过期
2026-07-15 10:45:20 +00:00
kuaifan
23b06327d6 feat(file): 添加文件内容提取 API 支持分页读取
- FileController: 新增 fetch API 通过路径获取文本内容
  - FileController: one API 支持 with_text 参数提取文本
  - ManticoreFile: 实现分页提取 extractFileContentPaginated
  - TextExtractor: 添加 truncate 参数支持内容截取
2026-01-16 01:41:28 +00:00
kuaifan
dd20711c04 refactor: 移除冗余日志记录,优化代码可读性 2026-01-14 09:41:06 +00:00
kuaifan
f496bc5fca feat: Optimize search functionality and AI module integration
- Refactor Manticore search classes for better performance
- Update AI module with enhanced processing capabilities
- Improve Apps module functionality
- Enhance SearchBox Vue component with new features
2026-01-06 07:25:23 +00:00
kuaifan
8ab810c670 feat: 将 Manticore 相关检查更新为使用 "search" 应用 2026-01-05 05:51:48 +00:00
kuaifan
4b0f4e388c feat: 优化 Manticore 相关描述 2026-01-04 13:30:03 +00:00
kuaifan
90a5624877 feat: 添加用户标签功能,更新用户索引以支持标签创建、更新和删除事件 2026-01-04 07:13:13 +00:00
kuaifan
ea0d27fdea feat: 添加 Manticore 同步命令通用锁机制,优化信号处理与锁管理 2026-01-03 23:09:50 +00:00
kuaifan
9a8304d595 feat: 增强 Manticore 向量更新逻辑,记录更新失败的 ID 2026-01-03 21:59:44 +00:00
kuaifan
e020a80020 feat: Add batch embedding retrieval and vector update methods for Manticore integration
- Implemented `getBatchEmbeddings` method in AI module for retrieving embeddings for multiple texts.
- Added vector update methods for messages, files, tasks, projects, and users in ManticoreBase.
- Enhanced ManticoreFile, ManticoreMsg, ManticoreProject, ManticoreTask, and ManticoreUser to support vector generation during sync operations.
- Introduced `generateVectorsBatch` methods for batch processing of vector generation in Manticore modules.
- Updated ManticoreSyncTask to handle incremental updates and vector generation asynchronously.
2026-01-03 15:19:23 +00:00
kuaifan
c08323e1ea feat: 迁移至 MVA 权限方案
- 表结构:为 file/project/task_vectors 添加 allowed_users MULTI 字段
- 删除关系表:file_users, project_users, task_users
- 搜索:使用 allowed_users = userid 进行权限过滤
- 同步:sync 时自动计算并写入 allowed_users
- 级联:项目成员变更异步级联 v=1 任务,任务成员变更递归更新子任务
- 覆盖场景:visibility/parent_id/project_id 变更、子任务升级主任务等
2026-01-02 02:03:21 +00:00
kuaifan
fdf5ceeaab feat: Enhance Manticore integration and AI model support
- Added support for specifying vector dimensions in AI payloads for compatible vendors.
- Updated default AI model from 'text-embedding-ada-002' to 'text-embedding-3-small'.
- Refactored ManticoreBase to bind parameters explicitly for PDO statements, improving type handling.
- Adjusted SQL queries across Manticore modules to remove content previews and ensure inline vector values.
- Updated content preview handling in ManticoreFile, ManticoreProject, ManticoreTask, and ManticoreUser to use substrings for better data management.
2026-01-01 08:59:54 +00:00
kuaifan
48ef4cfdef refactor: 使用 Manticore Search 替换 SeekDB 2026-01-01 03:17:27 +00:00