diff --git a/electron/main/ai/tools/definitions/deep-search-messages.ts b/electron/main/ai/tools/definitions/deep-search-messages.ts
new file mode 100644
index 0000000..cab27d1
--- /dev/null
+++ b/electron/main/ai/tools/definitions/deep-search-messages.ts
@@ -0,0 +1,49 @@
+import { Type } from '@mariozechner/pi-ai'
+import type { AgentTool } from '@mariozechner/pi-agent-core'
+import type { ToolContext } from '../types'
+import * as workerManager from '../../../worker/workerManager'
+import { parseExtendedTimeParams } from '../utils/time-params'
+import { formatTimeRange } from '../utils/format'
+import { timeParamProperties } from '../utils/schemas'
+
+const schema = Type.Object({
+  keywords: Type.Array(Type.String(), { description: 'ai.tools.deep_search_messages.params.keywords' }),
+  sender_id: Type.Optional(Type.Number({ description: 'ai.tools.deep_search_messages.params.sender_id' })),
+  limit: Type.Optional(Type.Number({ description: 'ai.tools.deep_search_messages.params.limit' })),
+  ...timeParamProperties,
+})
+
+export function createTool(context: ToolContext): AgentTool<typeof schema> {
+  return {
+    name: 'deep_search_messages',
+    label: 'deep_search_messages',
+    description: 'ai.tools.deep_search_messages.desc',
+    parameters: schema,
+    execute: async (_toolCallId, params) => {
+      const { sessionId, timeFilter: contextTimeFilter, maxMessagesLimit, locale } = context
+      const limit = Math.min(maxMessagesLimit || params.limit || 1000, 50000)
+      const effectiveTimeFilter = parseExtendedTimeParams(params, contextTimeFilter)
+
+      const result = await workerManager.deepSearchMessages(
+        sessionId,
+        params.keywords,
+        effectiveTimeFilter,
+        limit,
+        0,
+        params.sender_id
+      )
+
+      const data = {
+        total: result.total,
+        returned: result.messages.length,
+        timeRange: formatTimeRange(effectiveTimeFilter, locale),
+        rawMessages: result.messages,
+      }
+
+      return {
+        content: [{ type: 'text', text: JSON.stringify(data) }],
+        details: data,
+      }
+    },
+  }
+}
diff --git a/electron/main/ai/tools/definitions/index.ts b/electron/main/ai/tools/definitions/index.ts
index 2d710f2..858e3dc 100644
--- a/electron/main/ai/tools/definitions/index.ts
+++ b/electron/main/ai/tools/definitions/index.ts
@@ -9,6 +9,7 @@ import type { ToolRegistryEntry } from '../types'
 
 import { createTool as createGetChatOverview } from './get-chat-overview'
 import { createTool as createSearchMessages } from './search-messages'
+import { createTool as createDeepSearchMessages } from './deep-search-messages'
 import { createTool as createGetRecentMessages } from './get-recent-messages'
 import { createTool as createGetMessageContext } from './get-message-context'
 import { createTool as createSearchSessions } from './search-sessions'
@@ -30,6 +31,7 @@ export const TOOL_REGISTRY: ToolRegistryEntry[] = [
   // ==================== Core 工具（始终加载） ====================
   { name: 'get_chat_overview', factory: createGetChatOverview, category: 'core' },
   { name: 'search_messages', factory: createSearchMessages, category: 'core' },
+  { name: 'deep_search_messages', factory: createDeepSearchMessages, category: 'core' },
   { name: 'get_recent_messages', factory: createGetRecentMessages, category: 'core' },
   { name: 'get_message_context', factory: createGetMessageContext, category: 'core' },
   { name: 'search_sessions', factory: createSearchSessions, category: 'core' },
diff --git a/electron/main/database/core.ts b/electron/main/database/core.ts
index 9b2af01..dcda1b3 100644
--- a/electron/main/database/core.ts
+++ b/electron/main/database/core.ts
@@ -120,6 +120,14 @@ function createDatabase(sessionId: string): Database.Database {
     CREATE INDEX IF NOT EXISTS idx_context_session ON message_context(session_id);
   `)
 
+  db.exec(`
+    CREATE VIRTUAL TABLE IF NOT EXISTS message_fts USING fts5(
+      content,
+      content='',
+      content_rowid=id
+    )
+  `)
+
   return db
 }
 
diff --git a/electron/main/database/migrations.ts b/electron/main/database/migrations.ts
index 8653a12..6ebd31e 100644
--- a/electron/main/database/migrations.ts
+++ b/electron/main/database/migrations.ts
@@ -12,6 +12,7 @@
 
 import type Database from 'better-sqlite3'
 import { t } from '../i18n'
+import { tokenizeForFts } from '../nlp/ftsTokenizer'
 
 /** 迁移脚本接口 */
 interface Migration {
@@ -35,7 +36,7 @@ export interface MigrationInfo {
 }
 
 /** 当前 schema 版本（最新迁移的版本号） */
-export const CURRENT_SCHEMA_VERSION = 3
+export const CURRENT_SCHEMA_VERSION = 4
 
 /**
  * 迁移脚本列表
@@ -138,6 +139,56 @@ const migrations: Migration[] = [
       }
     },
   },
+  {
+    version: 4,
+    descriptionKey: 'database.migrationV4Desc',
+    userMessageKey: 'database.migrationV4Message',
+    up: (db) => {
+      const hasTable = db
+        .prepare("SELECT name FROM sqlite_master WHERE type='table' AND name='message_fts'")
+        .get()
+      if (hasTable) return
+
+      db.exec(`
+        CREATE VIRTUAL TABLE IF NOT EXISTS message_fts USING fts5(
+          content,
+          content='',
+          content_rowid=id
+        )
+      `)
+
+      const BATCH_SIZE = 5000
+      const insertFts = db.prepare('INSERT INTO message_fts(rowid, content) VALUES (?, ?)')
+
+      const countRow = db
+        .prepare(
+          "SELECT COUNT(*) as total FROM message WHERE type = 0 AND content IS NOT NULL AND content != ''"
+        )
+        .get() as { total: number }
+
+      let offset = 0
+      while (offset < countRow.total) {
+        const rows = db
+          .prepare(
+            `SELECT id, content FROM message
+             WHERE type = 0 AND content IS NOT NULL AND content != ''
+             ORDER BY id ASC LIMIT ? OFFSET ?`
+          )
+          .all(BATCH_SIZE, offset) as Array<{ id: number; content: string }>
+
+        if (rows.length === 0) break
+
+        for (const row of rows) {
+          const tokens = tokenizeForFts(row.content)
+          if (tokens) {
+            insertFts.run(row.id, tokens)
+          }
+        }
+
+        offset += BATCH_SIZE
+      }
+    },
+  },
 ]
 
 /**
diff --git a/electron/main/i18n/locales/en-US.ts b/electron/main/i18n/locales/en-US.ts
index c221617..0de45aa 100644
--- a/electron/main/i18n/locales/en-US.ts
+++ b/electron/main/i18n/locales/en-US.ts
@@ -41,6 +41,8 @@ export default {
     migrationV2Message: 'Support member roles, message reply relationships and reply preview',
     migrationV3Desc: 'Add session index tables (chat_session, message_context) and session_gap_threshold field',
     migrationV3Message: 'Support session timeline browsing and AI-enhanced analysis',
+    migrationV4Desc: 'Create FTS5 full-text search index (message_fts) and build index data',
+    migrationV4Message: 'Enable full-text search for significantly faster keyword search',
     integrityError:
       'Database structure is incomplete: missing meta table. Please delete this database file and re-import.',
     checkFailed: 'Database check failed: {{error}}',
@@ -72,6 +74,20 @@ export default {
             'End time, format "YYYY-MM-DD HH:mm", e.g. "2024-03-15 18:30". Overrides year/month/day/hour when specified',
         },
       },
+      deep_search_messages: {
+        desc: 'Exact substring match search for chat records. Slower but never misses any message containing the keyword. Use when regular search (search_messages) results are insufficient, or when searching for partial words or single characters.',
+        params: {
+          keywords: 'List of search keywords, using substring match (LIKE). Returns messages matching any keyword',
+          sender_id: 'Sender member ID for filtering messages from a specific member',
+          limit: 'Message count limit, default 1000, max 50000',
+          year: 'Filter messages by year, e.g. 2024',
+          month: 'Filter messages by month (1-12), use with year',
+          day: 'Filter messages by day (1-31), use with year and month',
+          hour: 'Filter messages by hour (0-23), use with year, month, and day',
+          start_time: 'Start time, format "YYYY-MM-DD HH:mm". Overrides year/month/day/hour when specified',
+          end_time: 'End time, format "YYYY-MM-DD HH:mm". Overrides year/month/day/hour when specified',
+        },
+      },
       get_recent_messages: {
         desc: 'Get chat messages within a specified time period. Suitable for overview questions like "what has everyone been chatting about recently" or "what was discussed in month X". Supports minute-level time queries.',
         params: {
diff --git a/electron/main/i18n/locales/ja-JP.ts b/electron/main/i18n/locales/ja-JP.ts
index 97dc5bb..0b2224f 100644
--- a/electron/main/i18n/locales/ja-JP.ts
+++ b/electron/main/i18n/locales/ja-JP.ts
@@ -42,6 +42,8 @@ export default {
     migrationV3Desc:
       'セッションインデックス関連テーブル（chat_session、message_context）と session_gap_threshold フィールドを追加',
     migrationV3Message: 'セッションのタイムライン表示と AI 拡張分析に対応',
+    migrationV4Desc: 'FTS5 全文検索インデックス（message_fts）を作成しインデックスデータを構築',
+    migrationV4Message: '全文検索に対応し、キーワード検索速度が大幅に向上',
     integrityError:
       'データベース構造が不完全です：meta テーブルがありません。このデータベースファイルを削除して再インポートすることをお勧めします。',
     checkFailed: 'データベースチェックに失敗しました: {{error}}',
@@ -73,6 +75,20 @@ export default {
             '終了時刻。形式 "YYYY-MM-DD HH:mm"（例："2024-03-15 18:30"）。指定すると year/month/day/hour パラメータを上書きする',
         },
       },
+      deep_search_messages: {
+        desc: '完全部分文字列マッチでチャット履歴を検索する。速度は遅いが、キーワードを含むメッセージを漏らさない。通常の検索（search_messages）の結果が不十分な場合や、部分的な単語・単一文字を検索する場合に使用する。',
+        params: {
+          keywords: '検索キーワードリスト。部分文字列マッチ（LIKE）を使用し、いずれかのキーワードにマッチしたメッセージを返す',
+          sender_id: '送信者のメンバー ID。特定メンバーの送信メッセージをフィルタリングする',
+          limit: '返却メッセージ数の上限。デフォルト 1000、最大 50000',
+          year: '指定年のメッセージをフィルタリング（例：2024）',
+          month: '指定月のメッセージをフィルタリング（1-12）。year と併用する必要がある',
+          day: '指定日のメッセージをフィルタリング（1-31）。year と month と併用する必要がある',
+          hour: '指定時間のメッセージをフィルタリング（0-23）。year、month、day と併用する必要がある',
+          start_time: '開始時刻。形式 "YYYY-MM-DD HH:mm"。指定すると year/month/day/hour パラメータを上書きする',
+          end_time: '終了時刻。形式 "YYYY-MM-DD HH:mm"。指定すると year/month/day/hour パラメータを上書きする',
+        },
+      },
       get_recent_messages: {
         desc: '指定期間内のグループチャットメッセージを取得する。「最近みんな何を話していた？」「X月にグループで何が話題だった？」などの概要的な質問に適している。分単位の精度で時間クエリをサポートする。',
         params: {
diff --git a/electron/main/i18n/locales/zh-CN.ts b/electron/main/i18n/locales/zh-CN.ts
index e2f4690..e936c65 100644
--- a/electron/main/i18n/locales/zh-CN.ts
+++ b/electron/main/i18n/locales/zh-CN.ts
@@ -41,6 +41,8 @@ export default {
     migrationV2Message: '支持成员角色、消息回复关系和回复内容预览',
     migrationV3Desc: '添加会话索引相关表（chat_session、message_context）和 session_gap_threshold 字段',
     migrationV3Message: '支持会话时间轴浏览和 AI 增强分析功能',
+    migrationV4Desc: '创建 FTS5 全文搜索索引（message_fts）并构建索引数据',
+    migrationV4Message: '支持全文搜索，大幅提升关键词搜索速度',
     integrityError: '数据库结构不完整：缺少 meta 表。建议删除此数据库文件后重新导入。',
     checkFailed: '数据库检查失败: {{error}}',
   },
@@ -67,6 +69,20 @@ export default {
           end_time: '结束时间，格式 "YYYY-MM-DD HH:mm"，如 "2024-03-15 18:30"。指定后会覆盖 year/month/day/hour 参数',
         },
       },
+      deep_search_messages: {
+        desc: '精确子串匹配搜索聊天记录，速度较慢但不会遗漏任何包含关键词的消息。当普通搜索（search_messages）结果不足、需要搜索部分词或单个字符时使用。',
+        params: {
+          keywords: '搜索关键词列表，使用子串匹配（LIKE），任一关键词匹配即返回',
+          sender_id: '发送者的成员 ID，用于筛选特定成员发送的消息',
+          limit: '返回消息数量限制，默认 1000，最大 50000',
+          year: '筛选指定年份的消息，如 2024',
+          month: '筛选指定月份的消息（1-12），需要配合 year 使用',
+          day: '筛选指定日期的消息（1-31），需要配合 year 和 month 使用',
+          hour: '筛选指定小时的消息（0-23），需要配合 year、month 和 day 使用',
+          start_time: '开始时间，格式 "YYYY-MM-DD HH:mm"。指定后会覆盖 year/month/day/hour 参数',
+          end_time: '结束时间，格式 "YYYY-MM-DD HH:mm"。指定后会覆盖 year/month/day/hour 参数',
+        },
+      },
       get_recent_messages: {
         desc: '获取指定时间段内的群聊消息。适用于回答"最近大家聊了什么"、"X月群里聊了什么"等概览性问题。支持精确到分钟级别的时间查询。',
         params: {
diff --git a/electron/main/i18n/locales/zh-TW.ts b/electron/main/i18n/locales/zh-TW.ts
index 6146c9a..aa971e7 100644
--- a/electron/main/i18n/locales/zh-TW.ts
+++ b/electron/main/i18n/locales/zh-TW.ts
@@ -41,6 +41,8 @@ export default {
     migrationV2Message: '支援成員角色、訊息回覆關係和回覆內容預覽',
     migrationV3Desc: '新增會話索引相關表（chat_session、message_context）和 session_gap_threshold 欄位',
     migrationV3Message: '支援會話時間軸瀏覽和 AI 增強分析功能',
+    migrationV4Desc: '建立 FTS5 全文搜尋索引（message_fts）並構建索引資料',
+    migrationV4Message: '支援全文搜尋，大幅提升關鍵字搜尋速度',
     integrityError: '資料庫結構不完整：缺少 meta 表。建議刪除此資料庫檔案後重新匯入。',
     checkFailed: '資料庫檢查失敗：{{error}}',
   },
@@ -67,6 +69,20 @@ export default {
           end_time: '結束時間，格式 "YYYY-MM-DD HH:mm"，如 "2024-03-15 18:30"。指定後會覆蓋 year/month/day/hour 參數',
         },
       },
+      deep_search_messages: {
+        desc: '精確子串匹配搜尋聊天紀錄，速度較慢但不會遺漏任何包含關鍵詞的訊息。當普通搜尋（search_messages）結果不足、需要搜尋部分詞或單個字元時使用。',
+        params: {
+          keywords: '搜尋關鍵詞清單，使用子串匹配（LIKE），任一關鍵詞匹配即回傳',
+          sender_id: '傳送者的成員 ID，用於篩選特定成員傳送的訊息',
+          limit: '回傳訊息數量限制，預設 1000，最大 50000',
+          year: '篩選指定年份的訊息，如 2024',
+          month: '篩選指定月份的訊息（1-12），需要配合 year 使用',
+          day: '篩選指定日期的訊息（1-31），需要配合 year 和 month 使用',
+          hour: '篩選指定小時的訊息（0-23），需要配合 year、month 和 day 使用',
+          start_time: '開始時間，格式 "YYYY-MM-DD HH:mm"。指定後會覆蓋 year/month/day/hour 參數',
+          end_time: '結束時間，格式 "YYYY-MM-DD HH:mm"。指定後會覆蓋 year/month/day/hour 參數',
+        },
+      },
       get_recent_messages: {
         desc: '取得指定時間段內的群聊訊息。適用於回答「最近大家聊了什麼」、「X月群裡聊了什麼」等概覽性問題。支援精確到分鐘級別的時間查詢。',
         params: {
diff --git a/electron/main/nlp/ftsTokenizer.ts b/electron/main/nlp/ftsTokenizer.ts
new file mode 100644
index 0000000..159fe49
--- /dev/null
+++ b/electron/main/nlp/ftsTokenizer.ts
@@ -0,0 +1,112 @@
+/**
+ * FTS5 专用分词器
+ *
+ * 与 NLP 词频分析不同，FTS 分词不做词性过滤和停用词过滤，
+ * 搜索场景需要保留所有词以保证召回率。
+ *
+ * 使用 jieba 处理中文（天然兼容中英混合文本），
+ * Intl.Segmenter 处理纯英文/日文。
+ */
+
+interface JiebaInstance {
+  cut: (text: string, hmm?: boolean) => string[]
+}
+
+let jiebaInstance: JiebaInstance | null = null
+
+function getJieba(): JiebaInstance {
+  if (!jiebaInstance) {
+    try {
+      // eslint-disable-next-line @typescript-eslint/no-require-imports
+      const { Jieba } = require('@node-rs/jieba')
+      // eslint-disable-next-line @typescript-eslint/no-require-imports
+      const { dict } = require('@node-rs/jieba/dict')
+      jiebaInstance = Jieba.withDict(dict)
+    } catch (error) {
+      console.error('[FTS] Failed to load jieba module:', error)
+      throw new Error('jieba 模块加载失败')
+    }
+  }
+  return jiebaInstance!
+}
+
+/**
+ * 对文本进行 FTS 分词，返回空格分隔的 token 字符串。
+ * 用于写入 FTS5 索引。
+ */
+export function tokenizeForFts(text: string | null | undefined): string {
+  if (!text || text.trim().length === 0) return ''
+
+  try {
+    const jieba = getJieba()
+    const tokens = jieba.cut(text, false)
+    return tokens
+      .map((t) => t.trim().toLowerCase())
+      .filter((t) => t.length > 0)
+      .join(' ')
+  } catch {
+    return fallbackTokenize(text)
+  }
+}
+
+/**
+ * 降级分词：jieba 不可用时按字符/空格切分
+ */
+function fallbackTokenize(text: string): string {
+  return text
+    .toLowerCase()
+    .split(/\s+/)
+    .filter((t) => t.length > 0)
+    .join(' ')
+}
+
+/**
+ * 转义单个 token 使其在 FTS5 MATCH 中安全使用。
+ * 用双引号包裹并转义内部双引号。
+ */
+function escapeToken(token: string): string {
+  return `"${token.replace(/"/g, '""')}"`
+}
+
+/**
+ * 将用户搜索关键词列表转换为 FTS5 MATCH 表达式。
+ *
+ * 语义：
+ * - 单个关键词先分词，分词结果之间为 AND（都要出现）
+ * - 多个关键词之间为 OR（任一匹配即可）
+ *
+ * 示例：
+ * - ["今天开心"]          -> '"今天" "开心"'             (AND)
+ * - ["今天开心", "难过"]  -> '("今天" "开心") OR "难过"' (OR)
+ * - ["hello"]             -> '"hello"'
+ */
+export function tokenizeQueryForFts(keywords: string[]): string {
+  if (keywords.length === 0) return ''
+
+  const groups = keywords
+    .map((kw) => {
+      const trimmed = kw.trim()
+      if (!trimmed) return ''
+
+      try {
+        const jieba = getJieba()
+        const tokens = jieba
+          .cut(trimmed, false)
+          .map((t) => t.trim().toLowerCase())
+          .filter((t) => t.length > 0)
+
+        if (tokens.length === 0) return ''
+        if (tokens.length === 1) return escapeToken(tokens[0])
+        return tokens.map(escapeToken).join(' ')
+      } catch {
+        const simple = trimmed.toLowerCase().trim()
+        return simple ? escapeToken(simple) : ''
+      }
+    })
+    .filter((g) => g.length > 0)
+
+  if (groups.length === 0) return ''
+  if (groups.length === 1) return groups[0]
+
+  return groups.map((g) => (g.includes(' ') ? `(${g})` : g)).join(' OR ')
+}
diff --git a/electron/main/worker/dbWorker.ts b/electron/main/worker/dbWorker.ts
index eeb7f08..3fd3fb0 100644
--- a/electron/main/worker/dbWorker.ts
+++ b/electron/main/worker/dbWorker.ts
@@ -32,6 +32,7 @@ import {
   getLaughAnalysis,
   getClusterGraph,
   searchMessages,
+  deepSearchMessages,
   getMessageContext,
   getRecentMessages,
   getAllRecentMessages,
@@ -160,6 +161,9 @@ const syncHandlers: Record<string, (payload: any) => any> = {
   getWordFrequency: (p) => getWordFrequency(p),
   segmentText: (p) => segmentText(p.text, p.locale, p.minLength),
   getPosTags: () => getPosTags(),
+
+  // 深度搜索（LIKE 子串匹配）
+  deepSearchMessages: (p) => deepSearchMessages(p.sessionId, p.keywords, p.filter, p.limit, p.offset, p.senderId),
 }
 
 // 异步消息处理器（流式操作）
diff --git a/electron/main/worker/import/incrementalImport.ts b/electron/main/worker/import/incrementalImport.ts
index dc25f91..a4a7e6c 100644
--- a/electron/main/worker/import/incrementalImport.ts
+++ b/electron/main/worker/import/incrementalImport.ts
@@ -243,6 +243,16 @@ export async function incrementalImport(
     // 更新 imported_at 时间
     db.prepare('UPDATE meta SET imported_at = ?').run(Math.floor(Date.now() / 1000))
 
+    // 重建 FTS5 索引（增量导入后需要重建以包含新消息）
+    if (newMessageCount > 0) {
+      try {
+        const { rebuildFtsIndex } = await import('../query/fts')
+        rebuildFtsIndex(sessionId)
+      } catch {
+        // FTS 重建失败不影响导入流程
+      }
+    }
+
     // 写入概览统计缓存文件
     try {
       const { computeAndSetOverviewCache } = await import('../../database/sessionCache')
diff --git a/electron/main/worker/import/streamImport.ts b/electron/main/worker/import/streamImport.ts
index 60283b3..89130a2 100644
--- a/electron/main/worker/import/streamImport.ts
+++ b/electron/main/worker/import/streamImport.ts
@@ -703,6 +703,44 @@ async function streamImportSingle(
     createIndexes(db)
     logPerf('Indexes created', totalMessageCount)
 
+    // 构建 FTS5 全文搜索索引
+    try {
+      const { createFtsTable } = await import('../query/fts')
+      const { tokenizeForFts } = await import('../../nlp/ftsTokenizer')
+
+      createFtsTable(db)
+
+      const FTS_BATCH = 5000
+      const insertFts = db.prepare('INSERT INTO message_fts(rowid, content) VALUES (?, ?)')
+      let ftsOffset = 0
+
+      while (true) {
+        const rows = db
+          .prepare(
+            `SELECT id, content FROM message
+             WHERE type = 0 AND content IS NOT NULL AND content != ''
+             ORDER BY id ASC LIMIT ? OFFSET ?`
+          )
+          .all(FTS_BATCH, ftsOffset) as Array<{ id: number; content: string }>
+
+        if (rows.length === 0) break
+
+        const batch = db.transaction(() => {
+          for (const row of rows) {
+            const tokens = tokenizeForFts(row.content)
+            if (tokens) insertFts.run(row.id, tokens)
+          }
+        })
+        batch()
+
+        ftsOffset += FTS_BATCH
+      }
+
+      logPerf('FTS index built', totalMessageCount)
+    } catch (ftsError) {
+      logError('FTS index build failed (non-fatal)', ftsError instanceof Error ? ftsError : undefined)
+    }
+
     // 最终 WAL checkpoint
     sendProgress(requestId, {
       stage: 'importing',
diff --git a/electron/main/worker/import/utils.ts b/electron/main/worker/import/utils.ts
index c75cb3f..3ee8075 100644
--- a/electron/main/worker/import/utils.ts
+++ b/electron/main/worker/import/utils.ts
@@ -64,7 +64,7 @@ export function createDatabaseWithoutIndexes(sessionId: string): Database.Databa
       group_id TEXT,
       group_avatar TEXT,
       owner_id TEXT,
-      schema_version INTEGER DEFAULT 3,
+      schema_version INTEGER DEFAULT 4,
       session_gap_threshold INTEGER
     );
 
diff --git a/electron/main/worker/query/fts.ts b/electron/main/worker/query/fts.ts
new file mode 100644
index 0000000..b9fc886
--- /dev/null
+++ b/electron/main/worker/query/fts.ts
@@ -0,0 +1,215 @@
+/**
+ * FTS5 全文搜索索引管理模块
+ *
+ * 使用 Contentless FTS5 虚拟表为 message.content 提供倒排索引。
+ * 通过应用层预分词（jieba + Intl.Segmenter），空格分隔后存入 FTS5。
+ *
+ * DELETE 策略：不在 FTS 做同步删除，依赖 JOIN message 自然过滤。
+ * 提供 rebuildFtsIndex() 用于手动清理无效条目。
+ */
+
+import Database from 'better-sqlite3'
+import { getDbPath, openDatabase } from '../core'
+import { tokenizeForFts, tokenizeQueryForFts } from '../../nlp/ftsTokenizer'
+
+const BATCH_SIZE = 5000
+
+const CREATE_FTS_TABLE_SQL = `
+  CREATE VIRTUAL TABLE IF NOT EXISTS message_fts USING fts5(
+    content,
+    content='',
+    content_rowid=id
+  )
+`
+
+/**
+ * 打开可写数据库（FTS 写入专用，不使用缓存池）
+ */
+function openWritableDb(sessionId: string): Database.Database | null {
+  const dbPath = getDbPath(sessionId)
+  try {
+    const db = new Database(dbPath)
+    db.pragma('journal_mode = WAL')
+    return db
+  } catch {
+    return null
+  }
+}
+
+/**
+ * 检查数据库是否已有 FTS 索引
+ */
+export function hasFtsIndex(sessionId: string): boolean {
+  const db = openDatabase(sessionId)
+  if (!db) return false
+  try {
+    const row = db
+      .prepare("SELECT name FROM sqlite_master WHERE type='table' AND name='message_fts'")
+      .get() as { name: string } | undefined
+    return !!row
+  } catch {
+    return false
+  }
+}
+
+/**
+ * 创建 FTS 虚拟表（如果不存在）
+ */
+export function createFtsTable(db: Database.Database): void {
+  db.exec(CREATE_FTS_TABLE_SQL)
+}
+
+/**
+ * 批量构建 FTS 索引（分批处理，每批 BATCH_SIZE 条）
+ * 用于迁移或首次导入后的全量构建。
+ */
+export function buildFtsIndex(sessionId: string): { indexed: number } {
+  const db = openWritableDb(sessionId)
+  if (!db) return { indexed: 0 }
+
+  try {
+    createFtsTable(db)
+
+    const insertFts = db.prepare('INSERT INTO message_fts(rowid, content) VALUES (?, ?)')
+
+    const countRow = db.prepare(
+      "SELECT COUNT(*) as total FROM message WHERE type = 0 AND content IS NOT NULL AND content != ''"
+    ).get() as { total: number }
+    const total = countRow.total
+
+    let indexed = 0
+    let offset = 0
+
+    while (offset < total) {
+      const rows = db
+        .prepare(
+          `SELECT id, content FROM message
+           WHERE type = 0 AND content IS NOT NULL AND content != ''
+           ORDER BY id ASC
+           LIMIT ? OFFSET ?`
+        )
+        .all(BATCH_SIZE, offset) as Array<{ id: number; content: string }>
+
+      if (rows.length === 0) break
+
+      const batchInsert = db.transaction(() => {
+        for (const row of rows) {
+          const tokens = tokenizeForFts(row.content)
+          if (tokens) {
+            insertFts.run(row.id, tokens)
+          }
+        }
+      })
+      batchInsert()
+
+      indexed += rows.length
+      offset += BATCH_SIZE
+    }
+
+    return { indexed }
+  } finally {
+    db.close()
+  }
+}
+
+/**
+ * 重建 FTS 索引（清空后重新构建）
+ * 用于清理无效条目（如删除成员后）或修复索引
+ */
+export function rebuildFtsIndex(sessionId: string): { indexed: number } {
+  const db = openWritableDb(sessionId)
+  if (!db) return { indexed: 0 }
+
+  try {
+    const hasTable = db
+      .prepare("SELECT name FROM sqlite_master WHERE type='table' AND name='message_fts'")
+      .get()
+
+    if (hasTable) {
+      db.exec('DROP TABLE message_fts')
+    }
+
+    db.close()
+
+    return buildFtsIndex(sessionId)
+  } catch {
+    db.close()
+    return { indexed: 0 }
+  }
+}
+
+/**
+ * 批量写入 FTS 条目
+ * 用于增量导入时同步写入
+ */
+export function insertFtsEntries(
+  sessionId: string,
+  entries: Array<{ id: number; content: string | null }>
+): void {
+  const db = openWritableDb(sessionId)
+  if (!db) return
+
+  try {
+    const hasTable = db
+      .prepare("SELECT name FROM sqlite_master WHERE type='table' AND name='message_fts'")
+      .get()
+    if (!hasTable) {
+      db.close()
+      return
+    }
+
+    const insertFts = db.prepare('INSERT INTO message_fts(rowid, content) VALUES (?, ?)')
+
+    const batchInsert = db.transaction(() => {
+      for (const entry of entries) {
+        if (entry.content) {
+          const tokens = tokenizeForFts(entry.content)
+          if (tokens) {
+            insertFts.run(entry.id, tokens)
+          }
+        }
+      }
+    })
+    batchInsert()
+  } finally {
+    db.close()
+  }
+}
+
+/**
+ * 通过 FTS5 搜索消息，返回匹配的 message rowids
+ */
+export function searchByFts(
+  sessionId: string,
+  keywords: string[],
+  limit: number = 1000,
+  offset: number = 0
+): { rowids: number[]; total: number } {
+  if (keywords.length === 0) return { rowids: [], total: 0 }
+
+  const db = openDatabase(sessionId)
+  if (!db) return { rowids: [], total: 0 }
+
+  const matchQuery = tokenizeQueryForFts(keywords)
+  if (!matchQuery) return { rowids: [], total: 0 }
+
+  try {
+    const countRow = db
+      .prepare('SELECT COUNT(*) as total FROM message_fts WHERE content MATCH ?')
+      .get(matchQuery) as { total: number }
+
+    const rows = db
+      .prepare(
+        `SELECT rowid FROM message_fts WHERE content MATCH ? ORDER BY rank LIMIT ? OFFSET ?`
+      )
+      .all(matchQuery, limit, offset) as Array<{ rowid: number }>
+
+    return {
+      rowids: rows.map((r) => r.rowid),
+      total: countRow.total,
+    }
+  } catch (error) {
+    console.error('[FTS] Search failed, query:', matchQuery, error)
+    return { rowids: [], total: 0 }
+  }
+}
diff --git a/electron/main/worker/query/index.ts b/electron/main/worker/query/index.ts
index 4d58deb..97aedec 100644
--- a/electron/main/worker/query/index.ts
+++ b/electron/main/worker/query/index.ts
@@ -44,6 +44,7 @@ export type { ClusterGraphData, ClusterGraphNode, ClusterGraphLink, ClusterGraph
 // 聊天记录查询
 export {
   searchMessages,
+  deepSearchMessages,
   getMessageContext,
   getRecentMessages,
   getAllRecentMessages,
@@ -94,3 +95,6 @@ export type {
 
 // NLP 查询
 export { getWordFrequency, segmentText, getPosTags } from './nlp'
+
+// FTS 索引管理
+export { hasFtsIndex, buildFtsIndex, rebuildFtsIndex } from './fts'
diff --git a/electron/main/worker/query/messages.ts b/electron/main/worker/query/messages.ts
index 96c9614..49c9861 100644
--- a/electron/main/worker/query/messages.ts
+++ b/electron/main/worker/query/messages.ts
@@ -6,6 +6,8 @@
 
 import { openDatabase, buildTimeFilter, type TimeFilter } from '../core'
 import { ensureAvatarColumn } from './basic'
+import { hasFtsIndex } from './fts'
+import { tokenizeQueryForFts } from '../../nlp/ftsTokenizer'
 
 // ==================== 类型定义 ====================
 
@@ -275,22 +277,110 @@ export function searchMessages(
   const db = openDatabase(sessionId)
   if (!db) return { messages: [], total: 0 }
 
-  // 构建关键词条件（OR 逻辑）
-  let keywordCondition = '1=1' // 默认条件（始终为真）
+  const useFts = keywords.length > 0 && hasFtsIndex(sessionId)
+  let matchQuery = ''
+  if (useFts) {
+    matchQuery = tokenizeQueryForFts(keywords)
+  }
+
+  // FTS5 路径：使用倒排索引加速搜索
+  if (useFts && matchQuery) {
+    return searchMessagesWithFts(db, sessionId, matchQuery, filter, limit, offset, senderId)
+  }
+
+  // LIKE 路径（fallback）：旧数据库无 FTS 索引或无关键词
+  return searchMessagesWithLike(db, keywords, filter, limit, offset, senderId)
+}
+
+/**
+ * FTS5 搜索路径
+ */
+function searchMessagesWithFts(
+  db: ReturnType<typeof openDatabase> & object,
+  _sessionId: string,
+  matchQuery: string,
+  filter?: TimeFilter,
+  limit: number = 20,
+  offset: number = 0,
+  senderId?: number
+): MessagesWithTotal {
+  const { clause: timeClause, params: timeParams } = buildTimeFilter(filter, 'msg')
+  const timeCondition = timeClause ? timeClause.replace('WHERE', 'AND') : ''
+  const { condition: senderCondition, params: senderParams } = buildSenderCondition(senderId)
+
+  try {
+    const countSql = `
+      SELECT COUNT(*) as total
+      FROM message msg
+      JOIN member m ON msg.sender_id = m.id
+      WHERE msg.id IN (SELECT rowid FROM message_fts WHERE content MATCH ?)
+      ${timeCondition}
+      ${senderCondition}
+    `
+    const totalRow = db.prepare(countSql).get(matchQuery, ...timeParams, ...senderParams) as { total: number }
+    const total = totalRow?.total || 0
+
+    const sql = `
+      SELECT
+        msg.id,
+        m.id as senderId,
+        COALESCE(m.group_nickname, m.account_name, m.platform_id) as senderName,
+        m.platform_id as senderPlatformId,
+        m.aliases,
+        m.avatar,
+        msg.content,
+        msg.ts as timestamp,
+        msg.type,
+        msg.reply_to_message_id,
+        reply_msg.content as replyToContent,
+        COALESCE(reply_m.group_nickname, reply_m.account_name, reply_m.platform_id) as replyToSenderName
+      FROM message msg
+      JOIN member m ON msg.sender_id = m.id
+      LEFT JOIN message reply_msg ON msg.reply_to_message_id = reply_msg.platform_message_id
+      LEFT JOIN member reply_m ON reply_msg.sender_id = reply_m.id
+      WHERE msg.id IN (SELECT rowid FROM message_fts WHERE content MATCH ?)
+      ${timeCondition}
+      ${senderCondition}
+      ORDER BY msg.ts DESC
+      LIMIT ? OFFSET ?
+    `
+
+    const rows = db
+      .prepare(sql)
+      .all(matchQuery, ...timeParams, ...senderParams, limit, offset) as DbMessageRow[]
+
+    return {
+      messages: rows.map(sanitizeMessageRow),
+      total,
+    }
+  } catch (error) {
+    console.error('[FTS] searchMessages FTS path failed, falling back to LIKE:', error)
+    return searchMessagesWithLike(db, [], filter, limit, offset, senderId)
+  }
+}
+
+/**
+ * LIKE 搜索路径（fallback 或 deep_search 使用）
+ */
+export function searchMessagesWithLike(
+  db: ReturnType<typeof openDatabase> & object,
+  keywords: string[],
+  filter?: TimeFilter,
+  limit: number = 20,
+  offset: number = 0,
+  senderId?: number
+): MessagesWithTotal {
+  let keywordCondition = '1=1'
   const keywordParams: string[] = []
   if (keywords.length > 0) {
     keywordCondition = `(${keywords.map(() => `msg.content LIKE ?`).join(' OR ')})`
     keywordParams.push(...keywords.map((k) => `%${k}%`))
   }
 
-  // 构建时间过滤条件（使用 'msg' 表别名避免多表 JOIN 时的列名歧义）
   const { clause: timeClause, params: timeParams } = buildTimeFilter(filter, 'msg')
   const timeCondition = timeClause ? timeClause.replace('WHERE', 'AND') : ''
-
-  // 构建发送者筛选条件
   const { condition: senderCondition, params: senderParams } = buildSenderCondition(senderId)
 
-  // 查询总数
   const countSql = `
     SELECT COUNT(*) as total
     FROM message msg
@@ -302,7 +392,6 @@ export function searchMessages(
   const totalRow = db.prepare(countSql).get(...keywordParams, ...timeParams, ...senderParams) as { total: number }
   const total = totalRow?.total || 0
 
-  // 查询消息
   const sql = `
     SELECT
       msg.id,
@@ -336,6 +425,24 @@ export function searchMessages(
   }
 }
 
+/**
+ * 深度搜索消息（LIKE 子串匹配，速度较慢但不会遗漏）
+ * 始终使用 LIKE 路径，不经过 FTS5。
+ */
+export function deepSearchMessages(
+  sessionId: string,
+  keywords: string[],
+  filter?: TimeFilter,
+  limit: number = 20,
+  offset: number = 0,
+  senderId?: number
+): MessagesWithTotal {
+  ensureAvatarColumn(sessionId)
+  const db = openDatabase(sessionId)
+  if (!db) return { messages: [], total: 0 }
+  return searchMessagesWithLike(db, keywords, filter, limit, offset, senderId)
+}
+
 /**
  * 获取消息上下文（指定消息前后的消息）
  * 使用消息 ID 方式获取精确的前后 N 条消息
diff --git a/electron/main/worker/query/session/aiTools.ts b/electron/main/worker/query/session/aiTools.ts
index f55bfe9..649eab7 100644
--- a/electron/main/worker/query/session/aiTools.ts
+++ b/electron/main/worker/query/session/aiTools.ts
@@ -5,6 +5,8 @@
 
 import { openReadonlyDatabase } from './core'
 import type { SessionSearchResultItem, SessionMessagesResult } from './types'
+import { hasFtsIndex } from '../fts'
+import { tokenizeQueryForFts } from '../../../nlp/ftsTokenizer'
 
 /**
  * 搜索会话（用于 AI 工具）
@@ -50,17 +52,31 @@ export function searchSessions(
 
     // 关键词过滤：只返回包含关键词的会话
     if (keywords && keywords.length > 0) {
-      const keywordConditions = keywords.map(() => `m.content LIKE ?`).join(' OR ')
-      sessionSql += `
-        AND cs.id IN (
-          SELECT DISTINCT mc.session_id
-          FROM message_context mc
-          JOIN message m ON m.id = mc.message_id
-          WHERE (${keywordConditions})
-        )
-      `
-      for (const kw of keywords) {
-        params.push(`%${kw}%`)
+      const useFts = hasFtsIndex(sessionId)
+      const matchQuery = useFts ? tokenizeQueryForFts(keywords) : ''
+
+      if (useFts && matchQuery) {
+        sessionSql += `
+          AND cs.id IN (
+            SELECT DISTINCT mc.session_id
+            FROM message_context mc
+            WHERE mc.message_id IN (SELECT rowid FROM message_fts WHERE content MATCH ?)
+          )
+        `
+        params.push(matchQuery)
+      } else {
+        const keywordConditions = keywords.map(() => `m.content LIKE ?`).join(' OR ')
+        sessionSql += `
+          AND cs.id IN (
+            SELECT DISTINCT mc.session_id
+            FROM message_context mc
+            JOIN message m ON m.id = mc.message_id
+            WHERE (${keywordConditions})
+          )
+        `
+        for (const kw of keywords) {
+          params.push(`%${kw}%`)
+        }
       }
     }
 
diff --git a/electron/main/worker/workerManager.ts b/electron/main/worker/workerManager.ts
index c84e349..0c0ad22 100644
--- a/electron/main/worker/workerManager.ts
+++ b/electron/main/worker/workerManager.ts
@@ -468,6 +468,20 @@ export async function searchMessages(
   return sendToWorker('searchMessages', { sessionId, keywords, filter, limit, offset, senderId })
 }
 
+/**
+ * 深度搜索消息（LIKE 子串匹配，速度较慢但不会遗漏）
+ */
+export async function deepSearchMessages(
+  sessionId: string,
+  keywords: string[],
+  filter?: any,
+  limit?: number,
+  offset?: number,
+  senderId?: number
+): Promise<{ messages: SearchMessageResult[]; total: number }> {
+  return sendToWorker('deepSearchMessages', { sessionId, keywords, filter, limit, offset, senderId })
+}
+
 /**
  * 获取消息上下文
  * 支持单个或批量消息 ID，返回合并去重后的上下文消息
diff --git a/src/i18n/locales/en-US/ai.json b/src/i18n/locales/en-US/ai.json
index cd4d428..773f285 100644
--- a/src/i18n/locales/en-US/ai.json
+++ b/src/i18n/locales/en-US/ai.json
@@ -49,6 +49,7 @@
       "tools": {
         "get_chat_overview": "Get Chat Overview",
         "search_messages": "Search Messages",
+        "deep_search_messages": "Deep Search Messages",
         "get_recent_messages": "Get Recent Messages",
         "get_member_stats": "Get Member Stats",
         "get_time_stats": "Get Time Stats",
@@ -310,6 +311,7 @@
     "builtinToolDesc": {
       "get_chat_overview": "Get chat overview information",
       "search_messages": "Search chat messages",
+      "deep_search_messages": "Deep search chat messages",
       "get_recent_messages": "Get recent messages",
       "get_message_context": "Get message context",
       "get_conversation_between": "Get conversation between two members",
diff --git a/src/i18n/locales/ja-JP/ai.json b/src/i18n/locales/ja-JP/ai.json
index eb4798a..cf23fd6 100644
--- a/src/i18n/locales/ja-JP/ai.json
+++ b/src/i18n/locales/ja-JP/ai.json
@@ -49,6 +49,7 @@
       "tools": {
         "get_chat_overview": "チャット概要を取得",
         "search_messages": "チャット履歴を検索",
+        "deep_search_messages": "チャット履歴を深層検索",
         "get_recent_messages": "最近のメッセージを取得",
         "get_member_stats": "メンバー統計を取得",
         "get_time_stats": "時間分布を取得",
@@ -310,6 +311,7 @@
     "builtinToolDesc": {
       "get_chat_overview": "チャット概要情報を取得",
       "search_messages": "チャットメッセージを検索",
+      "deep_search_messages": "チャットメッセージを深層検索",
       "get_recent_messages": "最近のメッセージを取得",
       "get_message_context": "メッセージのコンテキストを取得",
       "get_conversation_between": "2人のやり取りを取得",
diff --git a/src/i18n/locales/zh-CN/ai.json b/src/i18n/locales/zh-CN/ai.json
index 3138fcc..97c2922 100644
--- a/src/i18n/locales/zh-CN/ai.json
+++ b/src/i18n/locales/zh-CN/ai.json
@@ -49,6 +49,7 @@
       "tools": {
         "get_chat_overview": "获取聊天概览",
         "search_messages": "搜索聊天记录",
+        "deep_search_messages": "深度搜索聊天记录",
         "get_recent_messages": "获取最近消息",
         "get_member_stats": "获取成员统计",
         "get_time_stats": "获取时间分布",
@@ -310,6 +311,7 @@
     "builtinToolDesc": {
       "get_chat_overview": "获取聊天概览信息",
       "search_messages": "搜索聊天消息",
+      "deep_search_messages": "深度搜索聊天消息",
       "get_recent_messages": "获取最近消息",
       "get_message_context": "获取消息上下文",
       "get_conversation_between": "获取两人对话",
diff --git a/src/i18n/locales/zh-TW/ai.json b/src/i18n/locales/zh-TW/ai.json
index 39ad636..e598061 100644
--- a/src/i18n/locales/zh-TW/ai.json
+++ b/src/i18n/locales/zh-TW/ai.json
@@ -49,6 +49,7 @@
       "tools": {
         "get_chat_overview": "取得聊天概覽",
         "search_messages": "搜尋聊天紀錄",
+        "deep_search_messages": "深度搜尋聊天紀錄",
         "get_recent_messages": "取得最近訊息",
         "get_member_stats": "取得成員統計",
         "get_time_stats": "取得時間分佈",
@@ -310,6 +311,7 @@
     "builtinToolDesc": {
       "get_chat_overview": "取得聊天概覽資訊",
       "search_messages": "搜尋聊天訊息",
+      "deep_search_messages": "深度搜尋聊天訊息",
       "get_recent_messages": "取得最近訊息",
       "get_message_context": "取得訊息上下文",
       "get_conversation_between": "取得兩人對話",