feat(ai): 新增聊天助手、智能检索与关注提醒，修复联系人加载及导出失败

.github/workflows/ai-cross-platform.yml

@@ -0,0 +1,70 @@
+name: AI Cross-Platform Checks
+
+on:
+  workflow_dispatch:
+  pull_request:
+    paths:
+      - '.github/workflows/ai-cross-platform.yml'
+      - 'src/wechat_decrypt_tool/ai/**'
+      - 'src/wechat_decrypt_tool/local_search/**'
+      - 'src/wechat_decrypt_tool/routers/ai*.py'
+      - 'src/wechat_decrypt_tool/routers/local_search.py'
+      - 'src/wechat_decrypt_tool/resources/local_search*.json'
+      - 'desktop/scripts/ai-packaging.cjs'
+      - 'desktop/scripts/build-backend.cjs'
+      - 'desktop/src/ai-notifications.cjs'
+      - 'desktop/src/renderer-startup.cjs'
+      - 'desktop/src/main.cjs'
+      - 'desktop/tests/ai-*.test.cjs'
+      - 'desktop/tests/renderer-startup.test.cjs'
+      - 'tools/verify_ai_runtime.py'
+      - 'tools/build_ai_smoke.cjs'
+      - 'frontend/components/**'
+      - 'frontend/composables/useAiApi.js'
+      - 'tests/test_ai*.py'
+      - 'tests/test_local_search*.py'
+      - 'pyproject.toml'
+      - 'uv.lock'
+
+permissions:
+  contents: read
+
+jobs:
+  ai:
+    strategy:
+      fail-fast: false
+      matrix:
+        os: [windows-2022, macos-14]
+    runs-on: ${{ matrix.os }}
+    timeout-minutes: 30
+    env:
+      # macOS setup-python builds may lack SQLite loadable-extension support.
+      # Use uv-managed CPython for both source checks and the frozen runtime.
+      UV_MANAGED_PYTHON: 'true'
+    steps:
+      - uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - uses: actions/setup-python@a26af69be951a213d495a4c3e4e4022e16d87065 # v5
+        with:
+          python-version: '3.11'
+      - uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
+        with:
+          node-version: 22
+      - run: python -m pip install uv
+      - name: Install Python with SQLite extension support
+        run: uv python install 3.11
+      - run: uv sync --locked --extra build
+      - name: AI runtime and media (synthetic, no API calls)
+        run: uv run python tools/verify_ai_runtime.py
+      - name: AI and local search regression
+        shell: bash
+        run: uv run pytest -q tests/test_ai*.py tests/test_local_search*.py
+      - run: npm ci
+        working-directory: frontend
+      - run: npx vitest run
+        working-directory: frontend
+      - run: npm run generate
+        working-directory: frontend
+      - name: Desktop notification and packaging contracts
+        run: node --test desktop/tests/ai-notifications.test.cjs desktop/tests/ai-packaging.test.cjs desktop/tests/renderer-startup.test.cjs
+      - name: Frozen AI runtime (no account or API credentials)
+        run: node tools/build_ai_smoke.cjs

.github/workflows/installer-unicode.yml

@@ -0,0 +1,58 @@
+name: Windows Installer Unicode
+
+on:
+  workflow_dispatch:
+  pull_request:
+    paths:
+      - 'desktop/**'
+      - '.github/workflows/installer-unicode.yml'
+      - '.github/workflows/release.yml'
+      - '.gitattributes'
+  push:
+    branches: [main]
+    paths:
+      - 'desktop/**'
+      - '.github/workflows/installer-unicode.yml'
+      - '.github/workflows/release.yml'
+      - '.gitattributes'
+
+permissions:
+  contents: read
+
+concurrency:
+  # 不同提交分别完成回归检查，避免连续推送让旧提交留下 Cancelled 状态。
+  # 同一提交的重复触发仍可取消，减少重复运行。
+  group: installer-unicode-${{ github.workflow }}-${{ github.ref }}-${{ github.sha }}
+  cancel-in-progress: true
+
+jobs:
+  unicode-paths:
+    name: Windows installer Unicode paths
+    runs-on: windows-2022
+    timeout-minutes: 15
+    steps:
+      - name: Checkout
+        uses: actions/checkout@11d5960a326750d5838078e36cf38b85af677262 # v4
+      - name: Setup Node
+        uses: actions/setup-node@49933ea5288caeca8642d1e84afbd3f7d6820020 # v4
+        with:
+          node-version: 22
+          cache: npm
+          cache-dependency-path: desktop/package-lock.json
+      # 仅测试安装器函数，不构建完整应用，也不需要签名或真实用户数据。
+      - name: Install locked test dependencies
+        working-directory: desktop
+        run: npm ci --ignore-scripts
+      - name: Verify Unicode install, upgrade, settings and data migration
+        working-directory: desktop
+        shell: pwsh
+        run: |
+          node --test `
+            tests/desktop-settings.test.cjs `
+            tests/output-dir.test.cjs `
+            tests/installer-output-dir.test.cjs `
+            tests/installer-safety.test.cjs `
+            tests/installer-unicode-integration.test.cjs
+          if ($LASTEXITCODE -ne 0) {
+            throw 'Windows installer Unicode regression tests failed.'
+          }

.github/workflows/release.yml

@@ -563,6 +563,11 @@ jobs:
         shell: pwsh
         run: |
           node --test `
+            tests/desktop-settings.test.cjs `
+            tests/output-dir.test.cjs `
+            tests/installer-output-dir.test.cjs `
+            tests/installer-safety.test.cjs `
+            tests/installer-unicode-integration.test.cjs `
             tests/package-config.test.cjs `
             tests/native-core-before-pack.test.cjs `
             tests/native-core-packaging.test.cjs `

.gitignore

@@ -104,6 +104,17 @@ pnpm-lock.yaml
 
 # Local scratch file accidentally generated during development
 /bento-summary.html
+/tmp/local-search-validation/
+
+# 本地跨平台验收的传输包、运行目录与远程连接辅助脚本
+/tmp/macos-ai-*
+/tmp/macos-public-manifest.json
+/tmp/mac-desktop-probe.cjs
+/tmp/mac-ui-check.cjs
+/tmp/mac-user-ui.cjs
+/tmp/mac-e2e-private.json
+/tmp/ai-frozen-check/
+/tmp/ai-macos-regression-*/
 
 # 高级版演示走片的默认输出目录（website/dev/shot.mjs），只是本地调试产物
 pro-shots/

README.md

@@ -215,6 +215,12 @@ npm run dev
 - API服务(默认): http://localhost:10392 （可通过环境变量 WECHAT_TOOL_PORT 修改）
 - API文档(默认): http://localhost:10392/docs
 
+## 聊天 AI 总结与关注提醒
+
+在「设置 → AI 服务」配置模型后，点击聊天右上角 AI，可按条数或时间批量总结群聊与好友消息，并设置定时总结、消息阈值和 AI 语义关注提醒。支持图片、常见文档分析及桌面通知。详见 [使用与开发说明](docs/chat-ai.md)。
+
+可在「设置 → AI 服务 → 本地检索」按账号开启可选的语义检索，使用 Hugging Face 固定版本模型，支持 CPU 与 NVIDIA GPU 自动回退。使用方法、下载来源和兼容性实测见 [本地语义检索说明](docs/local-semantic-search.md)。
+
 ## MCP 服务
 
 设置页中的“AI 接入提示词”会包含 endpoint 和 Bearer token，可直接复制给客户端作为接入指令。

THIRD_PARTY_NOTICES.md

@@ -1,5 +1,22 @@
 # Third-Party Notices
 
+## AI provider icons
+
+AI provider SVG brand marks in `frontend/assets/icons/ai-providers/` come from
+[`@lobehub/icons-static-svg` 1.95.0](https://www.npmjs.com/package/@lobehub/icons-static-svg/v/1.95.0),
+part of [Lobe Icons](https://github.com/lobehub/lobe-icons), copyright (c) 2023 LobeHub.
+The SVG files are distributed unchanged under the MIT License. Color variants
+retain their original colors; monochrome variants use themed CSS masks. The license is included in
+`docs/licenses/lobe-icons-MIT.txt`. Brand marks identify their respective services.
+
+The PNG app icons are original assets served by the respective official websites:
+
+- Doubao: https://lf-flow-web-cdn.doubao.com/obj/flow-doubao/favicon/new-doubao/128x128.png
+- LM Studio: https://lmstudio.ai/assets/marketing/logo-192x192.png
+
+These PNG assets are not covered by the Lobe Icons MIT License. All brand names
+and trademarks belong to their respective owners and are used to identify services.
+
 ## Optional local voice transcription
 
 The optional local voice transcription feature uses the following Python packages:
@@ -31,3 +48,18 @@ The complete upstream license text is included at
 The optional Windows QQ feedback bridge reuses WeQ's `nt_helper.node` and QQ
 flash-transfer protocol implementation. WeQ is copyright H3CoF6 and licensed under
 CC BY-NC-SA 4.0: https://github.com/H3CoF6/WeQ
+
+## Optional local semantic search
+
+Local retrieval uses ONNX Runtime (MIT), Hugging Face Hub (Apache-2.0),
+Hugging Face tokenizers (Apache-2.0), and sqlite-vec (MIT / Apache-2.0 dual license).
+Embedding weights are downloaded separately from fixed Hugging Face revisions:
+BAAI BGE Small/Base Chinese converted to ONNX by Xenova, and intfloat
+multilingual E5 Small. Their selected repositories declare MIT; source,
+revision and file hashes are recorded in `local_search_models.json`.
+
+The optional NVIDIA component downloads pinned official Python wheels for CUDA
+Runtime, cuBLAS, cuDNN and cuFFT. These libraries retain their NVIDIA license
+terms and bundled notices; they are not relicensed under this repository's
+license. ONNX Runtime GPU remains MIT. See `docs/local-semantic-search.md` and
+`local_search_gpu.json` for the exact component list and validation status.