diff --git a/README.md b/README.md
index 75180640..bd31b67b 100644
--- a/README.md
+++ b/README.md
@@ -63,7 +63,7 @@ DataFoundry 0.2 turns the first usable workbench into a more complete, stateful
 
 - **Branchable, concurrent analysis** — Keep multiple sessions running, queue follow-up prompts, restore completed work, and branch from an earlier question or checkpoint without overwriting the original path.
 - **Evidence-first follow-ups** — Reference a complete output or a selected table/text region in the next question; resolved evidence is carried into the governed run context with diagnostics.
-- **Semantic trace and first-party Data Link** — Inspect checkpoint-backed run structure in a semantic Trace DAG. With our newly open-sourced [Data Link](https://github.com/datagallery-lab/datalink), connect tables and columns to business concepts, entities, joinable paths, and confidence-scored relationships for stronger agent grounding.
+- **Semantic trace and DataLink integration** — Inspect checkpoint-backed run structure in a semantic Trace DAG. Connect an external DataLink service to map tables and columns to business concepts, entities, joinable paths, and confidence-scored relationships for stronger agent grounding.
 - **Reusable outputs and workspace assets** — Preview and export tables, charts, reports, SQL, and files; upload files into an active session, then promote supported files for reuse across sessions.
 - **Production-facing Web foundation** — Built-in password authentication, same-origin API proxying, bilingual UI, model connection tests, onboarding, and an auto-provisioned DTC growth analysis case.
 
@@ -75,7 +75,7 @@ Formal mode has two paths (do **not** run `npm run dev`). Docker / Compose is no
 
 ### Recommended: Ubuntu / Debian one-click
 
-`./deploy.sh` generates configuration, installs dependencies, builds (Web, API, and TUI), and starts Web + API as a detached background process — closing the terminal does not stop the stack. The TUI is built and ready but is a foreground client: start it in another terminal with `./deploy.sh tui` (or `npm run start:tui`); it does **not** stay running with the stack. Semantic Data Link is an **external** component (not started by deploy) — connect it later via MCP in the Web UI if needed. No model key is required during deploy — create and enable a model profile in the Web UI after login. Does **not** support native Windows / macOS.
+`./deploy.sh` generates configuration, installs dependencies, builds (Web, API, and TUI), and starts Web + API as a detached background process — closing the terminal does not stop the stack. The TUI is built and ready but is a foreground client: start it in another terminal with `./deploy.sh tui` (or `npm run start:tui`); it does **not** stay running with the stack. DataLink is an **external** component (not started by deploy) — connect it later via MCP in the Web UI if needed. No model key is required during deploy — create and enable a model profile in the Web UI after login. Does **not** support native Windows / macOS.
 
 ```bash
 git clone https://github.com/datagallery-lab/datafoundry.git
@@ -154,7 +154,7 @@ You will see the full chain: schema inspection → read-only SQL → SQL audit 
 
 For real production, also configure SMTP and a reverse proxy: [`deploy/nginx.datafoundry.conf.example`](deploy/nginx.datafoundry.conf.example) (gzip/brotli static assets; keep `/api/copilotkit` uncompressed and unbuffered for SSE).
 
-> Full steps, the two-environment matrix, and optional external Data Link notes: [Quick Start](docs/en/quick-start.md). Contributor hot-reload (`npm run dev`) is appendix-only. Connect your own PostgreSQL / MySQL / CSV and more: [Data Sources guide](docs/en/guides/data-sources.md).
+> Full steps, the two-environment matrix, and optional external DataLink notes: [Quick Start](docs/en/quick-start.md). Contributor hot-reload (`npm run dev`) is appendix-only. Connect your own PostgreSQL / MySQL / CSV and more: [Data Sources guide](docs/en/guides/data-sources.md).
 
 
 ## 🆚 How It Differs From Coding Agents And SQL Chatbots
diff --git a/README_zh.md b/README_zh.md
index dc52939d..a3878b4b 100644
--- a/README_zh.md
+++ b/README_zh.md
@@ -63,7 +63,7 @@ DataFoundry 0.2 在首个可用版本上，进一步补齐了有状态、可追
 
 - **可分支的并发分析** — 多个会话可同时运行，运行中可排队后续问题；恢复历史后，可从早期问题或 checkpoint 创建新分支，不覆盖原分析路径。
 - **证据驱动的追问** — 可将完整产出，或选中的表格区域、文本片段引用到下一个问题；证据解析结果和诊断信息会进入受控 run context。
-- **语义 Trace 与自研 Data Link** — 通过基于 checkpoint 的语义 Trace DAG 复盘执行结构；结合我们新近开源的 [Data Link](https://github.com/datagallery-lab/datalink)，把表和字段连接到业务概念、实体、可 JOIN 路径和带置信度的关系，为 Agent 提供更强语义支撑。
+- **语义 Trace 与 DataLink 集成** — 通过基于 checkpoint 的语义 Trace DAG 复盘执行结构；连接外部 DataLink 服务，把表和字段映射到业务概念、实体、可 JOIN 路径和带置信度的关系，为 Agent 提供更强语义支撑。
 - **可复用的产出与工作区资产** — 统一预览和导出表格、图表、报告、SQL 和文件；文件可先上传到当前会话，再提升为跨会话复用的工作区资产。
 - **面向正式部署的 Web 基础** — 内置密码认证、同源 API 代理、中英双语界面、模型连接测试、首次引导，并自动准备 DTC 增长分析案例。
 
@@ -75,7 +75,7 @@ DataFoundry 0.2 在首个可用版本上，进一步补齐了有状态、可追
 
 ### 推荐：Ubuntu / Debian 一键部署
 
-`./deploy.sh` 自动生成配置、安装依赖、构建（含 Web、API 与 TUI），并以 detached 后台进程启动 Web + API——关闭终端一般不会停止服务。TUI 会在部署时构建就绪，但它是前台交互客户端：请另开终端执行 `./deploy.sh tui`（或 `npm run start:tui`），**不会**随 stack 后台常驻。语义 Data Link 为**外置**组件（deploy 不会安装或启动）——如需使用，稍后在 Web 的 MCP 配置中连接外部服务即可。部署时不要求填写模型 Key——登录后在 Web 中创建并启用模型即可。**不支持**原生 Windows / macOS。
+`./deploy.sh` 自动生成配置、安装依赖、构建（含 Web、API 与 TUI），并以 detached 后台进程启动 Web + API——关闭终端一般不会停止服务。TUI 会在部署时构建就绪，但它是前台交互客户端：请另开终端执行 `./deploy.sh tui`（或 `npm run start:tui`），**不会**随 stack 后台常驻。DataLink 为**外置**组件（deploy 不会安装或启动）——如需使用，稍后在 Web 的 MCP 配置中连接外部服务即可。部署时不要求填写模型 Key——登录后在 Web 中创建并启用模型即可。**不支持**原生 Windows / macOS。
 
 ```bash
 git clone https://github.com/datagallery-lab/datafoundry.git
diff --git a/docs/en/guides/datalink.md b/docs/en/guides/datalink.md
new file mode 100644
index 00000000..fb74333c
--- /dev/null
+++ b/docs/en/guides/datalink.md
@@ -0,0 +1,52 @@
+# DataLink semantic service
+
+DataLink is an optional semantic graph service for DataFoundry. It connects schemas and profiles to business concepts, entities, join paths, and confidence-scored relationships. The implementation is included under `services/datalink`, but it runs as a separate Python service.
+
+## Runtime topology
+
+| Process | Default endpoint | Purpose |
+| --- | --- | --- |
+| DataLink MCP | `http://127.0.0.1:8080/mcp` | Exposes `datalink_explore` for agent grounding |
+| DataLink REST | `http://127.0.0.1:8081` | Graph management and visualization API |
+
+DataFoundry's Web and API processes remain unchanged. DataLink is not started by `deploy.sh` or `npm run start`; start it separately when semantic grounding is needed.
+
+## Install and start
+
+From the service directory, install Python 3.10+ and [uv](https://docs.astral.sh/uv/):
+
+```bash
+cd services/datalink
+uv pip install -e .
+cp datalink_config.example.json datalink_config.json
+```
+
+Set the LLM and embedding provider in `datalink_config.json`, then start both processes in separate terminals:
+
+```bash
+uv run datalink serve --port 8080 --transport streamable-http
+uv run datalink api --port 8081
+```
+
+The service README contains provider examples and CLI commands for building and exploring a graph.
+
+## Connect it in DataFoundry
+
+In the Web workbench, open MCP settings and add an external server with:
+
+| Field | Example |
+| --- | --- |
+| `serverUrl` | `http://127.0.0.1:8080/mcp` |
+| `apiUrl` | `http://127.0.0.1:8081` |
+| `transport` | `streamable-http` |
+| `toolManifest` | `[{ "name": "datalink_explore" }]` |
+
+Use a name or id containing `datalink` so the DataLink panel recognizes the server.
+
+## Verify
+
+```bash
+curl http://127.0.0.1:8081/healthz
+```
+
+If the panel reports the service as unavailable, check both processes, ports `8080` and `8081`, the configured graph path, and the MCP transport. Keep API keys in environment variables or a secret manager; do not commit them to `datalink_config.json`.
diff --git a/docs/en/quick-start.md b/docs/en/quick-start.md
index e7485d6c..7a69d451 100644
--- a/docs/en/quick-start.md
+++ b/docs/en/quick-start.md
@@ -13,7 +13,7 @@ After deploy, configure a model in the Web UI and run an analysis against the bu
 
 - **One-click deploy**: Ubuntu or Debian (x86_64 / aarch64); Node.js 22 (the script can help install it after consent)
 - **Manual npm**: Linux, macOS, or Windows; Node.js >= 22 and npm
-- Optional external [Data Link](https://github.com/datagallery-lab/datalink): run it as a separate process if you want semantic graph features (not required for deploy)
+- Optional external [DataLink](guides/datalink.md): run it as a separate process if you want semantic graph features (not required for deploy)
 
 Install and run the project in the same environment. On Windows, do not share `node_modules` between Windows and WSL.
 
@@ -64,11 +64,11 @@ Open `http://127.0.0.1:3000/login` (or the Web URL printed by the script if the
 
 `LLM_*` is not required during deploy. Set `AUTH_PUBLIC_BASE_URL` for remote hosts. Re-running deploy uses a maintenance window: stop the managed process group before `npm ci` and builds.
 
-### External Data Link (optional)
+### External DataLink (optional)
 
-One-click deploy starts **only** Web + API. It does **not** install, start, or health-check Data Link.
+One-click deploy starts **only** Web + API. It does **not** install, start, or health-check DataLink.
 
-To use semantic features, run [Data Link](https://github.com/datagallery-lab/datalink) as a separate service (typically MCP on `:8080` and REST on `:8081`), then in the Web workbench MCP settings add an external server, for example:
+To use semantic features, run [DataLink](guides/datalink.md) as a separate service (typically MCP on `:8080` and REST on `:8081`), then in the Web workbench MCP settings add an external server, for example:
 
 | Field | Example |
 | --- | --- |
@@ -77,7 +77,7 @@ To use semantic features, run [Data Link](https://github.com/datagallery-lab/dat
 | `transport` | `streamable-http` |
 | `toolManifest` | `[{ "name": "datalink_explore" }]` |
 
-Use a name/id containing `datalink` (or `datagraph`) so the Data Link panel can recognize it.
+Use a name/id containing `datalink` so the DataLink panel can recognize it.
 
 ## Windows / macOS / other: manual npm deploy
 
diff --git a/docs/javascripts/language-switcher.js b/docs/javascripts/language-switcher.js
index 80e05c48..a81c844d 100644
--- a/docs/javascripts/language-switcher.js
+++ b/docs/javascripts/language-switcher.js
@@ -105,6 +105,7 @@
     ["Web workbench", "Web 工作台"],
     ["TUI", "TUI"],
     ["Data sources", "数据源"],
+    ["DataLink", "DataLink"],
     ["Examples", "案例"],
     ["DTC growth operating review", "DTC 增长经营复盘"],
     ["Developer Guide", "开发者指南"],
diff --git a/docs/zh/guides/datalink.md b/docs/zh/guides/datalink.md
new file mode 100644
index 00000000..853172db
--- /dev/null
+++ b/docs/zh/guides/datalink.md
@@ -0,0 +1,52 @@
+# DataLink 语义服务
+
+DataLink 是 DataFoundry 的可选语义图服务，用于将 Schema 与数据画像连接到业务概念、实体、可 JOIN 路径和带置信度的关系。实现位于 `services/datalink`，但作为独立 Python 服务运行。
+
+## 运行拓扑
+
+| 进程 | 默认地址 | 用途 |
+| --- | --- | --- |
+| DataLink MCP | `http://127.0.0.1:8080/mcp` | 提供 `datalink_explore`，为 Agent 提供语义上下文 |
+| DataLink REST | `http://127.0.0.1:8081` | 图谱管理与可视化 API |
+
+DataFoundry 的 Web 与 API 进程不受影响。`deploy.sh` 和 `npm run start` 不会启动 DataLink；需要语义增强时请单独启动。
+
+## 安装与启动
+
+进入服务目录，准备 Python 3.10+ 与 [uv](https://docs.astral.sh/uv/)：
+
+```bash
+cd services/datalink
+uv pip install -e .
+cp datalink_config.example.json datalink_config.json
+```
+
+在 `datalink_config.json` 中配置 LLM 与 embedding 提供商，然后在两个终端分别启动：
+
+```bash
+uv run datalink serve --port 8080 --transport streamable-http
+uv run datalink api --port 8081
+```
+
+提供商示例、建图和探索命令见服务目录下的 README。
+
+## 在 DataFoundry 中连接
+
+在 Web 工作台打开 MCP 设置，添加外部服务：
+
+| 字段 | 示例 |
+| --- | --- |
+| `serverUrl` | `http://127.0.0.1:8080/mcp` |
+| `apiUrl` | `http://127.0.0.1:8081` |
+| `transport` | `streamable-http` |
+| `toolManifest` | `[{ "name": "datalink_explore" }]` |
+
+名称或 id 中包含 `datalink` 时，DataLink 面板会识别该服务。
+
+## 验证
+
+```bash
+curl http://127.0.0.1:8081/healthz
+```
+
+如果面板显示服务不可用，请检查两个进程、`8080`/`8081` 端口、图数据库路径和 MCP transport。API Key 应放在环境变量或 Secret 管理系统中，不要提交到 `datalink_config.json`。
diff --git a/docs/zh/quick-start.md b/docs/zh/quick-start.md
index 26daae26..c677ff77 100644
--- a/docs/zh/quick-start.md
+++ b/docs/zh/quick-start.md
@@ -13,7 +13,7 @@
 
 - **一键部署**：Ubuntu 或 Debian（x86_64 / aarch64）；Node.js 22（缺失时脚本可在确认后协助安装）
 - **手动 npm**：Linux、macOS 或 Windows；Node.js >= 22 与 npm
-- 可选外置 [Data Link](https://github.com/datagallery-lab/datalink)：需要语义图谱能力时单独运行（一键部署不依赖）
+- 可选外置 [DataLink](guides/datalink.md)：需要语义图谱能力时单独运行（一键部署不依赖）
 
 请在同一环境内安装和运行项目。Windows 用户不要在 Windows 和 WSL 之间共用 `node_modules`。
 
@@ -64,11 +64,11 @@ cd datafoundry
 
 部署时不要求填写 `LLM_*`。远程主机请设置 `AUTH_PUBLIC_BASE_URL`。重复部署会进入维护窗口：先停止受管进程组，再执行 `npm ci` 与构建。
 
-### 外置 Data Link（可选）
+### 外置 DataLink（可选）
 
-一键部署**只启动** Web + API，**不会**安装、启动或健康检查 Data Link。
+一键部署**只启动** Web + API，**不会**安装、启动或健康检查 DataLink。
 
-若需要语义能力，请单独运行 [Data Link](https://github.com/datagallery-lab/datalink)（常见为 MCP `:8080` + REST `:8081`），再在 Web 工作台的 MCP 设置中添加外部服务，例如：
+若需要语义能力，请单独运行 [DataLink](guides/datalink.md)（常见为 MCP `:8080` + REST `:8081`），再在 Web 工作台的 MCP 设置中添加外部服务，例如：
 
 | 字段 | 示例 |
 | --- | --- |
@@ -77,7 +77,7 @@ cd datafoundry
 | `transport` | `streamable-http` |
 | `toolManifest` | `[{ "name": "datalink_explore" }]` |
 
-名称/id 包含 `datalink`（或 `datagraph`）即可被 Data Link 面板识别。
+名称/id 包含 `datalink` 即可被 DataLink 面板识别。
 
 ## Windows / macOS / 其他：手动 npm 部署
 
diff --git a/mkdocs.yml b/mkdocs.yml
index e4aebf9d..b468d139 100644
--- a/mkdocs.yml
+++ b/mkdocs.yml
@@ -74,6 +74,7 @@ nav:
           - Web workbench: en/guides/web-workbench.md
           - TUI: en/guides/tui.md
           - Data sources: en/guides/data-sources.md
+          - DataLink: en/guides/datalink.md
       - Examples:
           - DTC growth operating review: en/examples/dtc-growth-demo.md
   - Developer Guide: