Step‑by‑Step Guide to Run OpenClaw Locally and Connect It to DingTalk

1. Prepare Prerequisites

The official documentation requires a working Node environment (Node.js 24 or 22.16+). Windows users are advised to use WSL2. Additionally, you should have:

A separate workspace directory (e.g., ~/openclaw-demo) for task verification.

A usable model key for the onboarding chat model.

DingTalk application permissions (access to the Open Platform or enterprise app management).

A clean directory that does not mix OpenClaw state with other projects.

Success at this stage means you can confirm the Node version, workspace, model key, and DingTalk permissions are ready.

2. Run the Official Install Script

Use the provided install script instead of manually assembling the environment. For macOS/Linux/WSL2:

# macOS / Linux / WSL2
curl -fsSL https://openclaw.ai/install.sh | bash

For Windows PowerShell:

# Windows PowerShell
iwr -useb https://openclaw.ai/install.ps1 | iex

After the script finishes, run a basic health check before opening the dashboard:

openclaw doctor
openclaw status

doctor

checks dependencies and environment; status shows service status. Any errors (e.g., wrong Node version or missing services) must be resolved before proceeding.

Success criteria: openclaw doctor runs without errors and openclaw status reports the service as started.

3. Execute Onboarding

Run the onboarding wizard to configure the first‑time settings automatically:

openclaw onboard
openclaw config file

Key items to verify during onboarding:

Model provider: Connect a basic chat model that can respond.

Gateway binding: The quick start binds the gateway to a local loopback address and fixed port for safety and easier debugging.

Workspace: Attach the previously created ~/openclaw-demo directory.

Configuration file location: Usually ~/.openclaw/openclaw.json.

Run the following three commands to confirm onboarding success:

openclaw config file
openclaw status
openclaw dashboard

If the config file exists, the service is running, and the dashboard opens, onboarding is considered passed.

4. Verify via Dashboard Before Adding a Channel

Open the dashboard and send a minimal verification message, for example:

你好，请先确认你已经正常启动。
然后告诉我当前 workspace 指向哪个目录。

If OpenClaw replies correctly and reports the workspace path, three things are confirmed: the model configuration works, the dashboard can reach the gateway, and the local configuration is effective.

Success criteria: the system not only opens the page but also returns the first message and reads the current configuration.

5. Configure the DingTalk Channel

After the previous steps succeed, set up the DingTalk channel. The required chain includes creating a DingTalk robot app, obtaining credentials, and aligning callback URLs.

Create a robot application in DingTalk Open Platform or enterprise backend to get Client ID / Client Secret (also called AppKey / AppSecret).

Enter these credentials in OpenClaw’s channel configuration.

Provide the callback URL generated by OpenClaw to DingTalk.

Test private chat first, then group chat, keeping the two scenarios separate.

The full DingTalk integration consists of five stages: app creation, model selection, channel parameter entry, private‑chat verification, and group‑chat verification. Private‑chat validation isolates channel connectivity and model response, making later group‑chat issues easier to locate.

Success criteria: a simple message sent in a DingTalk private chat receives a stable reply, followed by a successful @‑mention test in a group chat.

6. Perform a Real Task Verification

Even if chat works, the system is not fully validated until it can execute a concrete task. Create a file in the workspace, e.g., ~/openclaw-demo/brief.md, with some content, then issue the following task via the dashboard or DingTalk:

请读取 workspace 里的 brief.md，
整理成 3 条待办，并在同目录新建 todo.md。

Verification points:

If OpenClaw reads brief.md, the workspace is correctly linked.

If it creates todo.md, file handling and write permissions are functional.

If the task succeeds through DingTalk, the end‑to‑end channel from DingTalk to the local agent is closed.

Only when OpenClaw can read files, write results, and return them does the local run‑through count as truly complete.

7. Troubleshooting Checklist (Order of Checks)

Script installed but command not found: Verify the install script location and PATH before reinstalling the channel.

Dashboard opens but no reply: Check model provider configuration and onboarding first, not DingTalk.

DingTalk does not reply: Confirm app credentials, callback URL, and model settings; then isolate private‑chat vs group‑chat issues.

Task execution fails: Ensure the workspace is correctly paired and the target file exists; then verify tool permissions.