DeepSeek Harness

Follow the Windows or macOS steps for your system.

Windows

1. Install

Open PowerShell from the Start menu.

Check Node.js and npm:

node --version
npm --version

If either command is missing, install the LTS release from Node.js and reopen the terminal. Once both version numbers appear, run:

npm install --global @deepseek-ai/dsh
dsh --version

2. Configure

Prepare an API Key and copy a model ID from the model catalog that supports Chat Completions and is available to your Key. Replace YOUR_MODEL with that ID.

The configuration file is %USERPROFILE%\.dsh\settings.yaml.

1. Press Win + E to open File Explorer.
2. Enter %USERPROFILE%\.dsh in the address bar and press Enter.
3. Right-click settings.yaml and choose Open with → Notepad.
4. If the directory is missing, create the .dsh folders inside %USERPROFILE%. If the file is missing, create it in Notepad.

llm-pi-ai:
  providers:
    getrun:
      displayName: GetRun
      apiKeyEnv: GETRUN_API_KEY
      api: openai-completions
      baseURL: https://api.get-apikey.com/v1
      models:
        - id: YOUR_MODEL
          name: YOUR_MODEL

Replace both YOUR_MODEL values with the same model ID. Keep apiKeyEnv: GETRUN_API_KEY as shown; enter the API Key in the Web interface next. For an existing file, merge llm-pi-ai.providers.getrun and keep other settings. Use spaces for YAML indentation.

Save in the directory above. In Notepad, use Save As → All files → UTF-8 and name the file settings.yaml without an extra .txt extension.

3. Use

Open your project folder in File Explorer, enter powershell in the address bar, and press Enter.

dsh web

Open the local URL printed in the terminal, then:

1. Open Models and find the GetRun provider.
2. Enter and save your API Key.
3. Start a session and select the model you configured.
4. Send Reply only OK. Once you receive a reply, enter your task.

Keep the terminal running during use. Press Ctrl + C in that terminal to stop it.

macOS

1. Install

Press Command + Space to find and open Terminal.

Check Node.js and npm:

node --version
npm --version

If either command is missing, install the LTS release from Node.js and reopen the terminal. Once both version numbers appear, run:

npm install --global @deepseek-ai/dsh
dsh --version

2. Configure

Prepare an API Key and copy a model ID from the model catalog that supports Chat Completions and is available to your Key. Replace YOUR_MODEL with that ID.

The configuration file is ~/.dsh/settings.yaml.

1. Open Finder and press Command + Shift + G.
2. Enter ~/.dsh and select Go.
3. Open settings.yaml in TextEdit or VS Code.
4. If the folder or file is missing, choose File → Open Folder in VS Code, press Command + Shift + G, and enter ~ to open your home directory. Create the .dsh folders in the Explorer sidebar, then create settings.yaml inside.

llm-pi-ai:
  providers:
    getrun:
      displayName: GetRun
      apiKeyEnv: GETRUN_API_KEY
      api: openai-completions
      baseURL: https://api.get-apikey.com/v1
      models:
        - id: YOUR_MODEL
          name: YOUR_MODEL

Replace both YOUR_MODEL values with the same model ID. Keep apiKeyEnv: GETRUN_API_KEY as shown; enter the API Key in the Web interface next. For an existing file, merge llm-pi-ai.providers.getrun and keep other settings. Use spaces for YAML indentation.

Press Command + S to save. In TextEdit, choose Format → Make Plain Text and keep the filename settings.yaml.

3. Use

Open Terminal, type cd with a trailing space, drag your project folder from Finder into Terminal, and press Enter.

dsh web

Open the local URL printed in the terminal, then:

1. Open Models and find the GetRun provider.
2. Enter and save your API Key.
3. Start a session and select the model you configured.
4. Send Reply only OK. Once you receive a reply, enter your task.

Keep the terminal running during use. Press Ctrl + C in that terminal to stop it.

Configuration location

If DSH_HOME is set, edit settings.yaml in that directory. Linux uses the same installation and launch commands and defaults to ~/.dsh/settings.yaml. See the official DeepSeek Harness documentation for more usage details.