DeepSeek Harness with AI Gateway

DeepSeek Harness (dsh) is DeepSeek's terminal coding agent, built on the pi-ai LLM layer. You can configure it to use AI Gateway for unified model access and spend monitoring, while keeping its native model picker.

Run the Vercel CLI setup command for DeepSeek Harness:

npx vercel ai-gateway setup --agent deepseek

The command writes an llm-pi-ai section to $DSH_HOME/settings.yaml naming AI_GATEWAY_API_KEY as the key environment variable, plus a model shortlist that leads with the newest DeepSeek models and follows with the gateway's most-used models. Re-run it to refresh the shortlist.

If you can't use the Vercel CLI, configure DeepSeek Harness manually:

1. Create an API key

   Go to the AI Gateway section in the Vercel dashboard sidebar and click API keys to create a new API key.

2. Export the key

   Export the key in your shell:

   Terminal

   export AI_GATEWAY_API_KEY="your-ai-gateway-api-key"

   To persist the key, add the same line to your shell's startup file.

3. Point DeepSeek Harness at the gateway

   Add the gateway provider to the llm-pi-ai section of $DSH_HOME/settings.yaml (default ~/.dsh/settings.yaml):

   ~/.dsh/settings.yaml

   llm-pi-ai:
     providers:
       vercel-ai-gateway:
         apiKeyEnv: AI_GATEWAY_API_KEY
         models:
           - id: deepseek/deepseek-v4-flash
             name: DeepSeek V4 Flash
           - id: openai/gpt-6-astra
             name: GPT-6 Astra
    
   agent-default-model:
     provider: vercel-ai-gateway
     model: deepseek/deepseek-v4-flash

   The shortlist under models drives the picker. Replace pi-ai's full catalog with the gateway models you want, and add any gateway model ID to it. agent-default-model picks the model DeepSeek Harness starts with.

4. Start DeepSeek Harness

   Open a new terminal so the key is loaded, then run:

   Terminal

   dsh

   Headless runs use the gateway route too. Switch models with /model in a session.

View your usage, spend, and request activity in the AI Gateway section in the Vercel dashboard sidebar. See the observability documentation for more details.

• Browse the models catalog for available models
• Configure spend limits on your API key