DeepSeek Harness setup guide

Overview

DeepSeek Harness (the dsh command) is DeepSeek AI’s open-source, plugin-based agent harness. MoLoop can be configured as its custom OpenAI-compatible model provider.

• Official repository

• Official documentation

Quick configuration

Install Node.js first, then start the DeepSeek Harness Web UI with npm.

npx @deepseek-ai/dsh web

The default address is http://127.0.0.1:3080.

Compatibility

• OpenAI Chat Completions

• Streaming

• Tool calling

Step-by-step setup

Start the Web UI

Run dsh in a terminal. The first launch serves the Web UI on local port 3080.

npx @deepseek-ai/dsh web

Add a custom model provider

Open Settings → Model → Add model provider, choose “Custom model API”, and enter the fields below.

Provider ID: moloop
显示名称: MoLoop
基础 URL: https://api.moloop.io/v1
API 协议: OpenAI Chat Completions
API 密钥: YOUR_API_KEY
模型 ID: YOUR_MODEL_ID

Choose the API protocol

Set the protocol to OpenAI Chat Completions. Use the OpenAI-compatible base URL shown by MoLoop and do not append /v1 twice.

Add and select a model

Click Get available models. If the gateway does not expose GET /models, enter the model ID manually. Save it, then select it in a new session.

Recommended models

DeepSeek Harness applies model changes on the next request; restarting the server is not required.

• Browse the model catalog

Things to know

• The API protocol must match the gateway. Do not use OpenAI Responses or Anthropic Messages for a Chat Completions endpoint.

• If model discovery fails, check the key and URL first. You can manually add a model ID when the endpoint does not expose a model list.

• DeepSeek Harness is still in technical preview and may introduce compatibility changes.