Skip to main content
DeepSeek Harness is an open-source Agent Harness from DeepSeek, built around an “everything is a plugin” design. It provides a Web UI and can connect to OpenAI-compatible endpoints through custom providers, so you can use ScitiX models in DeepSeek Harness agent tasks. This guide covers the npm run method. DeepSeek Harness also supports running from source; see the official documentation for details.

Configure

1. Install Node.js

DeepSeek Harness requires Node.js. Open a terminal and run:
If the command returns a version number, Node.js is installed. If not, download and install it from the Node.js website.

2. Start DeepSeek Harness

Run:
The first run downloads the required dependencies and may take some time. After startup, DeepSeek Harness opens the local Web UI by default:
DeepSeek Harness is in developer preview, and version upgrades may introduce breaking changes. It can execute model-generated code and commands. Before running an agent, check the selected workspace, permissions, and requested actions.

3. Add ScitiX as a provider

Create an API Key on the API Keys page first. In the DeepSeek Harness Web UI, open Settings > Models, then choose Add a custom provider. Fill in the following fields, then create the provider. DeepSeek Harness custom provider configured with ScitiX

4. Select runtime settings

Return to the DeepSeek Harness main screen and check the following settings before starting a task:
  • Workspace: select the workspace the agent can access. DeepSeek Harness reads files, runs commands, and executes tasks within the selected workspace.
  • Model: select the ScitiX model you added from the model picker.
  • Mode: choose a runtime mode based on the task. DeepSeek Harness provides Standard mode, PTC mode, Minimal mode, and Creator mode. For first-time verification, you can keep the default Standard mode.
DeepSeek Harness runtime settings with workspace, mode, and model selected

Verify

With a ScitiX model selected, send a simple task:
A normal response means the configuration is successful.