#DeepSeek Harness

DeepSeek's official open-source agent runtime (command: dsh), built on an "everything is a plugin" architecture — one command starts a local Web UI. MIT-licensed, model-agnostic: add ByteSeek as an OpenAI-compatible provider and you are in.

#Install

Requires Node.js ^22.19.0 or >=24.0.0 (odd major versions such as Node 23 are unsupported) — check with node --version first.

cd your-project
npx @deepseek-ai/dsh web

npm install -g @deepseek-ai/dsh
dsh web

Website: deepseek.com/harness; official docs: deepseek-harness.github.io.

The Web UI starts at http://127.0.0.1:3080 by default and opens your browser; over SSH it only prints the URL. Pass --no-open to suppress the auto-open.

#Prerequisites

• dsh web starts and serves the Web UI
• An OpenAI-format group key from the API Keys page

#Configure

1. In the Web UI open Settings → Models.
2. Add an OpenAI-compatible provider:
   • Base URL: https://tokens.byteseek.ai/v1
   • API Key: your sk- key
   • Model: an exact ID from the model plaza, e.g. gpt-5.6-sol
3. Changes apply immediately — no restart.
4. Back in the session view, pick a workspace directory first, then switch the model selector to your new provider.

#Verify

Send a message to the configured model in the Web UI; a normal reply means success. Charges appear on the usage page.

#Common issues