DeepSeek Harness Desktop for macOS and Windows

Who Should Read

LLM developers choosing or extending a coding-agent harness

Core Mechanics

• DeepSeek Harness is an open-source harness
  • It is in worldwide public preview under the MIT license.
  • A harness is the runtime wrapper that lets a model use files, shell, and tools.
• Everything is a plugin
  • It is built on Cordis's "everything is a plugin" architecture.
  • Plugins extend tools, skills, and the UI.
• Build plugins through chat
  • "Creator mode" lets you create plugins by conversation.
  • The demo Pomodoro plugin was built, installed, and verified in 5m 24s.
• Official and experimental plugins
  • Agent loop, Subagents, and Terminal are official plugins.
  • Agent teams, Auto approval review, Scheduled tasks, and Voice input are Experimental.
• Scheduling and debugging built in
  • The Scheduled tasks plugin runs recurring jobs, such as a weekly report every Friday at 17:00.

Community Reactions

• Opinion that desktop build enables telemetry by default
  • Regular dsh web reportedly only sends explicit user feedback.
  • A tip: in $DSH_HOME/cordis.patch.yml, disable desktop-product-telemetry and product-analytics, and set session-log-deepseek enabled: false.
• Experience: customizing core is costly
  • Changing core behavior means maintaining downstream patches on core plugins.
  • One user reported about 25 downstream commits and 0 new plugins.
• Experience: AI built plugins on request
  • A macOS user said settings and workspaces carried over.
  • They asked DSH for font-size and task-finished bell plugins and it delivered both.
• Opinion that two-way subagent messaging is a strength
  • Subagents can message the parent mid-run and the parent can redirect them.
  • You can keep talking on the main thread meanwhile.
• Concerns about third-party plugins and fake sites

How to Apply

• Try the web UI with npx
  • Install Node.js and run npx @deepseek-ai/dsh web.
  • Starting with the web UI avoids the desktop telemetry concern.
• Disable telemetry first
  • If using the desktop app, add the three entries from the comment tip to $DSH_HOME/cordis.patch.yml before first launch.
  • $DSH_HOME defaults to ~/.dsh.
• Build an internal plugin via Creator mode
  • Pick one small repetitive task and request a plugin by chat.
  • After install, check tool calls in Developer tools traces.
• Download only from official sources
  • Get it from the official deepseek.com page or the GitHub repo deepseek-ai/deepseek-harness.
  • Read third-party plugin source before installing.

Try It

# Paste in a shell (requires Node.js; web UI command from the article)

npx @deepseek-ai/dsh web

# Next: in the web UI, ask Creator mode:
# Write a Pomodoro timer plugin for me.

# Disable telemetry (comment tip; add to $DSH_HOME/cordis.patch.yml before first launch)
# - id: desktop-product-telemetry
#   disabled: true
# - id: product-analytics
#   disabled: true
# - id: session-log-deepseek
#   config:
#     enabled: false

Terminology

Related Papers

• Same Simulation, Different Evidence: Where Retrieved Tool Output Costs LLM Agents Accuracy

  Saving long tool output to files cut GPT-5.1 accuracy by 11.3 points; naming the authoritative file restored it.

• Fake News, Real Crashes: Information Injection Attacks on LLM Financial Trading Agents

  Fake headlines hijacked LLM trading agents 62% of the time versus 14% for prompt injection.

• Harness Engineering 101: The LLM is the brain, the harness is the body

  An agent is a loop managing a JSON array; 16 chapters build one from scratch

• Harness2 by Google Research – Recursive Agent Harnessing

  Recursively refining the harness with fixed model weights raised benchmark scores by up to 11.8 points.

• Ousted Savannah police officer gave ICE agent husband access to Flock system

  A police officer shared her Flock login with her ICE-agent husband; an audit found 983 searches and she was fired

• Tmux and Claude Code: Sessions, Send-Keys, and the Ceiling

  Running Claude Code agents in tmux works well, but limits appear around four agents.

Related Resources

• https://www.deepseek.com/en/harness/
• https://github.com/deepseek-ai/deepseek-harness
• https://arxiv.org/abs/2608.25512
• https://frontierharness.org/