Vibeleaderboard
← All Intel
Intel / post

DeepSeek ships Harness desktop apps with plugin-based agent system

Source
x.com
Date
From the Daily Brief

DeepSeek Harness is the lab's first-party coding and agent harness, and the v0.2 preview now comes as installable desktop apps for macOS and Windows, with Linux supported through the dsh npm package. For engineers building on DeepSeek models, that means a vendor-made environment to compare against the third-party harnesses they already use. Separately, the Proteus project published a report on letting DeepSeek Harness evolve itself to handle audio transcription over 30 episodes. Every change passed through a staged gate, 27 snapshots were accepted and 3 failed candidates were repaired, with the full trace public. Proteus also judged a perfect external score to be insufficient evidence on its own, a useful habit for anyone grading self-modifying agents. The release is still a preview, so expect interfaces to change.

Read the 2026-10-02 Brief →
deepseek_ai@deepseek_ai

Please check out @DeepSeekHarness. We just released packaged desktop versions for macOS and Windows; Linux users can get it from the @​deepseek-ai/dsh package on npm. https://t.co/Cde0dZ6cPy https://t.co/mIvsRMsXEj

Context

DeepSeek has packaged its app, DeepSeek , as installable desktop programs for macOS and Windows, and Linux users can get it from the @deepseek-ai/dsh package on npm, according to DeepSeek's announcement. An agent harness is the software wrapped around a language model that runs its work loop, gives it tools and shows the results. DeepSeek first opened Harness as a developer preview in August, and the new post describes only the packaging, not new features.

What sets Harness apart is that it treats the parts of the app itself as plugins. The Harness product page, written in Chinese, states the premise as everything is a plugin and says users can install existing plugins or create new ones by conversation in a Create Mode, extending tools, skills and the interface. Its plugin list puts the terminal, the and subagents alongside add-ons marked experimental for agent teams, automated tasks and voice input.

The page's demonstration shows what building a plugin by conversation looks like. Asked for a Pomodoro timer plugin, the agent loads a plugin-development , writes the plugin's package and client files, installs the bundle through a plugin manager tool, edits it, and runs a check that the plugin works, finishing in 5 minutes 24 seconds. That is DeepSeek's own demo of one small interface plugin, not evidence of how well the approach handles larger changes.

Terms in this piece · Glossary
  • AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
  • agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
  • agentic loop — The cycle an agent runs in: decide, call a tool, read the result, decide again — repeating until the goal is met or a stop condition fires.
  • agent skill — A reusable instruction file that teaches an agent how to do one job well — the procedure, the tools, and what counts as done.
More from deepseek_ai
Recommended reads
Comments

Checking sign-in…

Loading comments…