Back to feed
News Story
APriority79
DeepTech深科技
1 sources

DeepSeek officially open-sources Harness: finally its own Vibe Coding entry point

DeepSeek has officially released and open-sourced DeepSeek Harness (DSH) as a developer preview, an agent runtime framework that connects models to file systems, terminals, web tools, and organizes context and task execution. Built on Cordis, it adopts an 'everything is a plugin' architecture and supports multiple agent presets, now available on GitHub. This move marks DeepSeek's first foray into the execution layer beyond models, providing an entry point for Vibe Coding.

SynthePulse Insight · AI deep readingMembers

DeepSeek Harness: From Model to Execution Layer, DeepSeek's Agent Ambition

Version 1 · 1 source

DeepSeek's open-source Agent Harness (DSH) marks its shift from model competition to execution-layer competition. This article examines its architecture, capabilities, and strategic significance.

  • DeepSeek open-sourced the DeepSeek Harness (DSH) developer preview on August 13, launchable via npm.
  • DSH is built on Cordis and adopts an 'everything is a plugin' architecture, with replaceable models, tools, and Agent Loops.
  • DSH offers Web UI, TUI, Headless modes, and interfaces such as ACP, JSON-RPC, and Python SDK.
Open section navigationOpen-Source Release and Positioning

Open-Source Release and Positioning

DeepSeek released and open-sourced the DeepSeek Harness (DSH) developer preview on August 13. This is an agent runtime framework, not a foundation model or API client. Users can install it via npm and run `npx @deepseek-ai/dsh web` to launch the Web UI, but the company emphasizes it is still in preview and may have incompatible updates.

The release of DSH was foreshadowed: the July 31 V4 Flash official update log mentioned that its testing framework was the 'upcoming' DeepSeek Harness minimal mode, indicating that Harness had already been involved in model evaluation.

Free for now

Read the full analysis

3 more sections of analysis, plus the full takeaway

Loading

Credibility boundary

This article is based on a report from DeepTech深科技, which includes official information (such as GitHub, npm, and API documentation) and observations from media early access. Some data (such as V4 Pro benchmark scores) comes from official statements but has not been independently verified.

Primary report

DeepTech深科技

Primary source