The fastest-starred project in GitHub history, DeepSeek Harness, surpassed 20,000 stars within just over an hour of its release.

Today DeepSeek has been very active, with the official release of the V4 Pro version experiencing a bumpy rollout, an announcement of significant API price increases, and the launch of its official companion tool, DeepSeek Harness (DSH).

Just about an hour and a half after release, its GitHub stars have rapidly surpassed 24,000, making it the fastest-growing star project in GitHub history.

How fast is that? DeepSeek's own previous record was DeepSeek R1, which took 5.7 days to exceed 20,000 stars, while the previous fastest project to hit 20,000 stars was xAI's Grok-1, which took about 1.2 days.

DeepSeek Harness (dsh) is an open-source Agent Harness developed by DeepSeek AI, featuring an everything-is-a-plugin architecture and powered by Cordis.

All agent capabilities—models, tools, skills, sessions, sandboxes, storage, loops, scheduling, UI, and more—are composed of plugins that can be freely replaced and flexibly reorganized.

The company emphasizes that DeepSeek Harness is currently in the developer preview stage and is iterating rapidly, with breaking changes expected in the future.

It's worth noting that the ultra-high-performance tests from the earlier V4 Flash-0731 release were run using this tool, and the V4 Pro likely follows the same approach.

Based on some user feedback since the release, DeepSeek Harness's design philosophy is quite unique, with a plugin-based extension setup offering a high degree of freedom. However, it differs significantly from software many people are already familiar with, such as Claude Code and Codex. Users who are unfamiliar with or don't understand what DeepSeek Harness can do may face a learning curve.