DeepSeek has released its agent development framework, DeepSeek Harness, and open-sourced the code. The framework supports highly pluggable models, tools, and strategies, with over a hundred built-in plugins, designed for self-evolution and DIY customization. The article also mentions an official price increase on the 17th, especially for caching.
SynthePulse Insight · AI deep readingMembers
Version 1 · 1 source
DeepSeek Harness is open-sourced, challenging Codex's closed philosophy with its Cordis plugin architecture, but its value for average users awaits ecosystem maturation.
DeepSeek Harness (DSH) has finally been released, and its source code is open-sourced. The author, who has been in beta testing for nearly half a month, migrated almost all Vibe Coding projects from Codex, calling it "a saddle thoroughly built for 'self-evolution' and 'DIY'."
Everything in DSH—models, tools, strategies, storage, context management—is a pluggable building block. The official version includes over a hundred built-in plugins, with more to come from the community. The author sees this as "the Android of the agent era."
This article is based on a deep-dive experience piece from QbitAI, authored by a beta tester, so the views are subjective. The mentioned price increases and plugin counts are as relayed by the author and have not been officially confirmed.
Primary report
Primary source