DeepSeek Releases Open-Source AI Agent Runtime Built on a Plugin-First Architecture
DeepSeek AI has released DeepSeek Harness (dsh), an open-source agent runtime currently in developer preview under the MIT license. The project provides a ready-to-use environment for building and running coding and automation agents, offering file editing, shell access, web search, sub-agent support, and a web UI out of the box. Built on a plugin framework called Cordis, developed with researchers from Peking University, the system treats every component — including the model adapter, tool registry, and agent loop — as a swappable plugin with no privileged core. The harness is model-agnostic, supporting providers such as Anthropic, OpenAI, AWS Bedrock, Azure, and Google Gemini, alongside DeepSeek's own models. The release highlights a broader industry shift toward treating the agent runtime itself — not just the underlying model — as the primary architectural product.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in