DeepSeek has released an early version of its open source agent harness, built on the Cordis meta-framework. The company said the harness is designed around one core idea: 'Everything is a plugin.' Models, tools, skills, sessions, sandboxes, filesystems, loops, orchestration, and UI are all implemented as plugins and can be mixed, matched, replaced, and extended.
The harness aims to support dynamic composability, allowing plugins to be added and removed on the fly without breaking the application. This is in contrast to systems like Microsoft's Visual Studio Code, where extensions cannot be removed without restarting the host process.
DeepSeek Harness also provides chain of thought traces, recording everything the model sees in an append-only session log. This includes system prompts, reasoning, tool calls, and results. The Trajectory view allows inspection of these records by source, with options to resume, fork, search, and replay.
The release is seen as a move by Chinese AI labs to compete beyond model benchmarks and pricing, focusing on the harness layer where much of the competition is now happening.