DeepSeek Harness developer preview launched
Original: DeepSeek Harness developer preview
Why This Matters
DeepSeek's open plugin architecture lowers the barrier for developers to build and customize production-grade AI agents.
DeepSeek has released Harness, a plugin-based agent framework, in developer preview. Built on the Cordis kernel, every agent capability — models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and UI — is a swappable plugin. Source code is publicly available on GitHub.
DeepSeek has launched DeepSeek Harness in developer preview, an open-source agent harness framework targeting developers worldwide. The system is built on the Cordis plugin kernel, which manages plugin mounting, unmounting, and dependency resolution. Every agent capability is implemented as a plugin, including models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI, allowing developers to swap or extend any component via configuration without modifying core source code.
All agent runs are fully traceable through an append-only session log that captures system prompts, reasoning steps, tool calls and results, subagent scheduling, and context injections. A Trajectory view enables inspection by source, with support for resume, fork, search, and replay on the same event stream.
Harness ships with four runtime modes: Standard (full toolset including file editing, shell, web search, planning, subagents, and workflows), Code (Standard capabilities plus a TypeScript SDK allowing the model to orchestrate multi-step tool calls in a single program), Minimal (shell tool and file editor only, for benchmarking), and Creator (for building custom agent presets with runtime inspection and plugin experimentation). Installation is available via npx or by cloning the GitHub repository.