What is DeepSeek Harness?
DeepSeek Harness (dsh) is an open-source agent harness developed by DeepSeek AI. It turns a “model” into an “agent”, built around one principle: everything is a plugin.
Agent = Model + Harness
A raw LLM API gives you tokens, not an agent. An agent needs a loop: it calls the model, observes results, invokes tools, stores state, and decides what to do next. That surrounding machinery is the harness.
- Model — the reasoning engine (e.g. a DeepSeek model via API).
- Harness — the runtime that composes prompts, tools, memory, and the agent loop itself.
DeepSeek Harness is that runtime. Unlike monolithic agent frameworks, every part of it — model adapters, the tool registry, session logging, the agent loop itself — is a plugin mounted on a shared context.
Powered by Cordis
DSH is built on Cordis, a plugin framework whose design is described in A Programming Paradigm for Spatiotemporal Composability. Plugins contribute services, typed events, and reversible effects to a shared context. There is no privileged core to patch — you extend DSH by mounting a plugin beside the others, and everything it registers is undone when it unloads.
Two distribution concepts
| Form | What it is |
|---|---|
| Profile | A named composition stored in the Harness home (built-in web and headless templates). |
| Bundle | A distributable package of Cordis config rows + code, declared under the dsh field in package.json. |
Running DSH
Install Node.js, then start the Web UI at http://127.0.0.1:3080 by default:
Or run from a source checkout (clone, pnpm install, pnpm run build, then pnpm dsh web). Full setup: Installation.
The plugin ecosystem
As of 2026-08-13, 600+ repositories carry the dsh-plugin topic on GitHub — vision tools, task boards, launchers, directories, and more. dsh.so indexes 490+ of them in the Plugin Registry.
Further reading
- Install DeepSeek Harness — step-by-step setup.
- Architecture — Cordis, profiles, bundles, and events.
- Plugin Development — write your first plugin.