DeepSeek Harness
DeepSeek Harness (dsh) is an open-source agent harness in developer preview. Every agent capability is implemented as a composable Cordis plugin for models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI.
Visit deepseek-ai/deepseek-harnessOverview
DeepSeek Harness is an open-source agent harness developed by DeepSeek AI and built on the Cordis plugin system. It is in developer preview with source code available. In DeepSeek Harness, every agent capability is implemented as a plugin that can be swapped or recomposed, and every run is recorded in an append-only session log for traceability.
Key Features
- Everything is a plugin: models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and the UI are provided by Cordis plugins that can be selected, swapped, or extended in configuration.
- Every run is traceable through an append-only session log that records system prompts, reasoning, tool calls and results, subagent scheduling, and context injections.
- Multiple runtime modes: Standard mode includes the full toolset; Code mode orchestrates multi-step tool calls through the Code Mode SDK; Minimal mode keeps only a shell tool and a file editor; Creator mode supports runtime inspection and plugin experiments.
- The Cordis kernel manages plugin mounting, unmounting, and dependencies, with Cordis services and events connecting plugins.
- A Web UI and command-line runtime are available, with quick launch through npx @deepseek-ai/dsh web or source installation options.
Use Cases
- Building and customizing agent harnesses for tasks such as file editing, shell commands, file and web search, skills, planning, goals, subagents, and workflows.
- Benchmarking models in a minimal environment with a two-tool coding agent.
- Creating custom agent presets and experimenting with Cordis plugins in Creator mode.
- Inspecting, resuming, forking, searching, and replaying agent runs from session logs.
Getting Started
- Install Node.js, then run npx @deepseek-ai/dsh web to start the Web UI.
- To run from source, clone the repository, run pnpm install, run pnpm run build, and then run pnpm dsh web.
Deployment & Requirements
- Node.js is required for the npm launch method.
- Running from source requires pnpm, a repository checkout, and built artifacts prepared by pnpm run build before pnpm dsh web.
Before You Adopt
- License: MIT. Review its terms before using, modifying, or distributing the project.
- The project remains in developer preview, is still being tested, and its core plugins and APIs will continue to evolve.
- The append-only session log records everything the model sees, including system prompts, reasoning, tool calls and results, subagent scheduling, and context injections, which is relevant when handling sensitive data.