DeepSeek Harness

About

DeepSeek Harness (`dsh` on the command line) is an MIT-licensed open-source agent harness from DeepSeek AI, currently in developer preview and open for testing by agent-harness developers worldwide with its source code published. Its thesis is Agent = Model + Harness: the model is the soul of an agent, while the harness lets that agent understand its environment, use tools, and keep working in real-world settings. Architecturally it is built on the Cordis plugin system — the Cordis kernel only manages plugin mounting, unmounting and dependencies and holds no capabilities of its own, while every agent capability (models, tools, skills, sessions, sandboxes, storage, loops, scheduling and the UI) is supplied by plugins that cooperate through Cordis services and events. Developers can therefore select, swap or extend any capability purely in configuration, without touching the DeepSeek Harness source. The second design principle is that every run is traceable: everything the model sees is written to an append-only session log — system prompts, reasoning, tool calls and results, subagent scheduling, and every context injection. The Trajectory view lets you inspect those records by source, and resume, fork, search and replay all operate on the same event stream. Four runtime modes ship out of the box: Standard mode is a full coding agent with file editing, shell, file and web search, skills, planning, goals, subagents and workflows; Code mode (called PTC mode on the Chinese page) exposes tools through the Code Mode SDK so the model can combine multi-step operations in a single TypeScript program; Minimal mode keeps only a persistent bash and str_replace_editor for benchmarking models in a minimal environment; and Creator mode is built for authoring custom agent presets, adding runtime inspection, in-memory plugin experiments and preset-authoring guidance. Getting started means installing Node.js and running `npx @deepseek-ai/dsh web`, which serves the Web UI at http://127.0.0.1:3080 by default; add a DeepSeek API key under Settings → Models, or point it at another provider or a custom OpenAI-compatible endpoint. You can also clone the repository and build from source with pnpm. A running instance is composed from profiles and bundles, with `web` and `headless` shipping as templates, and a Python SDK plus additional CLI modes are documented. The project is written in TypeScript, its plugin ecosystem is indexed by the GitHub `dsh-plugin` topic, and support lives in GitHub Discussions and Discord. Note that DeepSeek states core plugins and base APIs are still iterating rapidly during the preview, and there will be compatibility-breaking changes.

Key Features

• ✓Everything is a plugin: models, tools, skills, sessions, sandboxes, storage, loops, scheduling and the UI
• ✓The Cordis kernel only handles plugin mounting, unmounting and dependencies, holding no agent capabilities itself
• ✓Compose in configuration: select, swap or extend any capability without changing the source
• ✓Append-only session log: system prompts, reasoning, tool calls and results, subagent scheduling and every context injection
• ✓Trajectory view inspects run records by source, with resume / fork / search / replay on one event stream
• ✓Standard mode: a full coding agent with file editing, shell, file and web search, skills, planning, goals, subagents and workflows
• ✓Code mode (PTC mode): the Code Mode SDK lets the model combine multi-step tool calls in one TypeScript program
• ✓Minimal mode: just persistent bash and str_replace_editor, for benchmarking models in a minimal environment
• ✓Creator mode: inspect the runtime, experiment with Cordis plugins in memory and author custom agent presets
• ✓One command to the Web UI: install Node.js, run `npx @deepseek-ai/dsh web`, served at 127.0.0.1:3080 by default
• ✓Bring your own model: configure a DeepSeek API key, another provider, or a custom OpenAI-compatible endpoint
• ✓Runtime shape is layered from profiles and bundles, with web and headless shipping as templates
• ✓Built-in approval and sandbox policy, prompting before operations that need authorization
• ✓A Python SDK and additional CLI modes for embedding it into your own pipelines
• ✓Open plugin ecosystem: the GitHub `dsh-plugin` topic indexes community plugins, with Discussions and Discord for support