Skip to content
ML2

DeepSeek Harness: an open agent harness built from plugins

Added NewCopyable prompt below

Published Last verified Released

Our verdict

Try it if you want a coding agent you can take apart: DeepSeek Harness (dsh) runs locally with one npx command, works with a DeepSeek API key or other providers and OpenAI-compatible endpoints, and lets you replace almost any part with a plugin instead of maintaining a fork. Builders are already shipping plugins for it, such as desktop notifications and themes. It is a developer preview with breaking changes promised, and its own safety notice says it has not had a security audit and can run model-generated commands, so run it in a container or throwaway VM with only the files and keys it needs. Ignore posts calling it free frontier models: the harness is free, the models you plug in are not necessarily.

Who it's for
Developers who want an open, hackable coding agent they can extend with plugins, or who already use DeepSeek models and want an official harness for them
Leverage
A full coding agent you can extend or rewire with plugins instead of forking
Effort
Medium
Date added

What it does

DeepSeek Harness (deepseek-ai/deepseek-harness, MIT, TypeScript) is built on the Cordis plugin framework. Models, tools, skills, sessions, sandboxes, storage, the agent loop and the UI are all plugins grouped into profiles. npx @deepseek-ai/dsh web starts a Web UI on 127.0.0.1:3080; you add a model key under Settings, choose a workspace, and the agent reads and edits files, runs commands, delegates work and keeps a plan, asking before actions that need approval. Other modes include a headless runner, ACP, an SDK and a Python SDK. dsh plugin --profile <name> add <package> installs community plugins (tagged dsh-plugin on GitHub). Feedback goes through GitHub Discussions; issues are turned off.

How builders use it

  • Read SAFETY.md, then start it inside a container or VM: npx @deepseek-ai/dsh web.
  • Open Settings, Models, add a DeepSeek key or another provider, then choose your project folder as the workspace.
  • Start with a read-only task such as summarising the repository before letting it edit files.
  • Add a plugin to a profile with dsh plugin --profile web add <package>, after reading its source.
  • Pin a version: the project warns of compatibility-breaking changes during the preview.

What to expect

A local coding agent whose model, tools and behaviour you choose and extend with plugins, without maintaining a fork.

Copy this: Run the Web UI and add a plugin

# Official source: github.com/deepseek-ai/deepseek-harness
# Developer preview: read SAFETY.md first and run it in a container or VM
npx @deepseek-ai/dsh web          # Web UI on http://127.0.0.1:3080
# Settings > Models: add a DeepSeek API key (or another provider)
# Choose workspace: your project folder

# First task, read-only:
# "Summarize this repository and identify its main packages."

# Install a community plugin into a profile (review its code first):
npx @deepseek-ai/dsh plugin --profile web add <package>

Related discoveries

  • ToolCoding & Shipping Agents

    OpenAI Codex CLI

    OpenAI's open-source, lightweight coding agent for your terminal.

    Effort: Low#Open source#CLI

  • ToolCoding & Shipping Agents

    Aider

    Open-source A.I. pair programming in your terminal, with git commits built in.

    Effort: Low#Local / BYO model#Open source

  • ToolCoding & Shipping Agents

    Cline

    Open-source autonomous coding agent for your IDE, CLI or as an SDK.

    Effort: Low#Plugin / extension#Local / BYO model

  • ToolCoding & Shipping Agents

    Gemini CLI

    Google's open-source agent that brings Gemini into your terminal.

    Effort: Low#Open source#CLI

Want more like this?

Practical discoveries, without the noise.