DeepSeek Harness
An open-source agent harness built on a plugin-first design, run from a local Web UI.
- Stars
- 269k
- Forks
- 32.7k
- Licence
- MIT
- Language
- TypeScript
- Last update
- 7 days ago
- Type
- Dev tool
What it is
DeepSeek Harness (dsh) is an open-source agent harness from DeepSeek AI, built so that nearly every part is a plugin. It runs on Cordis, a plugin framework the README links to, and can be started as a local Web UI. The README describes it as being in developer preview, so the interface and APIs may change.
- Start the local Web UI with npx @deepseek-ai/dsh web
- Run the project from a source checkout for development and testing
- Follow the development guide and architecture docs to understand the plugin system
- Build or share a plugin tagged with the dsh-plugin topic
Great for
- Developers who want to experiment with an agent harness they can extend with plugins
- Developers already interested in the Cordis plugin model
- People comfortable running a local dev tool from the command line
- Plugin authors who want to publish to the dsh-plugin topic
Not ideal for
- Users who need a stable production system, since the README warns of compatibility-breaking changes during developer preview
- Anyone who has not read the SAFETY.md notice, which the README asks users to review before running it
- Non-technical users looking for a finished consumer app; the README assumes Node.js or a repository checkout
How to get it
- 1
Install with a package manager
You'll need: Node.js · pnpm (for running from source)
Run from npm
npx @deepseek-ai/dsh webClone the repository
git clone https://github.com/deepseek-ai/deepseek-harness.gitInstall dependencies
pnpm installBuild artifacts
pnpm run build - 2
Run it
- Install Node.js.
- Run npx @deepseek-ai/dsh web in a terminal.
- Open http://127.0.0.1:3080 if the browser does not open automatically.
- Pass --no-open to run the server without opening a browser.
- 3
Get the most out of it
- Read the Web UI guide linked from the README for available features
- Use pnpm run dev:web while editing source to rebuild client bundles automatically
- Use make help to list the Make targets for Web and Desktop
- Read the architecture documentation to understand how plugins fit together
- Join the Discord or GitHub Discussions to ask questions and share plugins
Downloads
Strengths
- Built on a stated everything-is-a-plugin architecture
- Can be started quickly with a single npx command, per the README
- Documentation, a Discord community and GitHub Discussions are linked from the README
Trade-offs
- The README explicitly says it is in developer preview with breaking changes expected
- The README is truncated and gives limited detail on features and models supported
- Requires understanding of Node.js or pnpm and a local dev workflow
How it compares
| Project | Stars | Type | Level | Updated | Best for |
|---|---|---|---|---|---|
| deepseek-harness (this) | 269k | Dev tool | intermediate | 7 days ago | Developers who want to experiment with an agent harness they can extend with plugins |
| langchain | 149k | Framework | intermediate | 4 days ago | Python developers building LLM-powered applications |
| crewAI | 60.4k | Library | intermediate | 4 days ago | Python developers prototyping multi-agent workflows |
| dsh-our-free-model | 56.6k | Dev tool | beginner | 4 days ago | Developers who already use dsh and want to try models without creating accounts or keys |
| dsh-desktop | 34.5k | App | beginner | 4 days ago | People who want to run DeepSeek Harness without using the command line |
| learn-harness-engineering | 22.9k | Learning resource | beginner | 14 days ago | Developers new to building or controlling AI coding agents |
Stats synced from GitHub. The assessment above was drafted with AI from the project's README on 2026-10-29 and may miss recent changes — check the README before relying on a command. MIT licence: free to use, modify and distribute, including commercially, as long as you keep the copyright notice.