DeepSeek Harness Quick Start
DeepSeek Harness (harness.vin) is an open-source AI Agent framework that keeps AI agents working in real-world scenarios. This tutorial walks you through installation, configuration, and creating your first AI Agent with the everything-is-a-plugin architecture.
Installation & Setup
Prerequisites
Make sure Node.js is installed (needed to launch the Web UI via npx). Verify with node -v.
One-Command Launch
Run npx @deepseek-ai/dsh web to start the web UI — the command prints its access address. No global install needed — npx handles everything.
Install from Source (Optional)
For deep customization, clone the source via git clone https://github.com/deepseek-ai/deepseek-harness, then follow the repository README to build and run with pnpm.
Configure Plugins
DeepSeek Harness follows an everything-is-a-plugin architecture. Pick the model, tool, skill, and sandbox plugins you need at the config layer — combine full agent capabilities without touching source code.
Run Your Agent
Once configured, your AI Agent is ready to run. DeepSeek Harness records a complete trajectory event stream with resume, fork, search, and replay support.
Get Started in Seconds
Start with one command, and experience DeepSeek Harness's plugin-based architecture and tool calling.
# Launch the Web UI with one command (prints its address)
npx @deepseek-ai/dsh web
# Headless mode: run one fresh session and print the final answer
dsh --profile headless "Summarize this repository."
# Call the same API from your own programs (Python 3.10+)
python -m pip install deepseek-harness-sdk
from deepseek_harness import DeepSeekHarness
with DeepSeekHarness(provider="deepseek-official") as harness:
result = harness.run("Summarize this repository.")
print(result.final_response)Core Capabilities
Python SDK
Install deepseek-harness-sdk via pip to call the same API as the Web UI from your own programs — run agents and get the final response.
Plugin Architecture
Models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and UI — every capability is a plugin, freely replaceable and recombinable.
Multiple Run Modes
Standard mode, PTC mode, minimal mode, and creator mode — pick the right one for each scenario.
Trajectory
A complete append-only event stream with resume, fork, search, and replay, making AI execution fully traceable.
Learn More
FAQ
How do I install DeepSeek Harness?
Run npx @deepseek-ai/dsh web for a quick start (Node.js required), or install from source via git clone and build with pnpm following the repository README.
Is DeepSeek Harness open source?
Yes. DeepSeek Harness (harness.vin) is an open-source AI Agent framework released under the MIT license and hosted on GitHub.
What run modes does DeepSeek Harness support?
Four run modes: standard, PTC, minimal, and creator mode.
How do I create a custom plugin?
DeepSeek Harness is built on the Cordis plugin system — a plugin is a TypeScript module that exports an apply function, registering capabilities through ctx and declaring dependencies. See the Plugins page for details.