Is DeepSeek Harness free to use?

DeepSeek Harness is released under the MIT license, so the full source is free to read, modify and redistribute. Running an agent still needs model access: the quickstart tells you to open Settings, go to Models, and enter a DeepSeek API key before starting a session, so the tokens you spend are billed separately.

Can DeepSeek Harness run models other than DeepSeek's?

Yes. The DeepSeek API is the default provider in the quickstart, but the documentation points to a model configuration guide that covers other providers and custom OpenAI-compatible endpoints. Because the model itself is just another plugin in DeepSeek Harness, swapping providers does not require changing the framework's source.

What runtime modes does DeepSeek Harness ship with?

DeepSeek Harness ships four presets. Standard is the full coding agent with the complete tool set. Code drives multi-step operations through TypeScript. Minimal cuts the agent down to a bash shell and a file editor. Creator is for building your own presets and inspecting the runtime while it works.

Is DeepSeek Harness ready for production use?

No. DeepSeek labels Harness a developer preview and warns in the README, in capital letters, that there will be compatibility-breaking changes. The project is iterating quickly, so pinning a version and expecting a stable plugin API across upgrades is not safe yet.

Where do you find plugins for DeepSeek Harness?

Community plugins for DeepSeek Harness are published on GitHub under the dsh-plugin topic, which acts as the informal registry. Discussion happens in the repository's GitHub Discussions and on the project's Discord. Anyone publishing a plugin is asked to tag it with dsh-plugin so it is discoverable there.

DeepSeek · 2026-08-13 · major

DeepSeek Harness — open-source agent framework built entirely from plugins

DeepSeek Harness is an open-source agent framework released today under the MIT license as a v0.1 developer preview. Models, tools, skills, sessions, sandboxes, storage, loops, scheduling and the UI are all swappable plugins.

GitHub repository card for deepseek-ai/deepseek-harness — 'Everything is a Plugin'

An MIT-licensed agent framework from DeepSeek where models, tools, sandboxes and even the interface are swappable plugins.

Key specs

Git hub stars (day one)	34,074

Quick facts

Maker	DeepSeek
Version	v0.1 developer preview
License	MIT
Built on	Cordis plugin framework
Language	TypeScript
Web UI	http://127.0.0.1:3080

What is it?

DeepSeek Harness arrives as a v0.1 developer preview: an agent runtime you assemble from plugins rather than a fixed product you configure. Models, tools, skills, sessions, sandboxes, storage, loops, scheduling and the user interface are each a plugin that can be replaced or recomposed. DeepSeek published the whole codebase under the MIT license on the day of the announcement.

How does it work?

The plugin system sits on Cordis, a meta-framework that resolves dependencies between plugins and wires them together at runtime. Every session is written to an append-only log, so each step an agent takes stays traceable after the fact. Four presets ship in the box — Standard, Code, Minimal and Creator — and each one is simply a different composition of the same plugins. The CLI is named dsh and starts a local web interface.

Why does it matter?

Teams building coding agents normally fork a closed tool or rebuild the agent loop from scratch. This framework hands them the middle layer — the part between a model and the files, shells and tools it needs — as MIT-licensed code they can edit and extend. The repository passed 34,000 stars on its first day, so a plugin ecosystem has a large audience from the start.

Who is it for?

agent and developer-tooling engineers

Frequently asked questions

Is DeepSeek Harness free to use?: DeepSeek Harness is released under the MIT license, so the full source is free to read, modify and redistribute. Running an agent still needs model access: the quickstart tells you to open Settings, go to Models, and enter a DeepSeek API key before starting a session, so the tokens you spend are billed separately.
Can DeepSeek Harness run models other than DeepSeek's?: Yes. The DeepSeek API is the default provider in the quickstart, but the documentation points to a model configuration guide that covers other providers and custom OpenAI-compatible endpoints. Because the model itself is just another plugin in DeepSeek Harness, swapping providers does not require changing the framework's source.
What runtime modes does DeepSeek Harness ship with?: DeepSeek Harness ships four presets. Standard is the full coding agent with the complete tool set. Code drives multi-step operations through TypeScript. Minimal cuts the agent down to a bash shell and a file editor. Creator is for building your own presets and inspecting the runtime while it works.
Is DeepSeek Harness ready for production use?: No. DeepSeek labels Harness a developer preview and warns in the README, in capital letters, that there will be compatibility-breaking changes. The project is iterating quickly, so pinning a version and expecting a stable plugin API across upgrades is not safe yet.
Where do you find plugins for DeepSeek Harness?: Community plugins for DeepSeek Harness are published on GitHub under the dsh-plugin topic, which acts as the informal registry. Discussion happens in the repository's GitHub Discussions and on the project's Discord. Anyone publishing a plugin is asked to tag it with dsh-plugin so it is discoverable there.

Try it

npx @deepseek-ai/dsh web