What is DeepSeek Harness?
DeepSeek Harness (dsh) is the agent app DeepSeek built around its own models. It runs as a desktop app or a local web UI, works inside a folder you choose, and can read and edit files, run commands and tests, research with citations, preview documents and spreadsheets, and hand work to subagents, asking first before actions your permission policy flags.
Where Claude Cowork or the Codex app are finished products, everything in dsh, down to the agent loop and terminal, is a plugin you can swap, and Creator mode builds new ones from a chat request. It is free, but still a developer preview with breaking changes and no security audit.
Who is DeepSeek Harness best for?
DeepSeek Harness is best for developers on the DeepSeek API, people who want to build or reshape their own agent, mixed coding and document work, and teams that want to choose their model.
Developers on the DeepSeek API
It defaults to DeepSeek-V4.1-Flash at $0.30 per million input tokens at peak, and users report cache hit rates of 98% or more.
Building your own agent
Creator mode writes and installs plugins from a chat request, and the Python SDK, headless and ACP modes embed the same agent elsewhere.
Mixed coding and desk work
One app handles repos and tests as well as Word, Excel, PDF and slide files, with previews and a diff for each turn.
Teams that pick their own model
Besides DeepSeek it takes Anthropic, OpenAI, Kimi and GLM keys, or any gateway that speaks the OpenAI or Anthropic API.
What's included with DeepSeek Harness?
Plugins for everything, Creator mode, scheduled tasks, a trace view, MCP servers and several ways to run it, from the desktop app to a Python SDK.
- Plugins for everything
- Tools, skills, the interface and the agent loop are plugins; official ones include subagents, a terminal, agent teams and voice input.
- Creator mode
- Describe a plugin in chat and the agent writes, installs and checks it.
- Scheduled tasks
- An experimental plugin runs prompts once, daily, weekly, on an interval or on a cron schedule while the app is open.
- Trace view
- Every turn, tool call and timing can be inspected to debug a run.
- MCP servers
- It connects MCP servers over stdio or HTTP, with ready configurations for three memory servers.
- Several entry points
- Desktop app, browser UI, headless one-shot runs, a Python SDK and ACP for other clients.
How much does DeepSeek Harness cost?
DeepSeek Harness is free and open source under MIT; you pay only for model usage, such as $0.30 per million input tokens and $1.20 per million output tokens for DeepSeek-V4.1-Flash at peak.
- Desktop app for macOS and Windows, web UI via npx
- Bring your own model key
- No paid tier
- Off-peak: $0.15 / $0.60
- Cached input $0.006 at peak, $0.003 off-peak
- 1M-token context, up to 384K output tokens
- Off-peak: $0.66 / $1.98
- Cached input $0.044 at peak, $0.022 off-peak
- No image input
The app is free and open source; what you pay for is the model it calls. On DeepSeek's API, peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, off-peak rates are half, and Chinese public holidays count as off-peak. Charges come out of a prepaid balance. Anthropic, OpenAI, Kimi, GLM and other providers bill at their own rates. Prices are in US dollars. Full pricing on api-docs.deepseek.com. We checked these prices on .
How do readers rate DeepSeek Harness?
No one has rated DeepSeek Harness yet. Be the first.
Sign in to rate. One rating per account.
DeepSeek Harness pros and cons
It is free, cheap to run on DeepSeek's models and easy to extend, but it is an unaudited developer preview with breaking changes, no Linux desktop app and few ready connectors.
Pros
- Free and MIT-licensed with the full source on GitHub, so you can audit it, fork it or build it into your own product.
- Cheap to run on DeepSeek's models, and several users report prompt cache hit rates of 98% or more, which cuts the token bill further.
- Unusually easy to change: everything is a plugin, and users say Creator mode builds missing ones, such as a GitHub plugin, in minutes.
- Not locked to DeepSeek: it accepts Anthropic, OpenAI, Kimi and GLM keys or any OpenAI- or Anthropic-compatible endpoint.
Cons
- A developer preview: DeepSeek warns of breaking changes, and users on Reddit and GitHub report runs failing after upgrades.
- Not security audited: DeepSeek's own safety notice says it can damage files or leak credentials and advises a virtual machine or container.
- No Linux desktop app, the most common complaint on Reddit; Linux users run the browser UI from npx instead, and Intel Macs are not supported.
- Few ready-made connectors: users ask for an official plugin store and integrations such as GitHub, Gmail and Google Drive, and new models on a gateway may need adding by hand.
Is DeepSeek Harness open source?
Yes. DeepSeek Harness's code is public under the MIT licence, so you can read it, change it and run it yourself.
On GitHub
285,079stars
- Last commit
- 18 Nov
- Repository age
- 3 months
- Licence
- MIT
- Repository
- deepseek-ai/deepseek-harness
What are the best DeepSeek Harness alternatives?
DeepSeek Harness is an open source alternative to MiniMax Agent, Kimi Work and Claude Cowork. goose, AionUi and Leon do the same job and are open source too.
MiniMax's general-purpose AI agent, now also called MiniMax Code, for coding, research, documents and media.
Moonshot AI's desktop agent for knowledge workers that handles local files, browser tasks and scheduled jobs.
Choose it when you want a finished agent for files, apps and scheduled jobs with Claude's connectors, and are happy to pay from $20 a month on Claude Pro.
AI agent on your own computer that writes code, runs commands and automates tasks with any model.
Open-source desktop app to run Claude Code, Codex and 20+ AI agents together on your files.
Runs on your own machine or server, remembers context, and can operate your browser and desktop apps.
Questions about DeepSeek Harness
What is DeepSeek Harness?
DeepSeek Harness is DeepSeek's free, MIT-licensed agent app for Mac, Windows and the browser that codes, works with documents and data, researches and runs scheduled tasks, with every part built as a plugin.
Who is DeepSeek Harness for?
DeepSeek Harness is best for developers on the DeepSeek API, people who want to build or reshape their own agent, mixed coding and document work, and teams that want to choose their model.
Is DeepSeek Harness free?
Yes. DeepSeek Harness has a free DeepSeek Harness plan: desktop app for macOS and Windows, web UI via npx.
How much does DeepSeek Harness cost?
DeepSeek Harness's paid plans are DeepSeek-V4.1-Flash on the API at $0.30 / $1.20 per million tokens in / out at peak and DeepSeek-V4-Pro on the API at $1.32 / $3.96 per million tokens in / out at peak.
Is DeepSeek Harness open source?
Yes. DeepSeek Harness's code is public under the MIT licence, so you can read it, change it and run it yourself.
What are the best DeepSeek Harness alternatives?
The closest alternatives on Recaply are MiniMax Agent, Kimi Work, Claude Cowork, goose, AionUi and Leon.
Where can I try DeepSeek Harness?
On its official website, deepseek.com.
Do I need a DeepSeek API key to use DeepSeek Harness?
You need a key from some model provider. The quickstart uses a DeepSeek API key, but Settings, then Models also accepts Anthropic, OpenAI, Kimi and GLM keys, or a custom endpoint that speaks the OpenAI or Anthropic API. Sign-ins such as a Codex subscription are not supported yet.
Is DeepSeek Harness safe to run on my computer?
Use it with care. DeepSeek says it has not been security audited and can run model-written commands, load third-party plugins and reach your files and credentials. It advises least privilege, backups, and a virtual machine or container.
Does DeepSeek Harness work on Linux?
Not as a desktop app: downloads cover Apple silicon Macs on macOS 13 or later and 64-bit Windows 10 or later. On Linux you run the browser UI with npx and Node.js, or the Python SDK, which supports Linux x64 and arm64.
Is DeepSeek Harness the same as the DeepSeek chat app?
No. The DeepSeek app is a chat assistant. DeepSeek Harness is a separate open-source agent that works in a folder on your machine, edits files, runs commands and plugins, and calls models through an API key you add.
Our verdict
DeepSeek Harness is worth trying if you want a free, hackable agent app with cheap DeepSeek tokens, as long as you treat it as the preview it is.
The app costs nothing, V4.1-Flash keeps running costs low and the plugin design lets you reshape almost everything; DeepSeek itself warns of breaking changes and no security audit, and early users report upgrade bugs.
- Skip it for
- Machines holding sensitive data or credentials, Linux desktop users who want a native app, and anyone who needs ready connectors to Gmail, Drive or GitHub.
- Check first
- Top up a small API balance and try it in a container or virtual machine before you point it at real folders.







