{"slug": "deepseek-harness-download-where-to-get-it-how-to-install-it-on-windows-mac-and", "title": "DeepSeek Harness download: where to get it, how to install it on Windows, Mac and Linux, and what it costs", "summary": "DeepSeek released its open-source AI agent DeepSeek Harness, or dsh, as a free download at version 0.2.0-rc.2, available as a desktop app for Apple silicon Macs and 64-bit Windows and via the command \"npx @deepseek-ai/dsh web\" on Linux, Intel Macs and other platforms. The MIT-licensed developer preview, first shipped in August 2026, is free but requires a paid DeepSeek API key, with DeepSeek V4.1 Flash priced at $0.30 per million input tokens ($0.006 cached) and $1.20 per million output tokens, and DeepSeek V4 Pro at $1.32 ($0.044 cached) and $3.96, halving off-peak. DeepSeek warns of \"compatibility-breaking changes\" and pitches the Cordis-based agent for file editing, command execution, bug fixing and everyday tasks such as spreadsheets and slides.", "body_md": "DeepSeek Harness’s official download page. Image: DeepSeek\n\nDeepSeek Harness is free to download, but only from DeepSeek itself: the [official page](https://www.deepseek.com/en/harness/) offers a desktop app for Apple silicon Macs and 64-bit Windows, and the open-source code is on [DeepSeek’s GitHub](https://github.com/deepseek-ai/deepseek-harness). On Linux, Intel Macs or anywhere else, you run it in your browser with one command. The software costs nothing, but you pay for the AI model it uses, usually through a DeepSeek API key.\n\n*Last checked: October 1, 2026 (version 0.2.0-rc.2).*\n\n## What is DeepSeek Harness?\n\nDeepSeek Harness, or `dsh`, is an open-source AI agent from the Chinese lab DeepSeek. Like Claude Code or OpenAI’s Codex, it can read and edit files, run commands, fix bugs, run tests and work through a plan, but DeepSeek also pitches it for everyday work such as analysing spreadsheets, drafting documents and making slides.\n\nIts big idea is that “everything is a plugin”: the tools, the interface and even the agent itself are plugins you can swap, add or write yourself. It is built on a framework called Cordis, which DeepSeek describes in a [research paper](https://arxiv.org/abs/2608.25512). It came out as a developer preview in August 2026 under the MIT licence, and DeepSeek warns in capitals that “there will be compatibility-breaking changes”.\n\n## DeepSeek Harness download: the official links\n\nSearch for “DeepSeek Harness download” and several unofficial sites with the product’s name in their address appear near the top, and some of them describe themselves as unofficial. Agent software can run commands and touch your files, so only install it from these official places:\n\n| What | Official source | \n|---|---|\n| Desktop app (Mac, Windows) | [deepseek.com/en/harness](https://www.deepseek.com/en/harness/) , which links to download.deepseek.com | \n| Source code | [github.com/deepseek-ai/deepseek-harness](https://github.com/deepseek-ai/deepseek-harness) | \n| npm package | [@deepseek-ai/dsh](https://www.npmjs.com/package/@deepseek-ai/dsh) | \n| Documentation | [deepseek-harness.github.io](https://deepseek-harness.github.io/deepseek-harness/) | \n\nThere is no DeepSeek Harness app for iPhone, iPad or Android, and no official VS Code extension.\n\n## How to install DeepSeek Harness on Windows, Mac and Linux\n\n| Platform | Desktop app | Browser version | \n|---|---|---|\n| Windows 10/11 (64-bit) | Yes | Yes | \n| Mac with Apple silicon (M1 and later) | Yes | Yes | \n| Intel Mac | No | Yes | \n| Linux, including WSL | No | Yes | \n| iPhone, Android | No | No | \n\n**Desktop app:** download the installer from the official page, open it and follow the prompts. The app is the full web interface in a desktop window and updates itself.\n\n**Browser version (any computer):** install [Node.js](https://nodejs.org/) version 22.19 or later, open a terminal in the folder you want it to work in and run:\n\n```\nnpx @deepseek-ai/dsh web\n```\n\nThat starts DeepSeek Harness on your own machine and opens it at `http://127.0.0.1:3080`. Nothing is hosted online; the address only works on your computer.\n\n**First steps:** open Settings, then Models, and paste in an API key from the [DeepSeek platform](https://platform.deepseek.com/). Then choose a workspace (the folder it’s allowed to work in), start a session and give it a task, such as “Summarise this repository”.\n\n## Is DeepSeek Harness free?\n\nThe software is free and open source. What you pay for is the model. By default it uses DeepSeek’s own API, which is pay as you go. DeepSeek’s [price list](https://api-docs.deepseek.com/quick_start/pricing) (per million tokens, standard rates) is:\n\n| Model | Input | Output | \n|---|---|---|\n| DeepSeek V4.1 Flash (the default) | $0.30 ($0.006 if cached) | $1.20 | \n| DeepSeek V4 Pro | $1.32 ($0.044 if cached) | $3.96 | \n\nPrices halve off-peak. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays, excluding Chinese public holidays.\n\nAgents use far more tokens than a chat, because they reread files and tool output on every step, so set a spending limit on your API account. If you want to try models without paying, our guide to the [free DeepSeek, Kimi and GLM API through NVIDIA](https://madrobot.blog/2026/09/26/free-deepseek-kimi-glm-api-nvidia/) explains one route, and you can plug that endpoint in as a custom model.\n\n## Can you use DeepSeek Harness with Claude, Kimi, GLM or a local model?\n\nYes. Under Settings, then Models, choose “Add model provider”. [Built-in providers](https://github.com/deepseek-ai/deepseek-harness/blob/main/docs/user/guide/providers.md) include Anthropic (Claude), OpenAI, Moonshot AI (Kimi) and Z.ai (GLM), and each needs that company’s API key. Signing in with a subscription instead of a key, as you would with Codex or a Claude plan, isn’t supported yet.\n\nFor anything else, choose “Custom model API” and enter a base URL that speaks the OpenAI Chat Completions, OpenAI Responses or Anthropic Messages format. Local tools such as Ollama, LM Studio and llama.cpp’s server offer an OpenAI-style endpoint, so you can point it at a model running on your own computer. Small local models struggle with agent work, though, so expect weaker results.\n\n## DeepSeek Harness CLI, terminal and plugins\n\nThe `dsh` command does more than open the web interface. The [CLI documentation](https://github.com/deepseek-ai/deepseek-harness/blob/main/apps/cli/README.md) lists modes including:\n\n- **Headless:**`dsh --profile headless \"run the tests\"` runs one job, prints the answer and exits, which suits scripts.\n- **ACP:**`dsh --profile acp` serves editors and other tools that speak the Agent Client Protocol.\n- **SDK:** a Python SDK and a JSON-RPC mode for building it into your own apps.\n- **Plugins:**`dsh plugin` manages a profile’s plugins.\n\nA full-screen terminal interface (TUI), like Claude Code’s, isn’t installed by default; the docs treat it as an optional profile. DeepSeek’s own plugins include agent teams, subagents, scheduled tasks, a terminal and voice input, several marked experimental. Community plugins are tagged [dsh-plugin](https://github.com/topics/dsh-plugin) on GitHub, and a “Creator mode” lets the agent write a plugin for you from a chat.\n\n## Is DeepSeek Harness safe?\n\nDeepSeek is unusually blunt about this. Its [safety notice](https://github.com/deepseek-ai/deepseek-harness/blob/main/SAFETY.md) says the software “has not undergone a security audit and must not be treated as secure or production-ready”, and that bad model output or untrusted plugins “may damage the host computer, modify or delete files, disclose data or credentials”. It recommends running it in a disposable virtual machine or container, keeping backups and reviewing commands before approving them. Our explainer on [AI sandboxes](https://madrobot.blog/2026/09/27/what-is-an-ai-sandbox-why-ai-agents-escape/) covers why that matters.\n\nOn privacy, DeepSeek’s [privacy policy](https://www.deepseek.com/en/harness/privacy/) says it collects and stores personal data “in People’s Republic of China”, including what you type and upload when you use its service. If you choose another model provider, the policy says your prompts go straight to that provider and DeepSeek Harness “does not store or process your Input and Output on the server side”.\n\n## Is DeepSeek Harness available in my country?\n\nDeepSeek says the preview is available “worldwide”, and because it runs on your own computer there is no country check in the software. People in the US, UK, Canada, Australia, Germany, India and the rest of Europe can install it the same way. What can block you is the model: you need a DeepSeek API account (or another provider’s) that works where you are. Some governments and many employers bar DeepSeek’s apps on official or work devices, so check your organisation’s rules before installing it on a work machine.\n\n## Frequently asked questions\n\n### Is DeepSeek Harness free?\n\nYes, the software is free and open source under the MIT licence. You pay for the AI model it uses, normally DeepSeek’s pay-as-you-go API, unless you connect a free or local model.\n\n### Where is the official DeepSeek Harness download?\n\nOnly on deepseek.com/en/harness (which links to download.deepseek.com), the deepseek-ai/deepseek-harness repository on GitHub, and the @deepseek-ai/dsh package on npm. Other sites with the name in their address are not DeepSeek’s.\n\n### Is there a DeepSeek Harness desktop app for Linux?\n\nNo. The desktop app is only for Apple silicon Macs and 64-bit Windows. On Linux, WSL or an Intel Mac, run npx @deepseek-ai/dsh web and use it in your browser.\n\n### Does DeepSeek Harness work with Claude?\n\nYes, with an Anthropic API key added under Settings, then Models. Signing in with a Claude subscription isn’t supported.\n\n### Can DeepSeek Harness use a local model with Ollama or LM Studio?\n\nYes. Add a Custom model API with your local server’s OpenAI-compatible address. Small local models tend to do poorly at agent tasks.\n\n### Is DeepSeek Harness open source?\n\nYes. The full code is on GitHub under the MIT licence.\n\n### Is DeepSeek Harness better than Claude Code or OpenCode?\n\nIt’s too early to say. It is a developer preview that changes quickly, and its strength is plugins and everyday tasks beyond coding. Claude Code and Codex are more mature; try it on a spare project first.\n\n### Is DeepSeek Harness safe to use?\n\nDeepSeek says it hasn’t been security audited and can delete files or leak credentials if something goes wrong. Run it in a virtual machine or container, keep backups and review commands before approving them.\n\n### Is there a DeepSeek Harness iOS or Android app?\n\nNo. It runs on computers only, as a desktop app or in a browser on the machine running it.\n\n*Sources: [DeepSeek Harness official page](https://www.deepseek.com/en/harness/), [DeepSeek Harness on GitHub](https://github.com/deepseek-ai/deepseek-harness) (README, safety notice, model and CLI guides), [npm](https://www.npmjs.com/package/@deepseek-ai/dsh), [DeepSeek API pricing](https://api-docs.deepseek.com/quick_start/pricing), [DeepSeek Harness privacy policy](https://www.deepseek.com/en/harness/privacy/).*", "url": "https://wpnews.pro/news/deepseek-harness-download-where-to-get-it-how-to-install-it-on-windows-mac-and", "canonical_source": "https://madrobot.blog/2026/10/01/deepseek-harness-download-install-windows-mac-linux-free/", "published_at": "2026-10-01 01:12:00+00:00", "updated_at": "2026-10-01 01:18:34.494607+00:00", "lang": "en", "topics": ["ai-agents", "ai-tools", "ai-products", "developer-tools", "artificial-intelligence"], "entities": ["DeepSeek", "DeepSeek Harness", "dsh", "Cordis", "DeepSeek V4.1 Flash", "DeepSeek V4 Pro", "Node.js", "GitHub"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/deepseek-harness-download-where-to-get-it-how-to-install-it-on-windows-mac-and", "markdown": "https://wpnews.pro/news/deepseek-harness-download-where-to-get-it-how-to-install-it-on-windows-mac-and.md", "text": "https://wpnews.pro/news/deepseek-harness-download-where-to-get-it-how-to-install-it-on-windows-mac-and.txt", "jsonld": "https://wpnews.pro/news/deepseek-harness-download-where-to-get-it-how-to-install-it-on-windows-mac-and.jsonld"}}