DeepSeek has released an official desktop application for its open-source agent harness, DeepSeek Harness, marking the arrival of version 0.2. The installer supports macOS on Apple silicon and Windows on 64-bit systems. Users can download the software from deepseek.com/harness or execute the command npx @deepseek-ai/dsh web to run the web interface. The company warns that compatibility-breaking changes will follow in future updates.
Features in the latest update
The software acts as a runtime that converts a model into an agent. It reads files, executes commands, and maintains a plan. Version 0.2 targets everyday work as well as coding tasks.
- Built-in capabilities: Common tools for office work and development come preloaded.
- Plugin management: A new page allows users to install, configure, enable, and disable plugins.
- File and diff review: A right-hand sidebar previews generated files and code changes.
- Document work: Users can send documents, spreadsheets, or PDFs and request charts or slides.
- Automation Task: A plugin schedules recurring prompts, displaying run history and allowing editable frequency.
The desktop build includes the dsh command. Users no longer need a separate Node.js or pnpm installation. Sign-in uses a DeepSeek account or an API key. According to the rc.1 notes, account models provide web search without an extra key.
Architecture based on plugins
DeepSeek first open-sourced dsh in August 2026 under the MIT licence. It runs on Cordis, as described in the research paper ‘A Programming Paradigm for Spatiotemporal Composability’. The model adapter, tool registry, and agent loop are all plugins. Any layer can be replaced.
The harness is not locked to DeepSeek models. The provider guide covers third-party providers and custom OpenAI-compatible endpoints. The repository has passed 240,000 GitHub stars and 29,000 forks.
dsh also functions as DeepSeek’s evaluation rig. The API changelog states Code Agent benchmarks ran on ‘DeepSeek Harness minimal mode.’ DeepSeek-V4-Flash-0731 scored 82.7 on Terminal Bench 2.1 in that setup.
Four session modes
Sessions run in one of four modes, per 36Kr’s hands-on report and the release notes:
- Standard: General work, and now the built-in default.
- Creator: Describe a plugin in chat; the agent builds and installs it.
- PTC: Programmatic tool calling, with Node code run in a separate process.
- Minimal: The lean loop DeepSeek uses for benchmark runs.
Experimental compatibility layer
The v0.2.1-alpha.1 build adds an experimental Claude Code Mods compatibility layer. DeepSeek frames it as a test of whether the Mods API is broadly a subset of dsh plugins. It does not promise full compatibility. The build also adds a ‘Let Agent create a plugin’ entry, an optional Developer Tools bundle, and a –public-url flag for reverse proxies. It removes runtime invariant plugins, a breaking change for some extensions.
Comparison table
| Feature | DeepSeek Harness v0.2 | Claude Code | OpenAI Codex |
|---|---|---|---|
| Developer | DeepSeek | Anthropic | OpenAI |
| License | MIT | Proprietary | Apache-2.0 |
| Status | Developer preview | Generally available | Generally available |
| Surfaces | Desktop app (macOS Apple silicon, Windows x64), Web UI, headless CLI | Terminal, IDE, GitHub (@claude) | CLI, IDE extension, desktop app, Codex Web |
| Model access | DeepSeek account or API key, plus custom OpenAI-compatible providers | Anthropic Claude models | ChatGPT plan sign-in or OpenAI API key |
| Extension model | Everything is a plugin, including the agent loop | Plugins with custom commands and agents | MCP servers, plugins, skills |
| Quick start | npx @deepseek-ai/dsh web | curl -fsSL https://claude.ai/install.sh | bash | npm install -g @openai/codex |
What it means
Developers can now manage agents directly from their operating system without relying solely on a terminal. The ‘Creator’ mode is particularly notable because it allows the agent to write and install its own plugins, reducing manual configuration. Users are not forced to use DeepSeek models, as the system accepts custom OpenAI-compatible endpoints. However, the software remains a developer preview, meaning the interface and functionality will likely change between builds.
