DeepSeek Harness v0.2 Ships Official Desktop Apps for Its Open-Source Agent Harness

According to MarkTechPost, a harness is the runtime that turns a model into an agent, reading files, running commands and keeping a plan. The v0.2 preview targets everyday work as well as coding. Built-in capabilities include commonly used tools for office work and development; a new plugin-management page for installing, configuring, enabling and disabling plugins; and a right-hand sidebar that previews generated files and code changes. Users can send documents, spreadsheets or PDFs and ask for charts or slides. An Automation Task plugin schedules recurring prompts, with run history and editable frequency. The desktop build bundles the dsh command, so users no longer need a separate Node.js or pnpm install. They can sign in with a DeepSeek account or add an API key; according to rc.1 notes, account models get web search without an extra key.

DeepSeek first open-sourced dsh in August 2026 under MIT. MarkTechPost reported it runs on Cordis, described in the research paper "A Programming Paradigm for Spatiotemporal Composability." The model adapter, tool registry and agent loop are all plugins, and any layer can be replaced. The harness is not locked to DeepSeek models; a 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 doubles as DeepSeek's evaluation rig. The API changelog says Code Agent benchmarks ran on "DeepSeek Harness minimal mode," where DeepSeek-V4-Flash-0731 scored 82.7 on Terminal Bench 2.1 in that setup.

Sessions run in one of four modes, per 36Kr's hands-on and the release notes. Standard is for general work and is now the built-in default. Creator lets a user describe a plugin in chat, after which the agent builds and installs it. PTC, or programmatic tool calling, runs Node code in a separate process. Minimal is the lean loop DeepSeek uses for benchmark runs.

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 and 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.

MarkTechPost contrasted the release with Claude Code and OpenAI Codex. DeepSeek Harness is MIT-licensed, while Claude Code is proprietary and Codex is Apache-2.0. DeepSeek Harness is a developer preview; the other two are generally available. It offers desktop app, Web UI and headless CLI surfaces; Claude Code offers terminal, IDE and GitHub; Codex offers CLI, IDE extension, desktop app and Codex Web. Model access for DeepSeek Harness comes through a DeepSeek account or API key, plus custom OpenAI-compatible providers, while Claude Code uses Anthropic Claude models and Codex uses ChatGPT plan sign-in or an OpenAI API key. Its extension model treats everything as a plugin, including the agent loop; Claude Code uses plugins with custom commands and agents, and Codex uses MCP servers, plugins and skills.