DeepSeek

Yes. DeepSeek Harness is released under the MIT license and its full source is published in the deepseek-ai/deepseek-harness repository. Third-party dependencies and their licenses are disclosed separately in THIRD_PARTY_NOTICES.md.

Node.js 22.19 or newer, or 24 and above. Run npx @deepseek-ai/dsh web and the Web UI is served at http://127.0.0.1:3080 by default. Building from a source checkout additionally needs Corepack-enabled pnpm 11.7.0 and Git 2.26 or newer.

Not yet. DeepSeek Harness is explicitly in developer preview and iterating rapidly, and the repository warns that there will be compatibility-breaking changes. Its core plugins and APIs are still expected to evolve.

DeepSeek through a single API-key field, catalog providers such as Anthropic and OpenAI, and any custom OpenAI-compatible endpoint you configure with a provider ID, base URL, protocol, credential, and model list. Bedrock, Vertex, Azure, and Codex require their own native credentials instead of an API key.

Standard mode is the most complete starting point. Use Code mode for programmatic tool composition, Minimal mode for controlled benchmarks, and Creator mode for plugin experiments and preset authoring.

Start with the runtime mode closest to the job, then replace individual model, tool, skill, sandbox, storage, or UI plugins through configuration. Run dsh web --dump-config to print the composed tree, then patch any row it prints.

The plugin architecture is designed to make models, tools, skills, sessions, sandboxes, storage, loops, scheduling, and UI independently replaceable through configuration alone.

They share one append-only event stream containing prompts, model output, tool calls, results, subagent events, and context injections. Model-visible means logged, so anything reaching a model request is reconstructable from that stream.

Not currently. The CLI accepts --host and --port, but it intentionally does not support --host 0.0.0.0 yet and exits with a usage error. Repeatable --trusted-host adds named authorities accepted by the browser-trust fence.

, which forwards pnpm verbs to the profile directory. Plugin authors should add the dsh-plugin topic to their repository so others can discover the plugin on GitHub.

DeepSeek 是一个开源的大型语言模型（LLM），专注于提供高性能和可扩展的自然语言处理能力，适用于多种 AI 应用场景。

DeepSeek 可用于文本生成、对话系统、代码编写、数据分析以及多语言翻译等场景，特别适合需要定制化和高灵活性的 AI 应用。

DeepSeek 基于先进的深度学习架构，支持大规模参数量，并具备良好的推理能力和多任务处理能力，同时提供开源代码以便开发者进行二次开发和优化。