MIT 许可的 Rust AI 矢量设计工具，支持提示词生成画布设计、内置 MCP 服务器与 `op` CLI；README 未记载专门的 DeepSeek Harness 集成，DeepSeek 是内置模型供应商之一。
DSH 适配
生态相关
作者声明
安全审计
未审计
最后核验
2026-08-21
许可证
MIT
01它能帮你完成什么？
Generate UI and page designs from natural-language prompts on a live infinite canvas
Vector designs on the canvas in real-time with streaming animation; `.op` files (JSON, Git-friendly) plus code export to React + Tailwind, HTML + CSS, Vue, Svelte, Flutter, SwiftUI, Jetpack Compose, React Native
Designers and developers who want AI-generated UI as inspectable, diffable Design-as-Code files
Drive design automation from any MCP-compatible coding agent via the built-in MCP server
Read, create, and modify `.op` files through MCP tools — including a layered workflow (`design_skeleton` → `design_content` → `design_refine`), style guide tools, and multi-page support
Agent users who want a coding agent CLI to read, create, and modify design files from the terminal
Run batch design automation from the terminal with the `op` CLI
Batch designs from files or stdin (`op design @landing.txt`, `cat design.dsl | op design -`), node insertion, Figma import, and a headless file-backed server mode
Developers scripting design generation in CI or terminal workflows
02如何接入 DeepSeek Harness？
前置条件
- Platform package manager: Homebrew (macOS), Scoop (Windows), or direct download from GitHub Releases; Nix (Linux x86_64) is an alternative lane
- For the `op` CLI only: Homebrew on macOS/Linux, or the install script (macOS / Linux) / PowerShell script (Windows)
安装步骤
- 01
macOS: `brew tap zseven-w/openpencil && brew install --cask openpencil`
$ brew tap zseven-w/openpencil && brew install --cask openpencil
- 02
Windows: `scoop bucket add openpencil https://github.com/zseven-w/scoop-openpencil && scoop install openpencil`
- 03
Linux / Windows direct download: GitHub Releases — `.exe` (Windows), `.AppImage` / `.deb` (Linux)
- 04
CLI (`op`): `brew install zseven-w/openpencil/op`, or `curl -fsSL https://raw.githubusercontent.com/ZSeven-W/openpencil/main/scripts/install-op.sh | bash` (Windows: `irm https://raw.githubusercontent.com/ZSeven-W/openpencil/main/scripts/install-op.ps1 | iex`)
$ brew install zseven-w/openpencil/op
- 05
Optional Docker web host: `docker run -d -p 3100:3100 ghcr.io/zseven-w/openpencil-web:v${VERSION}` then open `http://localhost:3100/`
- 06
Optional — teach AI agents to design with `op`: install the OpenPencil Skill plugin, then run `op install` for detected agents (or `op install --target codex`)
验证接入成功
作者未说明
03DSH 适配与能力边界
Standalone AI-native vector design tool; no DeepSeek Harness/dsh integration is documented in the README — the only DeepSeek mention is as one of 9+ built-in model providers in the app's provider presets
Prompt → Canvas generation with Concurrent Agent Teams
natural-language prompts, optional screenshots/mockups (vision input), style guides→real-time vector UI on the infinite canvas; an orchestrator decomposes complex pages into spatial sub-tasks generated by multiple agents in parallel
AI requests are sent to the configured model provider over the network; browser-entered provider credentials are kept in same-origin localStorage by defaultBuilt-in MCP server (`op-mcp`)
any MCP-compatible agent CLI (one-click install into Claude Code / Codex / OpenCode / Kiro / Copilot CLIs)→MCP tools to read, create, and modify `.op` files — layered design workflow, segmented prompt retrieval, style guide tools, multi-page support
stdio transport via the desktop binary (`--mcp <path>`), plus a live HTTP endpoint at `127.0.0.1:<port>/mcp` from the running appDesign-as-Code: `.op` format and multi-platform code export
`.op` design files (JSON — human-readable, Git-friendly, diffable)→code export to React + Tailwind, HTML + CSS, Vue, Svelte, Flutter, SwiftUI, Jetpack Compose, React Native; design variables become CSS custom properties
Multi-model intelligence with DeepSeek among 9+ built-in providers
provider presets with region switcher — Anthropic, OpenAI, Google, DeepSeek, and more→prompts, thinking mode, and timeouts automatically adapted per model tier (full-tier, standard-tier, basic-tier)
04适合谁？何时不该用？
适合
- Designers and developers who want AI-generated UI as inspectable, diffable Design-as-Code files
- Agent users who want a coding agent CLI to read, create, and modify design files from the terminal
- Developers scripting design generation in CI or terminal workflows
不适合
- The README documents no DeepSeek Harness/dsh integration: MCP one-click install targets Claude Code, Codex, OpenCode, Kiro, and Copilot CLIs, and the LLM Skill's explicit install example targets codex. DeepSeek appears only as a model provider preset inside the app.
05兼容性、维护与安全提示
- The README documents no DeepSeek Harness/dsh integration: MCP one-click install targets Claude Code, Codex, OpenCode, Kiro, and Copilot CLIs, and the LLM Skill's explicit install example targets codex. DeepSeek appears only as a model provider preset inside the app.
- Nix builds are currently published for x86_64-linux only and do not produce a `.deb`; a Debian package requires the upstream release artifacts.
- The Windows installer has been flagged by antivirus as a known heuristic false positive on new, low-prevalence releases; the README points to a download-verification doc.
MIT · actively maintained (latest release v0.8.4, 2026-08-11)
06常见问题
OpenPencil 能配合 DeepSeek Harness 使用吗？
README 未记载专门的 DeepSeek Harness 集成。OpenPencil 提供内置 MCP 服务器和 `op` CLI，面向兼容 MCP 的 agent（一键安装目标为 Claude Code、Codex、OpenCode、Kiro 和 Copilot CLI）。DeepSeek 在应用内作为 9+ 内置模型供应商预设之一出现。
如何安装 OpenPencil？
macOS：`brew tap zseven-w/openpencil` 后执行 `brew install --cask openpencil`。Windows：`scoop bucket add openpencil https://github.com/zseven-w/scoop-openpencil` 后执行 `scoop install openpencil`。Linux：从 GitHub Releases 下载 `.AppImage` / `.deb`，或在 x86_64 上用 Nix。`op` CLI 可通过 `brew install zseven-w/openpencil/op` 或安装脚本安装。
AI 编码 agent 如何驱动 OpenPencil？
通过内置 MCP 服务器（`op-mcp` crate）——桌面二进制以 stdio 方式运行（`--mcp <path>`），或运行中的应用在 `127.0.0.1:<port>/mcp` 提供 HTTP 端点，无需 Node.js——以及支持批量设计 DSL 的 `op` CLI。OpenPencil Skill 插件可教 agent 用 `op` 做设计（`op install`）。
内置了哪些模型供应商？
9+ 家供应商，带区域切换的预设——Anthropic、OpenAI、Google、DeepSeek 等——还支持 Claude Code（本地 OAuth）、Codex CLI、OpenCode 和 GitHub Copilot 作为 agent 运行时。提示词和思考模式会按模型档位自动适配。
输出格式是什么？
`.op` 文件——JSON 格式，人类可读、对 Git 友好、可 diff。从一个 `.op` 文件可导出 React + Tailwind、HTML + CSS、Vue、Svelte、Flutter、SwiftUI、Jetpack Compose 和 React Native 代码，画布还支持导出 PNG/JPEG/WEBP/PDF。
07相关的 DSH 工作流
ruflo
作者 ruvnet
原版智能体元框架（meta-harness）：支持多智能体集群部署、自主工作流编排与对话式 AI 系统构建，内置自适应记忆、自学习智能与 RAG 检索，原生集成 Claude Code / Codex / Hermes 等大量工具。
reactive-resume
作者 amruthpillai
注重隐私的开源简历生成器，安全、可定制、可移植，永久免费，支持自托管。
everos
作者 evermind-ai
为所有 AI 智能体提供的便携统一记忆层，本地优先、Markdown 原生、数据归用户所有，跨应用与工作流自进化。
yao
作者 yaoapp
在桌面、移动端、浏览器或 API 上统一管理你的智能体与工作区，看板式任务追踪，支持自托管。
08数据与来源
Select from provider presets with region switcher — Anthropic, OpenAI, Google, DeepSeek, and more
Design automation from terminal: read, create, and modify `.op` files via any MCP-compatible agent
页面基于项目公开文档、仓库元数据和 DSH Plugins 的结构化解析生成；最后核验于 2026-08-21。发现错误？提交更正。
最佳 DeepSeek Harness 插件
从全目录挑出的 12 个值得优先安装的插件，覆盖各个分类。