* fix(ai): add icon name aliases and fix multi-path SVG concatenation Add 55+ common icon name aliases (burger→hamburger, sushi→fish, etc.) to both client icon-resolver and server icon API for robust AI-generated icon resolution. Register Lucide's own aliases for broader coverage. Fix SVG path concatenation bug where joining multiple <path> d-values caused incorrect rendering — a standalone <path> treats initial lowercase "m" as absolute, but after concatenation it becomes relative to the previous sub-path endpoint. Now ensures each sub-path starts with absolute "M". Add tryAsyncIconFontResolution for icon_font nodes that miss local lookup — fetches from server API, caches result, and triggers canvas re-render. * fix(canvas): preserve badge/overlay absolute positioning in auto-layout Add isBadgeOverlayNode() detector for badge, indicator, notification-dot, and overlay nodes. These nodes now retain their x/y coordinates instead of being stripped by layout sanitization. Update computeLayoutPositions to exclude badge nodes from the layout flow — they keep absolute positioning and render on top (prepended for correct z-order in reverse iteration). * fix(ai): prevent duplicate canvas objects and fix emoji-to-icon pipeline Streaming path: add ensureUniqueNodeIds before inserting nodes to prevent ID collisions across multiple AI generations. Track newly inserted IDs so subsequent streaming nodes don't collide either. Canvas sync: deduplicate Fabric objects sharing the same penNodeId — keep only the one tracked in objMap, remove stale duplicates. Badge nodes: use shared isBadgeOverlayNode() for z-order insertion and skip x/y stripping in layout parents. Fix emoji-to-icon pipeline: re-run applyIconPathResolution after applyNoEmojiIconHeuristic converts emoji text nodes to path nodes, so the icon resolver can match by name (e.g. "Pizza Emoji Path" → pizza). * fix(canvas): add async icon resolution fallback for icon_font nodes When lookupIconByName fails locally, queue tryAsyncIconFontResolution to fetch from server API. Cache result in ICON_PATH_MAP and trigger canvas re-render via store update. Store iconFontName and iconStyle on Fabric object for sync tracking. * fix(ai): strengthen emoji ban in prompts and improve orchestrator defaults Update all AI prompts to explicitly ban emoji characters with concrete examples and redirect to icon_font nodes instead of the previously incorrect "path nodes" guidance. Add z-order rule to orchestrator prompt: overlay elements must come before content they overlap. Add padding support to OrchestratorPlan rootFrame type. Default mobile root frame gap to 16 for consistent spacing. * feat(electron): add publisher name to Windows build configuration Updated the `electron-builder.yml` to include a publisher name for Windows builds, enhancing the identification of the application during installation. Additionally, revised the README files across multiple languages to reflect the new project description and features, emphasizing OpenPencil as the world's first AI-native open-source vector design tool with concurrent agent teams and design-as-code capabilities. --------- Co-authored-by: Fini <fini.yang@gmail.com>
10 KiB
OpenPencil
The world's first open-source AI-native vector design tool.
Concurrent Agent Teams • Design-as-Code • Built-in MCP Server • Multi-model Intelligence
English · 简体中文 · 繁體中文 · 日本語 · 한국어 · Français · Español · Deutsch · Português · Русский · हिन्दी · Türkçe · ไทย · Tiếng Việt · Bahasa Indonesia
Click the image to watch the demo video
Why OpenPencil
🎨 Prompt → CanvasDescribe any UI in natural language. Watch it appear on the infinite canvas in real-time with streaming animation. Modify existing designs by selecting elements and chatting. |
🤖 Concurrent Agent TeamsThe orchestrator decomposes complex pages into spatial sub-tasks. Multiple AI agents work on different sections simultaneously — hero, features, footer — all streaming in parallel. |
🧠 Multi-Model IntelligenceAutomatically adapts to each model's capabilities. Claude gets full prompts with thinking; GPT-4o/Gemini disable thinking; smaller models (MiniMax, Qwen, Llama) get simplified prompts for reliable output. |
🔌 MCP ServerOne-click install into Claude Code, Codex, Gemini, OpenCode, Kiro, or Copilot CLIs. Design from your terminal — read, create, and modify |
📦 Design-as-Code
|
🖥️ Runs EverywhereWeb app + native desktop on macOS, Windows, and Linux via Electron. Auto-updates from GitHub Releases. |
Quick Start
# Install dependencies
bun install
# Start dev server at http://localhost:3000
bun --bun run dev
Or run as a desktop app:
bun run electron:dev
AI-Native Design
Prompt to UI
- Text-to-design — describe a page, get it generated on canvas in real-time with streaming animation
- Orchestrator — decomposes complex pages into spatial sub-tasks for parallel generation
- Design modification — select elements, then describe changes in natural language
- Vision input — attach screenshots or mockups for reference-based design
Multi-Agent Support
| Agent | Setup |
|---|---|
| Claude Code | No config — uses Claude Agent SDK with local OAuth |
| Codex CLI | Connect in Agent Settings (Cmd+,) |
| OpenCode | Connect in Agent Settings (Cmd+,) |
| GitHub Copilot | copilot login then connect in Agent Settings (Cmd+,) |
Model Capability Profiles — automatically adapts prompts, thinking mode, and timeouts per model tier. Full-tier models (Claude) get complete prompts; standard-tier (GPT-4o, Gemini, DeepSeek) disable thinking; basic-tier (MiniMax, Qwen, Llama, Mistral) get simplified nested-JSON prompts for maximum reliability.
MCP Server
- Built-in MCP server — one-click install into Claude Code / Codex / Gemini / OpenCode / Kiro / Copilot CLIs
- Design automation from terminal: read, create, and modify
.opfiles via any MCP-compatible agent - Layered design workflow —
design_skeleton→design_content→design_refinefor higher-fidelity multi-section designs - Segmented prompt retrieval — load only the design knowledge you need (schema, layout, roles, icons, planning, etc.)
- Multi-page support — create, rename, reorder, and duplicate pages via MCP tools
Code Generation
- React + Tailwind CSS
- HTML + CSS
- CSS Variables from design tokens
Features
Canvas & Drawing
- Infinite canvas with pan, zoom, smart alignment guides, and snapping
- Rectangle, Ellipse, Line, Polygon, Pen (Bezier), Frame, Text
- Boolean operations — union, subtract, intersect with contextual toolbar
- Icon picker (Iconify) and image import (PNG/JPEG/SVG/WebP/GIF)
- Auto-layout — vertical/horizontal with gap, padding, justify, align
- Multi-page documents with tab navigation
Design System
- Design variables — color, number, string tokens with
$variablereferences - Multi-theme support — multiple axes, each with variants (Light/Dark, Compact/Comfortable)
- Component system — reusable components with instances and overrides
- CSS sync — auto-generated custom properties,
var(--name)in code output
Figma Import
- Import
.figfiles with layout, fills, strokes, effects, text, images, and vectors preserved
Desktop App
- Native macOS, Windows, and Linux via Electron
.opfile association — double-click to open, single-instance lock- Auto-update from GitHub Releases
- Native application menu and file dialogs
Tech Stack
| Frontend | React 19 · TanStack Start · Tailwind CSS v4 · shadcn/ui |
| Canvas | Fabric.js v7 |
| State | Zustand v5 |
| Server | Nitro |
| Desktop | Electron 35 |
| AI | Anthropic SDK · Claude Agent SDK · OpenCode SDK · Copilot SDK |
| Runtime | Bun · Vite 7 |
| File format | .op — JSON-based, human-readable, Git-friendly |
Project Structure
src/
canvas/ Fabric.js engine — drawing, sync, layout, guides, pen tool
components/ React UI — editor, panels, shared dialogs, icons
services/ai/ AI chat, orchestrator, design generation, streaming
services/figma/ Figma .fig binary import pipeline
services/codegen React+Tailwind and HTML+CSS code generators
stores/ Zustand — canvas, document, pages, history, AI, settings
variables/ Design token resolution and reference management
mcp/ MCP server tools for external CLI integration
uikit/ Reusable component kit system
server/
api/ai/ Nitro API — streaming chat, generation, validation
utils/ Claude CLI, OpenCode, Codex, Copilot client wrappers
electron/
main.ts Window, Nitro fork, native menu, auto-updater
preload.ts IPC bridge
Keyboard Shortcuts
| Key | Action | Key | Action | |
|---|---|---|---|---|
V |
Select | Cmd+S |
Save | |
R |
Rectangle | Cmd+Z |
Undo | |
O |
Ellipse | Cmd+Shift+Z |
Redo | |
L |
Line | Cmd+C/X/V/D |
Copy/Cut/Paste/Duplicate | |
T |
Text | Cmd+G |
Group | |
F |
Frame | Cmd+Shift+G |
Ungroup | |
P |
Pen tool | Cmd+Shift+E |
Export | |
H |
Hand (pan) | Cmd+Shift+C |
Code panel | |
Del |
Delete | Cmd+Shift+V |
Variables panel | |
[ / ] |
Reorder | Cmd+J |
AI chat | |
| Arrows | Nudge 1px | Cmd+, |
Agent settings | |
Cmd+Alt+U |
Boolean union | Cmd+Alt+S |
Boolean subtract | |
Cmd+Alt+I |
Boolean intersect |
Scripts
bun --bun run dev # Dev server (port 3000)
bun --bun run build # Production build
bun --bun run test # Run tests (Vitest)
npx tsc --noEmit # Type check
bun run electron:dev # Electron dev
bun run electron:build # Electron package
Contributing
Contributions are welcome! See CLAUDE.md for architecture details and code style.
- Fork and clone
- Create a branch:
git checkout -b feat/my-feature - Run checks:
npx tsc --noEmit && bun --bun run test - Commit with Conventional Commits:
feat(canvas): add rotation snapping - Open a PR against
main
Roadmap
- Design variables & tokens with CSS sync
- Component system (instances & overrides)
- AI design generation with orchestrator
- MCP server integration with layered design workflow
- Multi-page support
- Figma
.figimport - Boolean operations (union, subtract, intersect)
- Multi-model capability profiles
- Collaborative editing
- Plugin system
Contributors
Community
Star History
License
MIT — Copyright (c) 2026 ZSeven-W