Best local-first Claude Design alternative
SankiWork is the local-first vibe design workspace — it turns the coding agents you already run into a design engine that carries you from idea to prototype, web, slides, and HTML video, all finished on your own machine. Agent-native, with 21 coding agents and 152 design systems.

Why SankiWork?
Brief → Template → Visual direction → Artifact → Memory
Describe your goal in one line, or start from a template / plugin.
Once a direction is set, palette, type, and spacing flow into generation automatically.
The agent reads all context, produces real runnable files, and previews and edits them live in a sandbox.
Export it to engineering to keep building, or turn it into a marketing video with HyperFrames.Prototypes, landing pages, slide decks, dashboards, brand and design systems — even HTML video. SankiWork turns each into real, runnable files, driven by your own coding agent and ready to ship or hand off to engineering.
Every stage is iterative, visual, and research-driven — composable files, not opaque prompts.


Local-first · Agent-native
Local-first · Free
SankiWork is the official AI design workspace from the Sanki5211/sankiwork-community project. It turns a local coding agent — Claude Code, Codex, Cursor, Gemini CLI, OpenCode, or Qwen — into a design engine driven by composable skills and portable DESIGN.md systems.
Yes. The canonical project lives at https://sanki-ai.cloud/ and the source is on GitHub at https://gitea.com/Sanki5211/sankiwork-community. "SankiWork", "SankiWork", "sankiwork", and "SankiWork AI" all refer to this same project.
Claude Design is a hosted product tied to a single vendor. SankiWork is local-first and BYOK: you bring your own agent, credentials, and DESIGN.md system.
A vibe design workspace is where you design by describing intent to an AI agent — from prompt to prototype, web page, slides, or HTML video — instead of hand-placing every element. SankiWork is a local-first, agent-native vibe design workspace: it wires the coding agent you already run into a full design workflow, so one tool takes you from a rough idea to production-ready output you own.
Yes. The desktop app, daemon, and skill runtime run on your machine. Generated artifacts land in your project directory instead of being forced through a vendor cloud.
SankiWork ships 17 first-party BYOK adapters out of the box: Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot CLI, Grok, Hermes, Kimi, Devin for Terminal, OpenCode, Qwen, DeepSeek, Pi, Mistral Vibe, Kiro, Kilo, and Qoder. Any adapter that speaks the same skill protocol works — switching agents is a config change, not a redesign.
Yes. You can fork the repo, edit skills, add your own DESIGN.md systems, or run the daemon on your own machines.
Only your prompt and skill context goes to whichever provider you bring keys for (BYOK). SankiWork has no server of its own — the daemon talks to your provider directly. Generated artifacts land as files in your project directory, not in any vendor cloud.
Not today. SankiWork is local-first by design — the minimum is a local daemon plus an agent (Claude Code, Codex, Cursor, Gemini CLI, or one of the 17 supported adapters). A hosted sandbox is on the roadmap but not the priority: artifacts in your repo beat documents in someone else's database.
The product is free — there is no SankiWork subscription. You pay the API costs of whichever provider you use (Anthropic, OpenAI, Google, Mistral, xAI, Moonshot, etc.), billed directly to your account. BYOK keeps both the credentials and the spend on your side of the line.
Yes. The daemon runs anywhere Node 24 runs, and the landing page is a static Astro build that deploys to Cloudflare Pages, Vercel, or Netlify as-is. Teams running shared deployments typically pin the daemon to a machine inside their network and point each developer's CLI at it.
Drop a screenshot or a Figma export into the web UI and ask your agent to extract a brand into a DESIGN.md file. Save that file under design-systems/<your-brand>/ in your repo; every skill then renders in that brand without re-prompting. /alternatives/claude-design/ describes the same flow in step form.
Yes. Skills and DESIGN.md systems are agent-agnostic — the same SKILL.md file renders against Claude Code, Codex, Cursor, Gemini CLI, GitHub Copilot, Grok, Hermes, Qwen, or any other supported adapter. Switching agents is a config change in the daemon, not a redesign.
The live roadmap is at docs/roadmap.md in the GitHub repo, and weekly release notes ship through GitHub Releases. Major themes for the next quarter: more agent adapters, richer template families (3D, video, audio), and an optional shared-daemon mode for design teams.