artisign

An artificial artisan for UI design: agents build screens, a design system keeps them consistent.

Local-first UX design tool for AI agents. Agents design through MCP, the human reviews in the browser, state lives in plain files.

One Node process on 127.0.0.1. No database, no accounts, no hosting.

Status: Stable (2.2.0) — CLI, store, parser/model, the 25-tool MCP surface, browser preview with live reload/flows/comments, screen variants with side-by-side compare, the design system (tokens, variants, drift warnings, promote_to_system), and packaging are all implemented. artisign designs its own browser preview — see Designed with itself.

Why

1. Consistency across screens — the design system is first-class; screens store refs, never inline values
2. Token efficiency — tiered reads, field selection, diff mode on writes
3. Flows as first-class — click routes are data, not documentation
4. Render determinism — the preview renders the source, so preview equals output by construction

Install

Requirements. Node.js ≥ 22.4 on macOS or Linux — Windows is not supported (the store's atomic writes and path handling have never been verified there, so package.json makes npm install refuse the platform). For screenshots, a browser that is already installed: Chrome, Edge, Brave, Chromium or Firefox — see Screenshots below.

npm install -g artisign
artisign

1. npm install -g artisign — once.
2. artisign — starts artisign's background process (the daemon), opens the preview in your browser and, on first run, offers to connect your AI tool. After connecting, quit and reopen the AI tool so it picks up the new entry.
3. Ask your AI tool to create or open a project: "Create an artisign project in ~/work/acme". The project folder (and any missing parent folders) is created for you; the AI tool passes an absolute path, since artisign does not expand ~ itself. Writes are committed to git automatically (autoCommit, on by default) — artisign runs git init in the project on the first write, needs git installed for that, and skips the commit rather than fail when it is missing or the folder sits inside another repo. No git identity is required: on a machine where git has no user.name/user.email, artisign's commits are made as artisign <noreply@artisign.dev>; an identity you have configured (git config, or GIT_AUTHOR_*/GIT_COMMITTER_* in the environment) is used as it is, and artisign never writes to your git config.

Check it works. Ask your AI tool "List my artisign projects" — it calls list_projects and the project you just created appears, and in the preview's project picker too.

Two copies, one version. The global install is the artisign command you type. Your AI tool's entry runs the same version, pinned, through npx (downloaded once on its first launch) as a small stdio bridge; artisign update keeps the two in step. Whichever starts first — you with artisign, or the bridge when your AI tool connects — starts the daemon; both use the same ~/.artisign and port, so there is only ever one. The daemon inherits the environment of the process that started it: to make a variable such as ARTISIGN_BROWSER stick, run artisign from your shell before your AI tool does (artisign stop first if it is already running); putting it in the entry's env also works, but artisign setup rewrites entries, so add it again after an update.

Updating. Run artisign update (also artisign --update). It installs the latest version with npm install -g artisign@latest, re-pins the artisign entries in your AI tools to it and restarts a running artisign of an older version with its open projects. Then quit and reopen your AI tool, so it launches the re-pinned entry. It does nothing when you are already on the latest version, and never uses sudo. Nothing is updated automatically: `ar

... View full README on GitHub