The website your own agent runs.
spun.ink hosts your site. The agent you already use writes its content, changes its design and publishes updates through one connection — no content dashboard to learn, no site repository to maintain.
Real sessions, recorded.
Wrong turns left in — every call you see is one your own agent can make.
3:42
Live build / migration
Claude builds and publishes a website, 7 minutes on the clock
A fresh account, a 24-line brief and one prompt: the agent uploads four images, writes the layout, the page template and the page, and publishes to coffee4ai.myspun.ink. Sign-up 08:35, site public 08:42 on the menu-bar clock — the video is that session, cut to 3:42.
One change, start to finish
You ask. Your agent edits, shows you, publishes.
Content goes public when your agent publishes
A page changes publicly only at a publish, after you have seen the preview. Templates, settings and navigation are live when saved: a template write is render-checked first, navigation and forms are validated, and the version they replace is kept so it can be restored.
The hosting is not yours to run
Server, TLS, engine updates, backups, sitemap, feed, search and social meta are spun.ink's. On a paid plan, a custom domain is one DNS record. The content and the design stay yours to direct.
No per-edit meter
A flat plan fee, no AI credits, and the free plan includes unlimited edits. spun.ink runs no model; what your own agent costs stays between you and your model provider.
Who this is for — and who it isn’t.
Want an app built for you, or a dashboard you click through yourself? Those exist, and they are good at it. Want a website the agent you already use keeps running — writing it, restyling it, publishing it, rewinding it — that is spun.ink.
The same job, on each kind of setup
Every column takes a sentence from you. What differs is whose agent does the work and what it works on.
| The job | A vendor's agent + its hosting | A dashboard platform + an MCP | Your agent + a repo + a host | spun.ink |
|---|---|---|---|---|
| Set up a site | Prompt the vendor's agent; the vendor hosts the app. | Create an account, pick a theme, connect an MCP separately. | Scaffold a project, choose a host, wire the deploys. | Your agent calls sign_up; you confirm one email. |
| Change a paragraph | Ask the vendor's agent; it spends credits and rebuilds the app. | Edit it in the dashboard, or ask your agent through a side door with confirmations. | Ask your agent; commit; deploy. | Ask your agent: update_block, a preview link, publish_content. |
| Restyle the whole site | Regenerate; more credits. | Switch themes; the design is mostly out of your agent's reach. | Your agent rewrites the frontend; deploy. | Your agent rewrites templates. A new one previews first; a shared one goes live on save, render-checked, with the previous version kept. |
| See it before it is public | The vendor's preview. | The dashboard preview. | A preview deploy, if you set one up. | Draft content on a preview link. A shared template has no preview: the save is the change. |
| Recover from a mistake | The vendor's checkpoints, where offered. | Revisions for content. | git, for the code. | restore_revision for one item; restore_site to rewind the captured site to a moment. |
| What the intelligence costs | Credits per edit, paid to the vendor. | Your seat or site fee; your agent's cost on top. | Your agent's cost. | No per-edit meter; your agent's cost stays separate. |
| What you keep if you leave | Exported code, where offered. | Export tools vary. | The whole repository. | Every page as HTML and every record readable over MCP. No export tool today. |
The agent-operated website platform.
No dashboard. We never built one.
A website is data — its config, content, templates, and layout. The engine is code. The sole author is your own AI agent: that is what agent-operated means. Because everything that makes a site unique is structured data, a new page is written, not deployed, and it’s live in minutes.
Your agent invents content types, restyles the whole site, and rewrites navigation at runtime — no pull request, no build, no restart. Those writes keep the version they replace, so your agent can put a template, a page or the whole captured site back to an earlier moment.
A website is data. Dashboards were only ever the workaround for humans operating that data by hand — so we never built one.
The spun.ink thesis
Bring the agent you already use.
Claude or ChatGPT in the chat, or Claude Code, Cursor, VS Code or Codex in a terminal or editor — any MCP client, with the same URL. spun.ink does not try to be your AI. It gives your AI a website to run.
What people ask first
Isn't this just another AI website builder?
A builder's agent builds you an app that it hosts, and each change spends its credits. Here your own agent operates a site that spun.ink hosts: its content, its templates, its layout. It can come back tomorrow, or in year three, and change it again.
What if the agent breaks my site?
Content goes public only when your agent publishes, after a preview, and a publish that would not render rolls back. Template writes are render-checked before they save; navigation, records and forms are validated. Template, settings, navigation and form writes keep the version they replace, and restore_site rewinds the captured site to a moment. A render check proves the page renders, not that it reads well: that is still yours to judge on the preview.
Can I leave?
Every page is served as plain HTML, and your agent can read every record and revision over MCP. There is no export tool today, so moving means re-authoring the site elsewhere. We would rather say that here than after you have signed up.
Live in ten minutes. Still a sentence to change in year three.
Drop your email below and point your agent at spun.ink — it stays the operator from there. spun.ink charges nothing per edit; the tokens your agent spends are billed by your model provider, as they already are.