Instatic and the site you stop renting

A five-page marketing site with a blog, a working contact form, and two people on your team allowed to change the copy without filing a request. On Webflow that is 25 dollars a month for the Premium site plan that carries the CMS, 19 for the Core workspace that lets somebody stage a change before it goes live, and 39 a month for every additional full seat allowed to touch the design. Call it a thousand dollars a year for a brochure site, before the first word gets written and before anyone quotes you to build it. Instatic is the MIT-licensed version of that same shape. It went from nothing to 7,600 stars since the end of April, it ships an AI agent that edits the actual page instead of handing you a code block to paste, and you can have it live on a hosted container in about two minutes without opening a terminal.

What is actually in the box

Instatic is one Bun server that holds the whole website: the canvas editor, the content store, media, auth, forms, plugins, and the publisher. Backed by SQLite out of the box, Postgres when you want managed backups. The team behind it is CoreBunch, who also make Core Framework and Motion.page, tools that a large number of WordPress professionals already use to build sites for a living. Core Framework is not a plugin here. It is wired in as the design token engine, so you define one brand color and get the full tuned shade scale, one type ramp that scales with the viewport instead of forty hand-picked font sizes, and a spacing scale that keeps the rhythm consistent across breakpoints.

The output is the part that matters and the part most builders quietly compromise. A published Instatic page is semantic HTML and compact CSS baked to disk and swapped in atomically. No framework runtime, no editor attributes, no div soup. The few genuinely per-visitor parts get detected and lazy-loaded by a runtime that weighs about 1.1 kilobytes. Your visitor gets a file.

The AI agent is the reason this belongs in an AI library rather than a web design one. You describe what you want and it builds real editable nodes on the canvas, writing semantic HTML for structure and CSS for style through the same import pipeline you would use if you pasted markup yourself. There is a 35-tool site scope for building pages and a 15-tool content scope for editing entries. You bring your own model: Claude, OpenAI, OpenRouter, or a local Ollama instance. Your key, your model, your bill. That last part is the whole economic argument and I will come back to it.

Forms belong to your own database. You build a form out of semantic fields, Instatic reads the fields you placed and creates the matching table, and submissions land in your tables. No third-party form service, no embed script, no monthly fee for a contact form. And because every table you create is a content source a loop can render, a "Team" table you define once can be looped onto the about page, and a plain data table can quietly serve as a testimonials list, a product catalog, or a very lightweight CRM without installing anything.

The traction is real rather than a launch-day spike. Ten tagged releases between the middle of June and the end of July, running from v0.0.5 to v0.0.14, with new outside contributors appearing in almost every one. The most recent commits landed on 3 August. The LICENSE file is plain MIT, checked rather than taken from the README's word for it, copyright David Babinec, 2026. No tiers, no open-core asterisk, no contact-sales wall around the good parts.

The number that actually moves

For an owner-operator the displaced cost is the subscription plus whatever the agency charges to touch it. For the RevOps or marketing-ops reader it is something more specific and more annoying: the seat. Webflow's full seat, the one that lets a person design rather than just edit copy, is 39 dollars a month billed yearly. The limited seat, which can edit content and assemble components but not build, is 15. Framer is cheaper on the site plan at 10 or 30 dollars a month, and then charges 20 per additional editor and 10 per content editor, with agent usage metered against a monthly credit allowance. Either way, the number of people at your company allowed to change your own website is a line item, and it is the line item that produces the ticket queue. Somebody wants a paragraph changed on the pricing page. That person does not have a seat. So they message the one person who does.

Instatic's answer is roles built from 38 capabilities, token-based sessions, TOTP two-factor with the secrets encrypted at rest, account lockout with backoff, and step-up prompts before dangerous operations like deleting a user. You can hand out as many accounts as you want because nobody is counting them. That is not a small difference in a 40-person company. It is the difference between a website your marketing person owns and a website your marketing person requests changes to.

There is a second, quieter lock-in worth naming. Webflow's own FAQ says you can export your site and host it anywhere with a paid workspace plan, then adds two sentences: dynamic content must be exported collection by collection, and forms will stop working. Below that: you can only develop websites in Webflow, and if you export your code it cannot be reimported. So the export button exists, and what comes out the other side is a corpse. Instatic never has that problem, not because it is generous but because the published artifact was always just HTML and CSS. There is nothing to convert. Back up the database and the uploads folder and you have backed up the entire site.

What it takes to stand up, honestly

There is no hosted Instatic. Say that plainly, because "one-click deploy" invites the wrong assumption. What exists is a Railway template: pick SQLite or Postgres, press the button, wait about two minutes while it generates the secret keys, attaches the storage volume, and sets up health checks. You never open a terminal, and the first visit walks you through creating your site and your owner account. There are also documented paths for Render, and for Docker on your own VPS with Caddy handling TLS. Updating is redeploying the latest image; the database and uploads live on the attached volume, so the container gets replaced without rebuilding the site.

The deploy is not the work. The deploy is fifteen minutes including reading the docs. The work is the site, and if you are moving an existing one there is a Super Import that takes a whole static site, HTML and CSS and images and fonts, and turns it into pages, style rules, design tokens, and media, showing you every conflict before it writes and treating the entire import as a single undo. Budget half a day for a small site if you already know what you want it to look like, and a couple of days if you do not. That is not an Instatic number. That is what building a website costs no matter what tool you use, and any post telling you a repo makes it free is selling something.

Hosting is the cheap part and I want to be precise about it. On Railway the Hobby plan bundles five dollars of monthly usage, and the meter runs at 0.00000772 dollars per vCPU-second, 0.00000386 per gigabyte-second of memory, and six hundredths of a millionth of a dollar per gigabyte-second of volume. A small always-on Bun container with a modest volume lands inside or just past that five dollars. Egress is five cents a gigabyte, which is where a genuinely busy site starts to cost something. Call it five to fifteen dollars a month for a brochure site, and a plain VPS is in the same range.

Then the model bill, which is the part that is genuinely new and genuinely variable. The agent bills to your key, so the comparison is not free versus paid, it is metered by your provider versus metered by your CMS vendor. Webflow sells 2,000 AI credits for 20 dollars a month; Framer bundles 1,000 to 3,000 credits by plan and sells more. Against that, Claude's API rates are 2 dollars per million input tokens and 10 per million output for Sonnet 5 through the end of August, 1 and 5 for Haiku 4.5, 5 and 25 for Opus 5. An afternoon of building pages by description is a few hundred thousand tokens in each direction, which puts a real build session in single-digit dollars and a month of small copy edits under a dollar. Pointed at a local Ollama model it is zero and slower. The honest framing is not that the AI is cheaper. It is that you can see the meter, and you can turn it off without losing the editor.

The honest take

It is version 0.0.14, and the project says so on its own front page: pre-1.0 on purpose, APIs and workflows can still shift, and if that makes you nervous, wait. Take that at face value. Running a client's site on a 0.0.x release is a different decision from running your own.

The analytics story is the clearest gap. Instatic's dashboard is deliberately operational right now: status tiles, an append-only audit log of every meaningful admin action, and your own form submissions. There is no first-party visitor analytics yet; it is on the roadmap. Nor is there A/B testing, localization, or ecommerce, all of which Webflow sells as add-ons at 299, 9 to 29, and 29 dollars a month respectively. If your site is a revenue surface that gets tested and localized, you are not shopping in this aisle. If it is five pages and a blog, you were never using those anyway.

Then the maintainer question, which is the one nobody asks until it matters. The commit history is overwhelmingly one person. Outside contributors show up in every release, which is the healthy signal, but the bus factor is one. MIT means you keep the code no matter what happens; it does not mean you keep the roadmap, the security patches, or the person who knows why the publisher works the way it does. And there is no visible business model on the repo. No tiers, no open-core, no "contact sales." That is refreshing and it is also an unanswered question, because something has to pay for a full-time maintainer eventually, and the answer arriving later is usually a surprise.

On data, self-hosting means the site, the database, and the uploads are yours, which is the point. Two caveats. The agent sends your page structure and content to whichever model provider's key you configured, so the "your data stays home" claim ends at the API call unless you run Ollama locally. And backups are entirely on you: the whole disaster recovery plan is a Postgres dump or the SQLite file plus the uploads folder, which is admirably simple and completely worthless if nobody schedules it. Write that cron job the same afternoon you deploy.

The plugin system deserves credit and one caution. Backend plugin code runs in a per-plugin worker hosting a QuickJS-WASM sandbox with no filesystem, no environment variables, and no network at all unless you grant it, one host at a time. That is more careful than most CMS plugin ecosystems have ever been. But editor extensions and admin pages are different: they run in the admin window and require an explicit permission before install. That permission is the one to read twice.

None of this is the interesting part, though. The interesting part is what happened to the export button. Every website platform advertises one, and every one of them hands you something that no longer works, because the thing you were really renting was never the pages, it was the machinery that renders them.