Drive your site from your agent.

The entry point — every other tool needs a slug from here.

A new site, live the moment it is made, with starter content in it. Shape it afterwards with write_site_file.

One site at a glance — unpublished changes, live URL, active preview links.

The file tree: the working copy by default, or the live site, or one preview.

One file's contents, plus a content sha so a later write can catch conflicts.

Create a file or replace one whole. Writes to the working copy — nothing is public until you publish.

Change part of a file: replace one exact, unique snippet instead of resending the whole thing. Prefer it over write_site_file on a page that already exists — a full rewrite costs the entire file, and on a long one it can be cut off half-written.

Remove a file from the working copy. Without it a file written under a mistyped name could only ever be overwritten, never taken away.

What a VISITOR gets for one address — the response, not the file. On a pepita site the two differ enormously: a content tag is one line in the file and every item in the response, and an item's own address is in no file at all. Reads the unpublished working copy by default.

Live at <slug>.pepita.page, and on any custom domain.

A stable URL you can send to anyone, for review before going live.

Same URL, new content — so a link you already sent shows your latest changes.

Their names and URLs — the name is what the update and delete tools take.

The version itself is kept — restorable later from the editor's History.

A map of form name → how many submissions it holds. The cheap first look, before reading any of them.

One form's submissions, newest first. live or preview picks the source; neither means your own test submissions from the editor. Past 100 it answers with the count instead, so your agent can ask before filling the screen — then call again with confirm_large.

A time-limited download link — xlsx, csv or json — so your agent can hand you a file instead of reading a thousand rows out loud.

One per form, each matched to the form by name.

Subject, sender, HTML body, and the sha you need to change it.

Creates the template when the form has none and updates it when it does. Leave out html and pepita seeds a starter body. Updating an existing template changes the working copy — save it to change the emails people receive. A brand-new template is saved as it is created.

The same act as the editor's Save button on the template's row. Editing a template anywhere changes only the working copy; this is what makes those edits the email a visitor actually gets.

Confirmation emails for that form stop going out immediately.

The names to write in the body — as bare filenames, not as the URL beside them. pepita expands a filename at send time onto whichever of your domains matches the address the mail goes out from; a URL written in by hand is left exactly as it is, forever.

Emails sent from now on show a gap where it was. Mail already delivered may keep showing it for a long time — a picture in someone's inbox is a copy, and deleting it here cannot reach that.

The starting point — the names here are the names a page asks for.

The complete HTML document for one item — markup only. A <style> or <script> inside it is dropped; those live in the collection's own two files.

A brand-new template is saved as it is created; changing one that exists writes a working copy.

A collection's parts are edited together, so one save promotes them together. Nothing pending is a normal outcome, not a failure; a partial save reports what did not land.

With a layout it removes only that shape; without one, the template and every layout with it.

A collection's own style.css or script.js. Read before you write: writing replaces the file outright.

Where a collection's look and behaviour live — one of each per collection, applying to every layout. A <style> or <script> written inside a template is dropped silently, so this is the only place either belongs. Changes the working copy; save_content_template with no layout promotes it.

Newest first, each with its id, its address and whether it is live or still a draft.

All of them land or none do — a refusal names every bad item by its position.

The published state is unchanged — editing never publishes or unpublishes.

One item. The collection and its template are untouched.

A new item is a DRAFT — it shows in the editor and on preview links until this is called.

Nothing is deleted — the item stays in the editor.

Fetches a picture from a public address into the collection and hands back the pepita url to put in the item.

A short-lived address that takes one file with no other credential, for an agent that can run a command.

Each video with its streaming (HLS) URL and poster. Videos live outside the file tree, so this is how you find them.

A temporary link to the original file you sent, before streaming prep.

Its URLs never change, so pages that reference it keep working.

Removes the video for good, with its streaming files and poster.

The exact contract for overview, forms, confirmation-emails, video or headers-and-csp. The only tool that takes no slug, and the one to call before writing any of those — a guessed form action produces a page that looks right and silently misses the feature.

The connector signs itself in over OAuth the first time you use it — there is no key to paste. Every authorized client is listed in the editor under your avatar → Settings → Devices.