miow

Delete a wiki page and its whole history. This cannot be undone, and restore_wiki_page cannot bring it back: read it first with read_wiki_page if it may be wanted. A category's hub article is its category's path.

Replace the site's pages with an existing website's (Squarespace, Wix, or anything with a sitemap), word for word, at the same paths, with its images copied across. Runs in the background for a minute or two: poll get_site until `job` is null, or read its error.

Put a wiki page back as one of its revisions had it. The restore is a revision of its own, so it can be undone the same way.

Give the site a category tree and make it a wiki, or change its tree. A wiki's pages live in its categories, each with a version and a history of every save, and are written with save_wiki_page, not write_site. The tree is sent whole: each node's slug is its path segment, and its metadataKeys are the facts a page under it carries, a select key becoming a filter on the category's page.

Put the site back as it was before a change, undoing it and every change since. Take the change id from history. The undo is itself a change, so it can be undone too.

Create or replace pages, each sent whole, and delete others; optionally rename the site, change its theme, or set its header, replaced whole. A page you do not send is left as it is. Every write can be undone (history, undo). The input schema gives every block's fields. Pages - path "" is the home page and must exist. Other paths are lowercase slugs, slash-separated, no leading slash. - navOrder places a page in the header nav (0 first); null keeps it out. The home page is reached by the site name, so give it null. - path "search" is the site's search results page, which every site has; no page takes it. - path "_footer" is not a page to visit but the footer shown under every page: addresses, hours, contact details. Content and columns blocks, no hero and no h1, navOrder null. A site without one has no footer. - description is the page's search snippet: one sentence, under 160 characters. - Omit every id. Blocks - A page is a list of blocks, stacked vertically. Every page has exactly one h1: an opening hero's heading, a product's or a products list's with a menu, or else an h1 at the top of its first content block. - "hero" is a heading, a sentence or two of plain text, and optionally one button. It opens a page, its heading the h1: use one on the home page and on pages that sell something. With "call": true it is a call to action further down instead, its heading an h2: at most one on a page, usually last, asking for one action, with a button. Its heading and text are plain text that may wrap words in <strong> or <em>, as the source sets them bold or italic; words in <strong> in its heading are also drawn in the accent colour. Its optional image is a picture filling it behind the words, by https URL, under the same rule as a pictures block. Where its words are a subheading rather than a sentence (a line or two, often set as headings at the source), give "subheading": true. With no picture, "backdrop": "mesh", "grain", "flow" or "marble" sets a slowly moving field of the site's own colours behind the words, smooth, grainy, folding like poured paint, or hard-edged as marbled paper, and "backdropSpeed" and "backdropIntensity", each 0 to 1, set how fast it moves (0.25 unset) and how strongly it folds (0.5 unset): at most one on a site, on the home page's opening, for a brief that wants a modern or striking look, and never on a site moved across from elsewhere. Over a picture or backdrop its words sit on boxes of the page's background; "box": "panel" or "accent" sets them on that colour instead, and "none" sets light words on the picture itself. - "content" holds HTML, plain and semantic: h1–h4, p, ul, ol, li, a, strong, em, blockquote, table, hr. No class, style, script or img: they are stripped. After a hero, headings start at h2. - "columns" sets 2–4 columns side by side, each {"blocks": [...]} of the blocks a column may hold (content, pictures, figures, code, link groups, references, notices, lists): for features, services or contact details. They stack on a phone. To place them freely instead, as a page builder's grid does, give every column an "area": {"col", "span", "row", "rows"} on twelve columns, rows counted from 1 (up to 12 columns then); cells may sit lower than their neighbours or overlap. With "beside": true and one column, it is a facts panel beside the text instead: on a page about one thing (a product, a place, an organisation), placed before the page's content, its usual content a picture and a two-column table of facts. - "pictures" shows pictures by https URL, {"items": [{"src", "alt": what it shows, "caption"?}]}: one is a figure, several a row the visitor scrolls sideways. A video file is an item with "video": true, its src an MP4, and "poster"?: the picture it shows before it plays. Use only picture and video URLs the brief gives you; never invent one. For any gallery or slideshow on a site being moved across, one pictures block holding each picture once, in order, without the slideshow's counters ("Image 1 of 12") or its thumbnails. A YouTube link is an iframe in a content block. - "product" makes its page one thing for sale: {"images": [https URLs, the first on every listing], "price"?: as it reads ("£2,500"), "sold"?: true, "tags": [the lists it appears in], "text": HTML of what it is (materials, size, year, words about it)}. It draws the page's title as the h1, so the page has no hero; it goes first. One product per page. Write a price as an amount, "£2,500", "$40" or "€12.50", and the product can be bought online once the site takes payment; "from £40" or "POA" keeps it to the Enquire button. Where it is sold elsewhere, as a book is by its publisher, give it "action": {"label": "Buy from Salmon Poetry", "href": the page that sells it}. A service, booked rather than sold (a lesson, a treatment, a consultation), is a product with "service": true, "duration"?: how long it takes as it reads ("1 hour"), and "unit"?: what its price is for as it reads after it ("an hour", "per person"); its button reads Book, and where it is booked elsewhere, "action" goes to that calendar. "sold" on a service means fully booked. - "list" lists pages of the site, read as the page is: {"source": "under", "under"?: a path} the pages directly under it, a section's index page; {"source": "newest", "count"?, "under"?} the most recently changed, a blog's latest posts; {"source": "chosen", "pages": [paths]} those, in order; or {"source": "products", "tag"?} the product and service pages with that tag, or all, newest first. "show": "cards" (the default) draws each page's first picture, title and search snippet, "links" each title with how long ago it changed; products are always cards of their pictures, name and price, cropped square, "shape": "portrait" upright for books and posters, "landscape" wide. A shop with several lists gives each list's products block a "menu" of links to all of them, [{"label": "All", "href": "/shop"}, …], shown down the left; the block then heads the page with its … [10121 more characters not shown]

A link that makes whoever opens it the owner of a site nobody owns yet, after saving a passkey. Works once, for two weeks. Give it to the person who should own the site, never post it. Once the site has an owner, they invite others from its settings.

Make a new, empty site, live at once at https://<slug>.miow.me. Returns its token (pass it to every other tool as `token`) and a claim link. Give the claim link to your human: opening it on their phone saves a passkey that makes them the owner. An unclaimed site is deleted after seven days. Then fill the site with write_site.

Whether a pending domain from set_domain is live yet, and the DNS records still missing if not. DNS can take an hour to be seen; call again later rather than in a loop.

The last 30 changes to the site, newest first: what each did, who made it and when. Pass a change id to undo to put the site back as it was before that change.

Create a wiki page, or change one, as one revision of its history. Send the blocks whole, with base_version from read_wiki_page: a save over a page someone changed since is refused, and you read it again. A change to blocks or title moves its version (major for added, moved or removed blocks, minor for changed words); a metadata-only change moves nothing. Pages - path "" is the home page and must exist. Other paths are lowercase slugs, slash-separated, no leading slash. - navOrder places a page in the header nav (0 first); null keeps it out. The home page is reached by the site name, so give it null. - path "search" is the site's search results page, which every site has; no page takes it. - path "_footer" is not a page to visit but the footer shown under every page: addresses, hours, contact details. Content and columns blocks, no hero and no h1, navOrder null. A site without one has no footer. - description is the page's search snippet: one sentence, under 160 characters. - Omit every id. Blocks - A page is a list of blocks, stacked vertically. Every page has exactly one h1: an opening hero's heading, a product's or a products list's with a menu, or else an h1 at the top of its first content block. - "hero" is a heading, a sentence or two of plain text, and optionally one button. It opens a page, its heading the h1: use one on the home page and on pages that sell something. With "call": true it is a call to action further down instead, its heading an h2: at most one on a page, usually last, asking for one action, with a button. Its heading and text are plain text that may wrap words in <strong> or <em>, as the source sets them bold or italic; words in <strong> in its heading are also drawn in the accent colour. Its optional image is a picture filling it behind the words, by https URL, under the same rule as a pictures block. Where its words are a subheading rather than a sentence (a line or two, often set as headings at the source), give "subheading": true. With no picture, "backdrop": "mesh", "grain", "flow" or "marble" sets a slowly moving field of the site's own colours behind the words, smooth, grainy, folding like poured paint, or hard-edged as marbled paper, and "backdropSpeed" and "backdropIntensity", each 0 to 1, set how fast it moves (0.25 unset) and how strongly it folds (0.5 unset): at most one on a site, on the home page's opening, for a brief that wants a modern or striking look, and never on a site moved across from elsewhere. Over a picture or backdrop its words sit on boxes of the page's background; "box": "panel" or "accent" sets them on that colour instead, and "none" sets light words on the picture itself. - "content" holds HTML, plain and semantic: h1–h4, p, ul, ol, li, a, strong, em, blockquote, table, hr. No class, style, script or img: they are stripped. After a hero, headings start at h2. - "columns" sets 2–4 columns side by side, each {"blocks": [...]} of the blocks a column may hold (content, pictures, figures, code, link groups, references, notices, lists): for features, services or contact details. They stack on a phone. To place them freely instead, as a page builder's grid does, give every column an "area": {"col", "span", "row", "rows"} on twelve columns, rows counted from 1 (up to 12 columns then); cells may sit lower than their neighbours or overlap. With "beside": true and one column, it is a facts panel beside the text instead: on a page about one thing (a product, a place, an organisation), placed before the page's content, its usual content a picture and a two-column table of facts. - "pictures" shows pictures by https URL, {"items": [{"src", "alt": what it shows, "caption"?}]}: one is a figure, several a row the visitor scrolls sideways. A video file is an item with "video": true, its src an MP4, and "poster"?: the picture it shows before it plays. Use only picture and video URLs the brief gives you; never invent one. For any gallery or slideshow on a site being moved across, one pictures block holding each picture once, in order, without the slideshow's counters ("Image 1 of 12") or its thumbnails. A YouTube link is an iframe in a content block. - "product" makes its page one thing for sale: {"images": [https URLs, the first on every listing], "price"?: as it reads ("£2,500"), "sold"?: true, "tags": [the lists it appears in], "text": HTML of what it is (materials, size, year, words about it)}. It draws the page's title as the h1, so the page has no hero; it goes first. One product per page. Write a price as an amount, "£2,500", "$40" or "€12.50", and the product can be bought online once the site takes payment; "from £40" or "POA" keeps it to the Enquire button. Where it is sold elsewhere, as a book is by its publisher, give it "action": {"label": "Buy from Salmon Poetry", "href": the page that sells it}. A service, booked rather than sold (a lesson, a treatment, a consultation), is a product with "service": true, "duration"?: how long it takes as it reads ("1 hour"), and "unit"?: what its price is for as it reads after it ("an hour", "per person"); its button reads Book, and where it is booked elsewhere, "action" goes to that calendar. "sold" on a service means fully booked. - "list" lists pages of the site, read as the page is: {"source": "under", "under"?: a path} the pages directly under it, a section's index page; {"source": "newest", "count"?, "under"?} the most recently changed, a blog's latest posts; {"source": "chosen", "pages": [paths]} those, in order; or {"source": "products", "tag"?} the product and service pages with that tag, or all, newest first. "show": "cards" (the default) draws each page's first picture, title and search snippet, "links" each title with how long ago it changed; products are always cards of their pictures, name and price, cropped square, "shape": "portrait" upright for books and posters, "landscape" wide. A shop with several lists gives each list's products block a "menu" of links to all of them, [{"label": "A… [9442 more characters not shown]

Put a site nobody owns yet on a domain your human already has; once someone owns the site, its domain is set only from its settings. Returns the DNS records they must set where they bought it: first a TXT record proving the domain is theirs, then the records that point it here. The site moves to the domain once those records are seen; call domain_status to check.

Copy an image from a URL into miow's store and get back the URL to use in a pictures block. Up to 10 MB. Use it for an image on someone else's server, which may move or vanish.

A wiki page's revisions, newest first: version, kind of change, message, who and when, and what each changed block by block. Pass a revision id to restore_wiki_page to put the page back as it was.

A site's name, theme, address and every page by path and title, plus its status: a running or failed job (from import_site) and its domain. Call before changing a site you did not just create. With `site` and no token, any live site's name, address and pages, without its status.

The site's visitors, agents, page views, bounce rate and visit length over the last `days` days, the same figures for the same number of days before, visitors and agents by day, the agents by name, its most visited pages, entry pages, sources (a referring host or a link's utm_source), countries (ISO codes) and devices, and what visitors searched the site for and found nothing. A visitor is one browser on one day, counted without cookies. An agent (an AI crawler, an MCP client, or anything reading the site's markdown, llms.txt or .well-known files) is counted the same way and never among the visitors.

One page whole, in the shape write_site takes, so you can change it and send it back. Read every page you mean to change first: write_site replaces a page whole.

One wiki page whole: its title, blocks, metadata, version and how many revisions it has. Pass its version as base_version when you save it back.

The site's pages that hold some words, best first: titles starting with them, then titles holding them, then descriptions and page text; a question no page holds word for word is matched in English, stems and all. Each with its path, url, title, description and the passage the words are in, the first 20 of `total`: the same rows as the site's search page and /search.json?q=.