Release notes

The authoritative, commit-level history lives in CHANGELOG.md and on the releases page. This page summarises each version.

Unreleased

A round on input and output: more ways to feed yomi a page, and more shapes to get one back.

• An EPUB book. yomi pack --format epub (or a .epub output name) builds an EPUB 3 book from a crawl, readable on any e-reader. Each page becomes a well-formed XHTML chapter, in-scope links are rewired to sibling chapters, every referenced image is pulled into the book so it reads with no network, a generated navigation document lists every page, and a drawn-in-code cover stands in front. The book passes EPUBCheck, the official validator, with no errors or warnings, and carries accessibility metadata. --icon supplies your own cover, and --title, --language, and --date fill the book's metadata. The crawl keeps its SQLite store as a sidecar for the next incremental run, the same as a ZIM build.
• More ways in. yomi read - reads HTML from standard input and yomi read page.html reads a local file, so you can convert a page you already have without a fetch. --base sets the URL that relative links resolve against.
• More shapes out. yomi read -f json|jsonl emits the full page record instead of a Markdown document, and -f html emits a self-contained HTML article. yomi site --format json|jsonl writes one dataset file of every crawled page, for feeding a pipeline.
• Reach more pages. yomi site --sitemap (and pack --sitemap) seeds the crawl from the site's sitemap.xml and any robots.txt Sitemap: lines, following an index one level, so the crawl reaches pages that are listed but not linked.
• Resume a site crawl. yomi site --resume continues an interrupted Markdown crawl. The crawl records each page in a .yomi-state.jsonl sidecar as it goes, so a re-run skips the pages already done, the way pack resumes from its store.

v0.2.1

A pass to make a packed ZIM open nicely in Kiwix.

• A library icon. Every ZIM build now embeds a 48x48 icon, so Kiwix shows a real book tile instead of a blank placeholder. yomi draws a built-in reading icon by default; --icon with a PNG uses the site's own logo instead.
• Metadata Kiwix reads. The archive now carries a Counter of the packed pages for the library listing, Creator is the packed site rather than the tool version, and Scraper names yomi and its version, so the three keys read the way Kiwix expects.

v0.2.0

A new way to keep a whole site: one file instead of a folder.

• yomi pack <url> crawls a site and bundles it into one file. The default is a SQLite database with clean pages, links, and images tables you can query with SQL; --format zim (or a .zim output name) builds a ZIM offline archive you can read in Kiwix, with each page rendered to a self-contained HTML document, internal links rewired to the sibling entries, and a generated contents page.
• Resumable, incremental crawls. The crawl is backed by the SQLite store, so a pack resumes where it left off and a later run keeps every page already stored, fetching only what is new. --refresh re-reads everything; --max-age re-reads only the pages older than a cutoff, so a daily mirror stays current without reading the whole site. A ZIM build keeps its store as a sidecar for the next run.
• The format picks itself. A .db/.sqlite or .zim output name selects the format, so -o site.zim just builds a ZIM. ZIM metadata flags (--title, --description, --language, --date) and --no-compress round out the archive, and pack takes the same scope, limit, worker, and robots flags as yomi site.

v0.1.2

A small pass for the people actually typing the commands.

• Bare hosts read. yomi read, yomi meta, and yomi links now accept a URL with no scheme: yomi read example.com/post defaults to https://, the shorthand yomi site already took. Every example in the docs uses it, so now they all run exactly as written.
• Friendlier docs. A rewritten README with a terminal demo of the read, save, and metadata loop, and a warmer pass over the homepage and guides.

v0.1.1

A quality pass on the extracted Markdown, a cleaner yomi meta, and docs that match the real output.

• Cleaner article Markdown. A caption that only repeats an image's alt text is dropped, a picture linked to its own full-size file is unwrapped to a plain image, and a stray share or subscribe button left in the body is removed. A bare angle bracket the converter had escaped is restored in prose (so a -> b and <placeholder> read as written), a needless backslash before an underscore or tilde is dropped, and a link with no destination renders as its text alone. Code blocks keep their bytes exactly.
• yomi meta is just metadata. The JSON record no longer carries an empty markdown field. It is the page's url, title, byline, site_name, excerpt, lang, published, fetched, word_count, reading_time, and rendered, followed by links and images as arrays of objects.
• Docs match the code. The front-matter and meta examples now show quoted string values, reading_time as a whole-minute number, and links and images as arrays rather than counts.

v0.1.0

The first release. yomi reads a web page, or a whole website, into clean Markdown: fetch the page, render the JavaScript only when the page needs it, extract the main content, and convert what is left to GitHub-Flavored Markdown with a YAML front-matter block.

• yomi read <url> turns one page into Markdown, printing to stdout or writing to a file with -o. The default --render auto static-fetches first and escalates to headless Chrome only when the page looks JavaScript-gated, so most reads never launch a browser. Readability extraction drops the nav, cookie banners, footers, and share rails before conversion.
• yomi site <url> reads a whole site. The default output is a folder of .md files mirroring the URL paths, with a SUMMARY.md table of contents and a shared media/ folder; --single assembles one combined file with a table of contents, per-page sections and anchors, and demoted headings for a single clean outline.
• yomi meta <url> prints a page's metadata record as JSON, and yomi links <url> lists the outbound links in the article body, with --json for a structured list.
• yomi serve [dir] previews a folder of Markdown over a local static file server.
• Three image policies. --images remote (the default) leaves image URLs absolute, download fetches them next to the output, and inline embeds them as base64 data URIs, with a --max-image-mb cap.
• Polite by default. Honours robots.txt, scopes to the seed host, and reads pages with four parallel workers.
• Packaged everywhere. Archives, .deb/.rpm/.apk, a multi-arch GHCR image with Chromium bundled, plus Homebrew and Scoop.