The language, section by section.

For machines and coding agents

/docs/index.json

every section and heading, with its URL. Use it to map the whole reference or resolve a slug to a page.

/docs/search.json

one entry per heading (title, section, first sentence, keywords), as static data. Use it to search or browse offline, without a request per query.

/docs/anchors.json

the slug map, every address this reference can be linked at. Use it to check an anchor rather than guess one.

The search endpoint: GET https://vox-lang.dev/docs/search?q=<words>[&limit=<n>] returns JSON, {query, spec, slugRule, hits: [{url, title, section, excerpt, score}]}. q is capped at 120 characters and limit at 50, default 20.

spec is {commit, sha256}, the LANGUAGE.md commit and content hash the index was built from; slugRule is the rule for turning a heading into a URL, quoted under Anchors below; hits is ordered best match first, one entry per heading.

Matching is whole-word first, then a prefix match from 2 characters, then a one-typo match on words of 4 or more characters; exact beats prefix beats fuzzy. An empty hits array means those words are not in the reference at all, not that the syntax was wrong: do not guess further from it.

To read a hit: its url is /docs/<slug>/#<anchor>; the matching text is the Markdown twin at /docs/<slug>.md, where the anchor is a heading. Or GET the section page itself with an Accept: text/markdown header, no twin URL needed.

For example, GET /docs/search?q=file%20io currently ranks "File Properties" first, at /docs/file-io/#file-properties; the text is the twin at https://vox-lang.dev/docs/file-io.md, under its "File Properties" heading.

Every response's spec.commit names the exact source commit the index was built from; cite it alongside anything quoted from a hit.

Anchors: the slug rule below builds a link to any heading; /docs/anchors.json is the authority when in doubt.

A URL is the heading, lowercased, with punctuation dropped and spaces turned into hyphens.

Each "##" section of the spec is a page under /docs/, and every heading inside it is a fragment on that page, so "File I/O" is /docs/file-io/ and "Reading a whole file" inside it is /docs/file-io/#reading-a-whole-file.

The whole reference is also one file, for skimming or grepping in a single pass: /docs/all.md, 221 KB. /llms-full.txt is the larger bundle: overview and installation, this manual, every example, then the style guide.