@thallylabs/core

The framework-agnostic content engine behind the open-source Thally docs platform. It turns one directory of MDX into structured, queryable product knowledge for people, search engines, and AI tools through content parsing, search, embeddings, and brand-token utilities.

It has no dependency on Next.js, React, or any particular host — the host provides its list of pages through a small resolver seam, and core does the rest. Thally's own site is one host; you can build another.

Install

npm install @thallylabs/core

What's inside

| Export | Purpose | | --- | --- | | @thallylabs/core | Everything below in one barrel: content parsing, search (searchDocs, buildSearchCorpus), and embeddings (buildEmbeddingIndex, getRelevantChunks). | | @thallylabs/core/content | Just the content pipeline (parseMdxContent, getContentDocument, mdxToMarkdown, ContentDocument) — for consumers that transform content and don't need search or embeddings. | | @thallylabs/core/theme | Pure brand-token utilities (colors, CSS theme variables) — no Node or content dependencies, safe to import from client bundles. |

The package is side-effect free ("sideEffects": false), so bundlers tree-shake unused exports even when you import from the barrel. Size-sensitive consumers (edge workers, client bundles) should still prefer the subpath entries — they keep unused modules out of the module graph entirely, which also helps bundlers that can't prove the barrel's re-exports are unused.

The one thing a host must wire up

Search and the embedding index need to enumerate the site's pages, but how pages are enumerated is a host concern. Register a resolver once at startup, before the first search or embedding build:

import { registerDocEntriesSource } from '@thallylabs/core'

registerDocEntriesSource(() => myPages) // each page: { id, title, description, href, keywords }

Content parsing (getContentDocument, mdxToMarkdown) is self-contained and needs no registration.

License

MIT © Thally