01 / WORKFLOWImport. Inspect. Rehearse.
Select the documentation root so relative paths stay intact. Choose the entry page, review broken references and reverse links, then stage file moves or heading renames.
Confirm the mapping to download a plan. Make changes in your own editor, import a revised snapshot, and compare the expected destinations. Save a project JSON to resume later.
Snapshot JSON format
{"files":[
{"path":"README.md","text":"# Hello"},
{"path":"assets/logo.svg","size":163}
]}Only Markdown text is read. Other files are recorded by exact path and byte size. Project exports include original Markdown and asset metadata; keep them as private as your documentation.
Know the dialect.
CommonMark via remark-parse 11.0.0, GFM via remark-gfm 4.0.1, and github-slugger 2.0.0. Heading IDs use lowercase, punctuation removal and ordered duplicate suffixes. Unicode is retained where supported by that slugger.
Paths are case-sensitive. Root-relative paths resolve at the selected root. Percent encoding is decoded once; query strings do not affect file lookup. A fragment is checked against the exact generated ID. Code fences never create headings or links.
Inline links, images and defined references are inspected. Undefined Markdown references remain literal text under CommonMark. Shared definitions are one edit location. Images do not create navigation edges.
03 / LIMITS & PRIVACYUnknown means unchecked.
MDX, JS, raw HTML IDs/links, custom heading IDs, frontmatter, wiki links, templates, framework routes, redirects, extensionless routes and directory indexes are unsupported. No user code or HTML is run. External URLs are listed without requests. Assets are presence-checked only.
Up to 500 files, 250 KB and 20,000 lines per Markdown file, 5 MB Markdown / 20 MB total, 24 path levels, 64 AST levels, 100,000 nodes per file, 10,000 references, 12 MB project JSON, and an 8-second worker deadline.
Document text and filenames stay in memory until you download or close the page. No analytics, external fonts, paid APIs, keys or browser storage. Hosting may process ordinary page requests; no document data is sent. After loading, the workspace works offline.