LLM Wiki —
versioned,
self-improving
knowledge bases.

Multipart upload — PDF, DOCX, HTML, markdown. AI converts to markdown, commits to raw/docs/, and Claude Sonnet 4.6 synthesises pages in the background.

Frontmatter relations and inline [[Entity|relation]] wikilinks form a queryable graph. Walk it like data — what connects to what, what's orphaned, which cluster owns this idea.

Edit a source — every page that derives from it rewrites itself. Claude emits a partial diff with `// ... existing code ...` markers, Morph merges into the existing page server-side. ~10× fewer output tokens than a full rewrite.

Auto-generated llms.txt follows the llmstxt.org spec — pages, sources, summaries. Drop into any LLM context window or expose at your custom domain so other agents can ground on your wiki.

Every page edit is a real git commit — same branches, diffs and refs your engineering team already trusts. Browse the timeline, diff any two revisions, restore an old fact in one call.