LLM Wiki is a cross-platform desktop application that turns your documents into an organized, interlinked knowledge base — automatically. Instead of traditional RAG (retrieve-and-answer from scratch every time), the LLM incrementally builds and maintains a persistent wiki from your sources。
TypeScript
/**
* Convert Obsidian-style `[[target]]` and `[[target|alias]]` wiki
* links inside a markdown body to standard markdown links so a
* commonmark renderer (Milkdown / ReactMarkdown) styles them as
* links instead of dumping the bracket syntax as raw text.
*
* Output format: `[label](#target)` — using a fragment href so
* Tauri's webview doesn't try to navigate externally on click.
* In-app navigation can be wired up later via a click intercept
* on `<a href="#…">` elements; for now the goal is just to stop
* the wikilinks from looking like "raw code".
*
* Skips content inside fenced code blocks (```…```) and inline
* code spans (`…`) so wikilinks shown as code examples in
* documentation don't get mangled.
*/
/**
* Convert Obsidian-style image/file embeds `![[target]]` (and
* `![[target|alias]]`) into standard CommonMark image syntax
* `` so a commonmark renderer actually displays them.
*
* Obsidian's embed syntax is a superset of CommonMark that react-
* markdown does NOT understand — left untouched, `![[../assets/x.png]]`
* renders as literal text. The feishu-to-md export skill emits this
* form by default ("Obsidian mode"), so without this rewrite every
* exported image is invisible in the app.
*
* The `target` is preserved verbatim as the image URL (NOT URL-
* encoded) — `resolveMarkdownImageSrc` expects raw filesystem-style
* relative paths like `../assets/x.png` or `media/slug/img.png`.
* The alias (after `|`), if present, becomes the alt text.
*
* Must run BEFORE `transformWikilinks`, otherwise the generic
* `[[…]]` → `[label](#frag)` rule would mangle the embed target
* into a fragment link.
*
* Skips fenced code blocks and inline code spans so documentation
* showing the syntax literally isn't rewritten.
*/
export function transformImageEmbeds(body: string): string {
if (!body.includes("![[")) return body
const parts = body.split(/(```[\s\S]*?```)/g)
return parts
.map((part, idx) =>
idx % 2 === 1 ? part : transformImageEmbedsOutsideCode(part),
)
.join("")
}
const IMAGE_EMBED_RE = /!\[\[([^\]|\n]+)(?:\|([^\]\n]*))?\]\]/g
function transformImageEmbedsOutsideCode(text: string): string {
if (!text.includes("![[")) return text
const parts = text.split(/(`[^`\n]+`)/g)
return parts
.map((part, idx) => (idx % 2 === 1 ? part : replaceImageEmbeds(part)))
.join("")
}
function replaceImageEmbeds(text: string): string {
return text.replace(
IMAGE_EMBED_RE,
(_match, rawTarget: string, rawAlias?: string) => {
const target = rawTarget.trim()
const alias = rawAlias?.trim() ?? ""
// Sanitize alt text so `]` doesn't terminate the markdown image
// alt bracket early.
const alt = alias.replace(/]/g, ")")
// Wrap the URL in <…> so spaces / parens / non-ASCII in the
// path don't break the CommonMark image parser. The resolver
// strips no angle brackets — react-markdown removes them while
// parsing, so `src` arrives clean.
return ``
},
)
}
export function transformWikilinks(body: string): string {
if (!body.includes("[[")) return body
// Split on triple-backtick fences. The capturing group keeps
// the fence content in the output. Odd indices are inside a
// fence and must pass through untouched.
const parts = body.split(/(```[\s\S]*?```)/g)
return parts
.map((part, idx) => (idx % 2 === 1 ? part : transformOutsideCode(part)))
.join("")
}
const WIKILINK_RE = /\[\[([^\]|\n]+)(?:\|([^\]\n]*))?\]\]/g
function transformOutsideCode(text: string): string {
if (!text.includes("[[")) return text
// Split on inline-code spans so backticked content is preserved.
const parts = text.split(/(`[^`\n]+`)/g)
return parts
.map((part, idx) => (idx % 2 === 1 ? part : replaceWikilinks(part)))
.join("")
}
function replaceWikilinks(text: string): string {
return text.replace(WIKILINK_RE, (_match, rawTarget: string, rawAlias?: string) => {
const target = rawTarget.trim()
const alias = rawAlias?.trim() ?? ""
const label = alias.length > 0 ? alias : target
// Encode the target so spaces / parens / hashes don't break the
// markdown link parser. encodeURIComponent is overkill for a
// fragment but it's the safe default.
const href = `#${encodeURIComponent(target)}`
// Escape any closing brackets in the label that would otherwise
// terminate the markdown link text.
const escapedLabel = label.replace(/\[/g, "\\[").replace(/\]/g, "\\]")
return `[${escapedLabel}](${href})`
})
}