Skip to content

Fastest way to scaffold your LLM powered personal knowledge base - Obsidian, Claude Code, qmd, Github

1.5 KB
Raw Blame
/**
 * Re-index all collections and regenerate vector embeddings.
 * Run after bulk wiki ingest sessions (not after every single file edit).
 * First run downloads ~2GB of GGUF models — expected, one-time.
 *
 * Run from vault root:
 *   pnpm qmd:reindex
 *
 * Suitable as a cronjob:
 *   0 * * * * cd /home/user/my-brain && pnpm qmd:reindex
 */

import { createStore } from "@tobilu/qmd"
import { mkdir } from "node:fs/promises"
import { dirname, join } from "node:path"
import { fileURLToPath } from "node:url"

const VAULT = join(dirname(fileURLToPath(import.meta.url)), "../..")
const DB = join(VAULT, ".qmd", "index.sqlite")

try {
  await mkdir(dirname(DB), { recursive: true })
  const store = await createStore({ dbPath: DB })

  console.log("Updating file index...")
  const result = await store.update({
    onProgress: ({ collection, file, current, total }: any) => {
      process.stdout.write(`\r  [${collection}] ${current}/${total} ${file}`.padEnd(80))
    },
  })
  console.log(
    `\n  indexed: ${result.indexed} | updated: ${result.updated} | ` +
    `removed: ${result.removed} | needs embedding: ${result.needsEmbedding}`
  )

  console.log("Generating embeddings...")
  await store.embed({
    onProgress: ({ current, total, collection }: any) => {
      process.stdout.write(`\r  [${collection}] ${current}/${total}`.padEnd(60))
    },
  })
  console.log("\nDone.")
} catch (err: any) {
  console.error(`\n[qmd:reindex] ${err?.message || err}`)
  console.error(`  vault: ${VAULT}`)
  console.error(`  db:    ${DB}`)
  process.exit(1)
}