Documentation

Embeddings & search

SystemSculpt pairs fast lexical scans with embeddings-backed similarity matches for every vault.

Embeddings manager

When you enable embeddings, the plugin prepares storage, your chosen provider, and background workers before it starts watching the vault for changes.

  • Embeddings jobs prepare storage, your chosen provider, and the background indexer before scanning the vault.
  • Each provider and model gets its own on-disk cache so switching setups will not overwrite previous runs.
  • The indexer works in steps that respect your rate limit and concurrency settings.
Processing safeguards
Built-in safeguards pause duplicate runs and add cooldowns when you swap models so the vault is indexed once.

Search experience

The modal ships with a blended ranking algorithm that pulls the best of both worlds.

  • The search modal blends keyword results from the vault with semantic matches from your embeddings.
  • Results update as you type, reordering files based on both their content and embeddings score.
  • The status bar shows progress signals whenever a background indexing job runs.
ShortcutThe command palette entry Open SystemSculpt Search (⌘/Ctrl + K) loads the modal and prefetches recent files.
DiagnosticsWhen embeddings are disabled, the modal gracefully falls back to lexical-only scoring.

Configuration tips

  • Enable embeddings, choose a provider, and set exclusions under Settings → Embeddings.
  • Rate limits and batch sizes directly control how much work runs in parallel.
  • Use Rebuild Embeddings (Current Model) whenever you switch models so vectors stay current.