Local Blob / Vector Database

Overview

This article explains the local_blob/vector database storage layout and how to inspect it.

Folder Layout

local_blob/
  vector database/
    bmo_products/
    _vectrixdb.db

• bmo_products/: collection-level artifacts and snapshots.
• _vectrixdb.db: local metadata database for collection registry/state.

Script Code: Snapshot Persistence

These functions from product collections/create_collection.R control local blob writes:

collection_output_dir <- function(data_path) {
  normalized <- normalizePath(data_path, winslash = "/", mustWork = FALSE)
  if (tolower(basename(normalized)) == tolower(COLLECTION_NAME)) {
    return(normalized)
  }
  file.path(normalized, COLLECTION_NAME)
}

persist_collection_snapshot <- function(chunks, data_path) {
  out_dir <- collection_output_dir(data_path)
  if (!dir.exists(out_dir)) {
    dir.create(out_dir, recursive = TRUE)
  }

  saveRDS(chunks, file.path(out_dir, "chunks.rds"))
  out_dir
}

Script Code: Local Metadata Update

The local metadata DB update logic:

update_collection_tags_at_path <- function(data_path) {
  db_path <- file.path(data_path, "_vectrixdb.db")
  if (!file.exists(db_path)) return(FALSE)
  if (!requireNamespace("DBI", quietly = TRUE) || !requireNamespace("RSQLite", quietly = TRUE)) return(FALSE)

  con <- DBI::dbConnect(RSQLite::SQLite(), db_path)
  on.exit(DBI::dbDisconnect(con), add = TRUE)

  row_count <- DBI::dbGetQuery(
    con,
    "SELECT COUNT(*) AS n FROM collections WHERE name = ?",
    params = list(COLLECTION_NAME)
  )$n[[1]]

  if (is.null(row_count) || row_count == 0) return(FALSE)

  tags_json <- make_tags_json(COLLECTION_TAGS)
  DBI::dbExecute(
    con,
    "UPDATE collections SET tags = ?, description = ? WHERE name = ?",
    params = list(tags_json, COLLECTION_DESCRIPTION, COLLECTION_NAME)
  )
  TRUE
}

Open and Inspect

db <- VectrixDB::Vectrix$new(
  name = "bmo_products",
  path = "local_blob/vector database"
)

db$count()

Query Example

results <- db$search(
  query = "best cashback on groceries",
  mode = "hybrid",
  limit = 5
)

results$top()

Operational Notes

• Keep this path stable across scripts and dashboard runs.
• Back up the vector database folder if you need portable local state.