vrcatalogue/vrcatalogue-open-data

VRCatalogue Open Data

Data from VRCatalogue, a search engine for VRChat assets on BOOTH that shut down on 31 October 2026. It covers avatars, item ↔ avatar compatibility, categories, tags and search embeddings for about 216k BOOTH items. Items are identified by BOOTH item ID. Titles, descriptions and images are not included. ARCHITECTURE.md shows what the site ran on.

Files

Path	Contents	Rows
vrcatalogue.sqlite	All tables except embeddings
data/avatars.parquet	Base avatars	380
data/avatar_aliases.parquet	Alternative avatar names	608
data/items.parquet	BOOTH items	216k
data/item_avatar_compatibility.parquet	Item ↔ avatar compatibility	341k
data/categories.parquet	Categories	9
data/subcategories.parquet	Subcategories	109
data/item_categories.parquet	Item → subcategory	244k
data/unified_tags.parquet	Merged tags with en/ja/ko names	504
data/tags.parquet	Tags as written on BOOTH	122k
data/item_tags.parquet	Item → tag	2.9M
embeddings/*.parquet	Search vectors (3.3 GB, 11 files)	216k

The SQLite file and the Parquet files hold the same tables. The SQLite file also has two views: compatible_items and item_category_names. Timestamps are UTC, and SQLite booleans are 0/1. In the diagram below, --< means one-to-many.

Tables

items
 |
 +--< item_avatar_compatibility >-- avatars --< avatar_aliases
 |
 +--< item_categories >-- subcategories >-- categories
 |
 +--< item_tags >-- tags >-- unified_tags
 |
 +--- embeddings

avatars: avatar_id, slug, name_en, name_ja, name_ko, gender (female/male/unknown), booth_item_id (the avatar's own listing), compatible_item_count.

avatar_aliases: avatar_id, alias.

items: item_id, booth_url, shop_name, shop_url, booth_category, price_jpy, likes, is_adult, status (ok/delisted), published_at, content_rating (sfw/suggestive/nsfw), quest_compatible, uses_liltoon, uses_poiyomi, uses_vrcfury, uses_modular_avatar (all yes/no/n/a). The rating, Quest and uses_* values were labelled automatically and may contain mistakes.

item_avatar_compatibility: item_id, avatar_id, verified. When verified is true, the match was confirmed against the listing. When false, the listing mentions the avatar but the match wasn't confirmed.

categories: category_id, slug, name_en, name_ja, name_ko.

subcategories: subcategory_id, category_id, slug, name_*, and an optional group_* such as Outfits or Weapons.

item_categories: item_id, category_id, subcategory_id, is_primary. Each item has one primary subcategory and may have others.

tags: tag_id, name, unified_tag_id, item_count.

unified_tags: unified_tag_id, slug, name_*, type (general/avatar), avatar_id. A unified tag merges spelling and language variants of a tag.

item_tags: item_id, tag_id.

embeddings: item_id, title_vec, tags_vec, bio_vec. Each vector is float32[1536], made from the item's original title, its tags joined with ", ", and its description. Model: OpenAI text-embedding-3-large with dimensions=1536. Vectors are normalised, so compare them with cosine similarity or dot product. See examples/semantic_search.py.

Examples

Hair compatible with Manuka (SQLite):

SELECT ic.booth_url, ic.subcategory_en AS style, i.likes
FROM compatible_items ci
JOIN item_category_names ic USING (item_id)
JOIN items i USING (item_id)
WHERE ci.avatar_slug = 'manuka' AND ic.category_slug = 'hair' AND i.status = 'ok'
ORDER BY i.likes DESC;

Avatars by outfit count (DuckDB on Parquet):

SELECT a.name_en, count(*) AS outfits
FROM 'data/item_avatar_compatibility.parquet' c
JOIN 'data/avatars.parquet' a USING (avatar_id)
JOIN 'data/item_categories.parquet' ic USING (item_id)
JOIN 'data/categories.parquet' cat USING (category_id)
WHERE cat.slug = 'clothing' AND ic.is_primary
GROUP BY 1 ORDER BY 2 DESC LIMIT 20;

Source data

All original item data comes from BOOTH, which is run by pixiv Inc. That includes item IDs, shop names, tags, prices and likes. Items and their listings belong to their creators. This dataset is not affiliated with or endorsed by BOOTH or pixiv Inc.

Thank you

Thank you to everyone who used VRCatalogue, 7 March 2026 – 31 October 2026.