Setting categories

Categories are free text on the way in, resolved against your real category list on the way out. Send what a person would say ("coffee", "groceries", "the vet") rather than an exact category ID.

How resolution works

Every category can optionally have a parent. Matching considers both top-level categories and subcategories.

• match_type: "exact" — a case-insensitive exact category-name match
• match_type: "fuzzy" — close but not exact ("coffe" → Coffee & Cafes), with match_confidence
• match_type: "fallback" — below 0.7 confidence, routed to Other with suggested_categories

For a guaranteed exact match, call list_categories first and send one returned name verbatim.

Building your own categories

Every workspace starts with 12 top-level categories and their subcategories. This is a starting point, not a closed list: create your own category or subcategory any time and it becomes available for resolution.

For agents

Do not guess a category ID or invent an obvious duplicate. Send the natural-language category and trust the resolution. If it lands in Other, tell the person rather than silently choosing something close.