Hierarchical Taxonomy
Multi-level classification tree with parent-child relationships for complex categorization
Why do anything?
Complex domains need multi-level categorization. Flat taxonomies can't represent category hierarchies.
Why now?
Enterprise content organization requires structured hierarchies for navigation and filtering.
Why this feature?
Hierarchical taxonomy with parent-child relationships. Supports LLM inference, clustering-based, or hybrid discovery.
How It Works
Hierarchical taxonomy provides multi-level classification with tree navigation.
1
Root Classification
Classify into top-level category
2
Tree Navigation
Recursively classify down the tree
3
Path Assignment
Assign full path from root to leaf
4
Enrichment
Add hierarchy path to document payload
Why This Approach
Tree navigation enables drill-down filtering. Multiple inference strategies support different use cases.
Integration
taxonomy = client.taxonomies.create(type="hierarchical", hierarchy={...})Recent updates
Full changelog- Jul 24, 2026Studio renders multi-level taxonomy treesThe taxonomy detail view in Studio flattened hierarchical taxonomies, so a taxonomy with real parent and child depth showed as a single level. It now renders the full nested tree, so multi-level taxonomies display with their actual structure.
- Jul 23, 2026New explainer: turning a frozen embedding model into an open-vocabulary image taggerA new video walks through getting open-vocabulary image tagging out of a frozen embedding model with no fine-tuning, no classifier head, and no curated tag list, scoring every patch against the model's own vocabulary instead. The measured progression is the point: the global vector alone reaches 0.26 mAP on COCO-150, max-over-patches reaches 0.64, and re-encoding crops reaches 0.71. The takeaway generalizes past tagging: calibration tricks barely moved the result, while giving the model new input (patches, then crops) added 37 and then 7 more mAP points. It is the same shape as how Mixpeek taxonomies run classification.