Vector Clustering
Discover document groups using HDBSCAN and other algorithms on embedding vectors
Why do anything?
Large document collections have hidden structure. Without clustering, you can't discover natural groupings.
Why now?
AI enables automatic pattern discovery. Manual categorization misses emergent themes.
Why this feature?
8 clustering algorithms (HDBSCAN, K-Means, DBSCAN, etc.) with LLM-powered cluster labeling and dimensionality reduction.
How It Works
Vector clustering discovers document groups using embedding similarity.
1
Dimensionality Reduction
Optional t-SNE or UMAP for visualization
2
Clustering
Apply selected algorithm (HDBSCAN, K-Means, etc.)
3
Labeling
LLM generates descriptive cluster labels
4
Assignment
Assign cluster IDs to documents
Why This Approach
HDBSCAN handles variable density clusters. LLM labeling provides human-readable cluster names.
Where This Is Used
Integration
cluster = client.clusters.create(algorithm="hdbscan", min_cluster_size=5)
Recent updates
Full changelog- Aug 9, 2026Clustering accepts `enrich_source`, the field name it reads back to youFetching a cluster returned a field called `enrich_source`, while creating one required `enrich_source_collection`. Copying the name off a cluster you had just read meant the field was dropped on the way in, and enrichment quietly never ran. The run still finished as COMPLETED with its labeled clusters, and no source document carried any cluster fields, so nothing in the result told you what had gone missing. Create now accepts `enrich_source` as an alias for the same setting. Setting the canonical field to `false` still wins over an alias, so turning enrichment off keeps it off.
- Aug 9, 2026A clustering run submitted during an engine deploy now waits the deploy outSubmitting a cluster while the engine was rolling hit a deploy lock that asks callers to wait 120 seconds, and the retry envelope gave up after roughly 130 seconds in total. Any submission landing in a real cluster roll ran out of attempts before the engine came back. The task document was then left at IN_PROGRESS with nothing ever moving it, because the code meant to record the failure could not be reached. Retries now start at a 120 second floor and stretch across about 44 minutes, which outlasts a roll, and a submission that genuinely exhausts them is written as FAILED so you can see it and send it again.
- Jul 24, 2026Lasso and box selection restored in the cluster explorerIn the Studio cluster visualization, documents mode had lost its lasso and box selection tools, so you could not draw a region to grab a group of points. Both tools are back, so you can select clusters of documents directly on the plot again.
- Jul 24, 2026Clustering accepts more natural config shapesTwo clustering config papercuts are now handled for you. If you pass an `attribute_config` without setting `cluster_type`, the type is inferred as `attribute` instead of erroring. And a `multi_feature_strategy` placed at the root of the request is lifted into the `vector_config` where it belongs, rather than being ignored. Both make the clustering API more forgiving of the shapes people and agents naturally send.