Schema designer
Every registered schema, every ROW REF — laid out as a graph. Click a node to inspect its columns, indexes, and edges.
New table
Vector embeddings + full-text search are configured separately (not table columns) — see Index config. Enum variants, decimal precision, FK actions — switch to TOML.
Edit the TOML manifest directly. The engine validates it and the graph reloads with the new node.
Full-text search config
Per field: synonym expansion + stopword overrides. Each save replaces the whole list for that field.
Vector index config
Vector tables are created + configured on first insert — dim, metric, index family and quantization ride the write (no separate create step yet). Pick the settings; the exact insert call updates below.
—
IVF / IVF-PQ require trained centroids installed before the first insert — train them below. HNSW needs no pre-step.
Trains on the embeddings table's existing vectors (mini-batch k-means). Uses the table name from above.