vocabulary: LanceDB
version: '0.1'
description: >-
  Operational and capability vocabulary for the LanceDB multimodal lakehouse,
  covering Lance Namespace REST operations, table primitives, index types,
  search modes, embedding sources, and lakehouse lifecycle concepts.
maintainer: API Evangelist
references:
  - https://lance.org/format/
  - https://lance.org/format/namespace/
  - https://docs.lancedb.com/

concepts:
  - id: Namespace
    label: Namespace
    definition: Hierarchical container for Lance tables, materialized views, and child namespaces.
  - id: Table
    label: Table
    definition: A Lance-formatted columnar dataset addressed within a namespace.
  - id: Fragment
    label: Fragment
    definition: An immutable subset of a Lance table's data, the unit of compaction and read.
  - id: Field
    label: Field
    definition: A named, typed column in a Lance table, possibly an embedding vector or list of embeddings.
  - id: Version
    label: Version
    definition: A monotonically increasing snapshot identifier for time-travel and reproducibility.
  - id: Tag
    label: Tag
    definition: A named alias pointing at a specific table version.
  - id: Index
    label: Index
    definition: A persisted secondary structure accelerating filter or nearest-neighbor queries.
  - id: MaterializedView
    label: Materialized View
    definition: A precomputed projection or aggregation refreshed via an explicit refresh operation.
  - id: Transaction
    label: Transaction
    definition: A multi-operation unit that can be committed atomically across tables.
  - id: EmbeddingFunction
    label: Embedding Function
    definition: A pluggable model that converts text or media into vectors at insert and query time.
  - id: Reranker
    label: Reranker
    definition: A second-stage scorer that combines vector and full-text candidate sets.

indexTypes:
  - id: BTREE
    description: Ordered scalar index for range and equality predicates.
  - id: BITMAP
    description: Low-cardinality scalar index for set membership.
  - id: LABEL_LIST
    description: Index for array-of-label columns supporting `contains` and `any-of` filters.
  - id: FTS
    description: BM25 full-text search index backed by Tantivy.
  - id: IVF_FLAT
    description: Inverted-file vector index with exact distance scoring.
  - id: IVF_PQ
    description: Inverted-file vector index with product quantization.
  - id: IVF_HNSW_SQ
    description: HNSW-on-IVF vector index with scalar quantization.
  - id: IVF_HNSW_PQ
    description: HNSW-on-IVF vector index with product quantization.

searchModes:
  - id: vector
    description: Approximate nearest-neighbor search over an embedding column.
  - id: fullText
    description: BM25 full-text search using an FTS index.
  - id: hybrid
    description: Combined vector + FTS search with optional reranking.
  - id: sql
    description: Filter / project / aggregate via Arrow SQL.
  - id: multivector
    description: Late-interaction search over multivector columns (e.g. ColBERT-style).

distanceMetrics: [l2, cosine, dot]

storageBackends:
  - aws-s3
  - google-cloud-storage
  - azure-blob-storage
  - local-disk
  - http-object-store

operations:
  namespace: [Create, Describe, Drop, Exists, List]
  table: [Create, Describe, Exists, Drop, Register, Deregister, Rename, Restore, Stats, CountRows]
  data: [Insert, MergeInsert, Update, Delete, Query]
  schema: [AddColumns, AlterColumns, DropColumns, BackfillColumns, UpdateSchemaMetadata]
  version: [List, Create, Describe, BatchCreate, BatchDelete, BatchCommit]
  index: [Create, CreateScalar, List, DescribeStats, Drop]
  tag: [List, GetVersion, Create, Update, Delete]
  transaction: [Describe, Alter]
  materializedView: [Create, Refresh]

embeddingProviders:
  - OpenAI
  - Cohere
  - Anthropic
  - Jina
  - Hugging Face
  - Sentence Transformers
  - Ollama
  - Amazon Bedrock
  - Voyage AI

lifecycle: [draft, published, deprecated, sunset]