generated: '2026-07-21'
method: searched
source: https://docs.topk.io/concepts + https://docs.topk.io/collections + https://docs.topk.io/datasets
description: >-
  TopK exposes two primary storage abstractions: Collections (structured documents with typed,
  indexed fields for hybrid search) and Datasets (unstructured document files ingested for
  semantic search and grounded question answering). Derived from the documented concepts and
  API surface; no OpenAPI is published.
entities:
- name: Collection
  description: A named container of structured documents with a typed schema. Fields can carry keyword (BM25), semantic (multi-vector), and dense/sparse vector indexes, plus metadata for filtering. Supports on-demand isolated partitions (namespaces) for multi-tenancy.
  key: name
  operations:
  - create
  - list
  - get
  - delete
  relationships:
  - has_many: Document
    via: collection
- name: Document
  description: A record stored in a collection, keyed by `_id`, with schema-typed fields (text, struct, vectors, metadata). Written via upsert/update/delete and read via query or get-by-id.
  key: _id
  operations:
  - upsert
  - update
  - delete
  - query
  - get
  relationships:
  - belongs_to: Collection
    via: collection
- name: Dataset
  description: A container of ingested unstructured document files (PDF, Markdown, HTML, and more) used for semantic search and grounded question answering (Ask).
  key: name
  operations:
  - create
  - list
  - get
  - update
  - delete
  - ingest
  - search
  - ask
  relationships:
  - has_many: DatasetDocument
    via: dataset
- name: DatasetDocument
  description: An ingested source file within a dataset; passages are retrieved during search and cited in grounded answers.
  relationships:
  - belongs_to: Dataset
    via: dataset
- name: Partition
  description: An on-demand, fully isolated namespace within a collection enabling multi-tenancy.
  relationships:
  - belongs_to: Collection
    via: collection