Chroma Collection

{
  "$schema": "https://json-schema.org/draft/2020-12/schema",
  "$id": "https://trychroma.com/schemas/chroma/collection.json",
  "title": "Chroma Collection",
  "description": "A Chroma collection stores embeddings, documents, and associated metadata. Collections are the primary unit for organizing and searching vector data within a database.",
  "type": "object",
  "required": ["id", "name"],
  "properties": {
    "id": {
      "type": "string",
      "format": "uuid",
      "description": "The unique identifier of the collection, assigned by the server on creation"
    },
    "name": {
      "type": "string",
      "minLength": 1,
      "maxLength": 512,
      "pattern": "^[a-zA-Z0-9_-]+$",
      "description": "The name of the collection, used as a human-readable identifier within a database"
    },
    "metadata": {
      "type": ["object", "null"],
      "additionalProperties": true,
      "description": "Arbitrary key-value metadata associated with the collection, used for organizing and describing the collection's purpose and configuration"
    },
    "tenant": {
      "type": "string",
      "description": "The name of the tenant this collection belongs to"
    },
    "database": {
      "type": "string",
      "description": "The name of the database this collection belongs to"
    }
  },
  "$defs": {
    "CollectionConfiguration": {
      "type": "object",
      "description": "Configuration options that can be set in collection metadata to control embedding and indexing behavior",
      "properties": {
        "hnsw:space": {
          "type": "string",
          "enum": ["l2", "ip", "cosine"],
          "default": "l2",
          "description": "The distance function used for nearest neighbor search. l2 is Euclidean distance, ip is inner product, cosine is cosine similarity."
        },
        "hnsw:construction_ef": {
          "type": "integer",
          "minimum": 1,
          "default": 100,
          "description": "The size of the dynamic candidate list during HNSW index construction. Higher values improve recall at the cost of indexing speed."
        },
        "hnsw:search_ef": {
          "type": "integer",
          "minimum": 1,
          "default": 10,
          "description": "The size of the dynamic candidate list during search. Higher values improve recall at the cost of search speed."
        },
        "hnsw:M": {
          "type": "integer",
          "minimum": 2,
          "default": 16,
          "description": "The maximum number of bi-directional links per element in the HNSW graph. Higher values improve recall at the cost of memory."
        },
        "hnsw:num_threads": {
          "type": "integer",
          "minimum": 1,
          "default": 4,
          "description": "Number of threads to use during HNSW index construction"
        }
      }
    }
  }
}