Skip to content

Search Vector Embeddings¤

Python Plugin

This operator is part of a Python Plugin Package. In order to use it, you need to install it, e.g. with cmemc.

This workflow task searches for the top-k nearest vectors stored in a collection in Postgres Vector Store (PGVector).

Each incoming entity supplies a query embedding on the single input path. The task compares it against the vectors in the configured collection and attaches the top-k nearest matches to the entity as a JSON array on an additional output path, together with each match’s metadata, its stored embedding source, and its distance to the query.

The result is structured like this:

[
{
   "id": "...",
   "metadata": "..",
   "_embedding_source": "..",
   "distance": ".."
}
...
]

Use together with Store Vector Embeddings, which populates the collection this task searches.

Parameter¤

Database Host¤

The hostname of the postgres database service.

  • ID: host
  • Datatype: string
  • Default Value: pgvector

Database Port¤

The port number of the postgres database service.

  • ID: port
  • Datatype: Long
  • Default Value: 5432

Database User¤

The account name used to login to the postgres database service.

  • ID: user
  • Datatype: string
  • Default Value: pgvector

Database Password¤

The password of the database account.

  • ID: password
  • Datatype: password
  • Default Value: None

Database Name¤

The database name.

  • ID: database
  • Datatype: string
  • Default Value: pgvector

Collection Name¤

The name of the collection in Postgres Vector Store (PGVector).

  • ID: collection_name
  • Datatype: string
  • Default Value: None

Search Result Path¤

The path containing the search result in the output entities.

  • ID: search_result_path
  • Datatype: string
  • Default Value: _search_result

Embedding Query Path¤

The path containing the query embedding to search with. Its value must be a string representation of a list of floats, e.g. [0.1, 0.2, ...]; the task fails if it cannot be parsed as one.

  • ID: embedding_query_path
  • Datatype: string
  • Default Value: _embedding

Top-k¤

The number of entries to be returned in the search result.

  • ID: top_k
  • Datatype: Long
  • Default Value: 10

Advanced Parameter¤

Distance Strategy¤

The distance strategy used to rank search matches.

  • ID: distance_strategy
  • Datatype: enumeration
  • Default Value: COSINE

Comments