Zilliz Vector Store Load icon

Zilliz Vector Store Load

Search and load vectors from Zilliz vector database

Zilliz Vector Store Load

AI-generated

Overview

This node enables searching and loading vectors from a Zilliz vector database. It is designed to perform similarity searches by querying a specified collection within a Zilliz database using a given query vector. The node returns the most similar vectors based on the search criteria, which can be filtered and customized.

Common scenarios where this node is beneficial include:

  • Retrieving items similar to a given vector representation in recommendation systems.
  • Searching for nearest neighbors in high-dimensional vector spaces such as embeddings from machine learning models.
  • Filtering and fetching relevant data points from large vector datasets stored in Zilliz.

Practical example:

  • Given a user embedding vector, find the top 10 most similar product embeddings in a product catalog stored in Zilliz, optionally filtering by category or other metadata fields.

Properties

Name Meaning
Database Name Name of the Zilliz database to query (default is "default").
Collection Name Name of the Zilliz collection to use. Must start with a letter and contain only letters, numbers, underscores.
Query Vector Vector to search for; either an explicit array of numbers (e.g., [0.1, 0.2, 0.3]) or a field reference containing the vector.
Top K Number of most similar vectors to return (integer).
Options Additional options including:
- Filter Expression: A filter expression to apply (e.g., ID > 100 and category == 'tech').
- Output Fields: Comma-separated list of fields to return (default is all fields).
- Score Threshold: Minimum similarity score threshold between 0 and 1.

Output

The node outputs an array of JSON objects representing the search results. Each result includes:

  • id: Identifier of the matched vector.
  • score: Similarity score (distance) of the match.
  • Other fields returned from the collection entity as specified by the output fields option.
  • _metadata: An object containing:
    • collection: The collection name used.
    • database: The database name used.
    • searchVector: The query vector used for the search.

If no results are found matching the criteria, the output contains a single JSON object with a message indicating "No results found" along with the search vector, collection, and database info.

If the node encounters errors and is configured to continue on failure, it outputs an error message per item.

The node does not output binary data.

Dependencies

  • Requires connection to a Zilliz vector database cluster.
  • Needs an API key credential and cluster endpoint configuration for authentication.
  • Uses an internal Zilliz client library to perform vector search operations.

Troubleshooting

  • Invalid Query Vector: If the query vector is not a valid JSON array or is empty, the node throws an error stating "Query vector must be a non-empty array of numbers." Ensure the input vector is correctly formatted as an array of numbers or a valid field reference.
  • Filter Expression Errors: Incorrect filter expressions may cause the search to fail or return no results. Verify the syntax and field names used in the filter.
  • No Results Found: This indicates that no vectors matched the search criteria or passed the score threshold. Adjust the query vector, filters, or lower the score threshold if needed.
  • Credential Issues: Missing or invalid API key or cluster endpoint will prevent the node from connecting to the Zilliz service. Confirm credentials are properly set up.
  • Continue On Fail Behavior: When enabled, errors for individual items do not stop execution but are returned as error messages in the output.

Links and References

Discussion