How to use the Milvus plugin

Use the Milvus plugin to manage vector collections, upsert embeddings, run similarity search, and delete rows in Milvus, an open-source vector database, or Zilliz Cloud.

Authentication

All tasks inherit common connection properties from uri, token, username, password and databaseName.

• token is the API token for authentication, typically used with Zilliz Cloud or a Milvus instance with token-based auth enabled.
• username / password are credentials for a self-hosted Milvus instance with username/password authentication enabled.

token and password are alternative authentication methods, not to be combined — set one or the other depending on how your Milvus deployment is configured.

token and password are secrets, so they should never be hardcoded in a flow. Store them using Kestra's secrets and reference them with {{ secret('MILVUS_TOKEN') }}, or set them once for every Milvus task in a namespace via plugin defaults.

• databaseName: The Milvus database to operate against. Defaults to the server's default database when not set.

Tasks

• CreateCollection: Creates a collection with a single vector field and a primary key field, using Milvus's quick-setup schema (dimension, metricType, primaryFieldName, idType, autoId, vectorFieldName). For a custom multi-field schema, create the collection with the Milvus SDK or console first and use the other tasks against it directly.
• Upsert: Inserts or updates rows in configurable batch sizes (batchSize). Each row is a flat map with the collection's vector field, primary key field (unless auto-generated), and any scalar fields. Reads from an inline list or a Kestra internal storage URI.
• Search: Runs an approximate nearest-neighbor vector similarity search (vector, limit), optionally narrowed with a scalar filter expression (filter). Outputs results inline (FETCH), as a single record (FETCH_ONE), or to Kestra internal storage (STORE). Searching right after an Upsert or Delete may not reflect that write under the collection's default consistency level; set consistencyLevel: STRONG to always read the latest writes.
• Delete: Deletes rows by a list of primary key IDs (ids) or by a scalar filter expression (filter). Exactly one of the two must be set.
• DropCollection: Permanently deletes a collection and all of its rows. This cannot be undone.