Qdrant: Upsert

Stream a JSONL file and upsert each row into a Qdrant collection as a point. Each row contributes the point id, its dense vector (and optional sparse vector for hybrid collections), and any selected payload metadata.

For storing vectors via the legacy embeddings format see AI::QdrantVectorEmbed.

Pre-requisite: Install Qdrant application in Profile > {Organization} > Applications. The target collection must already exist with vector configuration matching Mode.

Parameters

Application—REQUIRED

Configured Qdrant application.

Collection—REQUIRED

Target Qdrant collection. Must already exist.

Mode—REQUIRED

Storage topology:

• Single Embedding — plain dense vectors only.
• Hybrid Embedding — named dense + sparse vectors (collection must be configured for hybrid).

Dense Vector Name

Hybrid mode only — name of the dense vector. Defaults to dense.

Sparse Vector Name

Hybrid mode only — name of the sparse vector. Defaults to sparse.

Sparse Method

Hybrid mode only — algorithm for deriving sparse vectors when not provided explicitly (BM25, default).

Sparse Vector Attribute

Hybrid mode only — field on each input row holding the sparse vector (shape { "indices": [u32, ...], "values": [f32, ...] }). When missing the row contributes an empty sparse vector.

ID Attribute—REQUIRED

Field on each row used as the Qdrant point id. Strings that parse as u64 are sent as numeric ids; everything else is sent as a string id.

Vector Attribute—REQUIRED

Field on each row holding the dense vector (array of numbers). Rows missing this field, or whose value is not a non-empty array, are skipped.

Payload Attributes

List of field names from each row to copy into the Qdrant payload (the metadata stored alongside the vector). Other fields on the row are dropped.

Input

File—REQUIRED

JSONL file. Each line contributes one Qdrant point.

Output

Trigger

Fires once the upsert completes successfully.