Parameters
Create a collection from scratch
The simplest way to create a collection is to provide the collection name and the field index configurations.201 Created. Collection creation also creates the default branch, main:
source.asOf, use numeric Unix epoch milliseconds.
The create response is intentionally compact. Use describe or list to retrieve indexConfigs, document counts, and the full collection model. 201 confirms catalog creation; it does not guarantee query-serving readiness or completed indexing. Handle loading and retry states using actual operations; there is no collection-status readiness field to poll.
Create a collection with managed embeddings
Managed embeddings let LambdaDB derive vector values from a source text field. Define the vector field withmanagedEmbedding: true and an embedding block. The source field must be a text field in the same collection.
For the current provider and model matrix, see Managed embeddings.
For managed embedding vector fields, do not send top-level
dimensions or similarity in the field config. LambdaDB resolves and stores those values under embedding.Create a partitioned collection
LambdaDB supports hash-based partitioning for a specified field.Currently, only
keyword type is supported for partitioning.Snapshot retention
Every collection retains committed snapshots for 30 days by default. SetsnapshotRetentionInDays to an integer from 1 through 31 when creating the collection, or change it later with the collection update API.
Omitting retention on create selects the default; explicitly sending null is rejected. See Collection PATCH rules for update and metadata-tag semantics.
Point-in-time history is available by creating a branch rather than cloning another collection. A source.asOf value is a Unix epoch timestamp in milliseconds and must refer to a branch in the same collection. See Branches, tags, and aliases.