Concepts

Workspace

A workspace holds your buckets, API keys, and members. It is the unit of access and billing. You start with one workspace and add members later.

Bucket

A bucket is a container for objects, like an S3 bucket. A smart bucket also indexes its objects for search and query. Set smart to false for plain storage.

Object

An object is a file plus its key. Spark accepts any file type: audio, video, documents, images, and structured records. Spark reads each object and builds the index that search and query use.

Tenant partition

A tenant partition isolates data by your own customer. Set the X-Spark-Tenant header on writes, search, and query. A query scoped to one tenant never reads another tenant's data.

Query modes

Spark returns data in the mode that fits the caller:

• Answer engine. A natural-language answer with citations to the source objects.
• Context retrieval. Ranked chunks for your own agent to use.
• Keyword search. A fast lexical match over objects.
• Vector search. A semantic match for RAG.

Citations

Every answer points at the objects that support it. Spark does not return an answer that hides its sources.