Skip to main content

Key Concepts

This page defines the terms used throughout the Knowledge Enrichment documentation. All terms refer specifically to Knowledge Enrichment behavior unless otherwise noted.


Action

An action is a named enrichment operation that the Context API can perform on content. Actions are selective; you choose which actions to run per request. Examples include textSummarization, textClassification, imageDescription, and namedEntityRecognitionText.

See Context API actions reference for the full list with input schemas.

Asynchronous Processing

Both of the core APIs process requests asynchronously. When you submit a request, the API immediately returns a job identifier (processingId for Context API, job_id for Data Curation API). You then poll a status or results endpoint until processing is complete.

Consider the following terminal states:

Terminal StateDescription
Done / SUCCESSProcessing completed successfully.
FAILURE / PARTIAL_FAILUREOne or more actions failed. Results contain per-action error details.
Wait For Upload (DC API only)The presigned URL has been issued, but the file has not yet been uploaded.

Chunks

In Knowledge Enrichment, chunks are semantically coherent segments of text extracted from a document during the data curation process. Each chunk includes positional metadata (start/end offsets) and optionally an embedding vector. Chunk size and overlap are configurable.

Content Lake

The Content Lake is the persistent, per-environment isolated storage for curated and enriched content. The DC API writes processed output to the Content Lake. The Context API reads from the Content Lake when processing already-curated content.

Guardrails

Guardrails are AI safety controls that monitor Context API enrichment operations and block responses containing harmful, offensive, or policy-violating content. When a guardrail is triggered, the API returns a structured error instead of a partial result.

See Guardrails for details and example violation responses.

Presigned URL

A presigned URL is a time-limited, pre-authenticated URL that allows a client to upload a file directly to cloud storage without exposing storage credentials. This is a consumer-facing step, not an internal infrastructure detail. Both core APIs use presigned URLs as the upload mechanism: the client requests a presigned URL from the API, and then it performs a PUT request to that URL with the file as binary data.

The objectKey returned alongside the presigned URL uniquely identifies the uploaded file and is used in subsequent processing requests.