Metadata handling

Metadata travels from indexable documents into vector payloads, retrieved documents, prompt context and frontend sources.

Important metadata fields

source_type: Type of source, for example page, file or product.
source_identifier: Stable identifier of the source object.
title: Human-readable source title.
url: Link to the source if available.
changed_at: Timestamp or date used for freshness-sensitive content.
Domain-specific metadata: Examples are product number, category, department, contact name or page uid.

Prompt metadata

Only selected metadata fields are written into the prompt. Configure prompt_metadata_fields per step to avoid bloating prompts.

Frontend metadata

The same prompt_metadata_fields selection is exposed as frontend source metadata. Include only fields that are useful and safe for both consumers.