The field data type - used for validation and filtering. date values are normalized to integer
epoch milliseconds at write time so range filters ($gt/$lt) work across vector store backends.
array is not yet supported and is rejected at knowledge-base upsert time.
OptionaldescriptionIn case that the field is required and not provided when uploaded to the knowledge base, this description will be used for extracting the value from the document. Also, it will be used for auto filtering when a prompt is sent to an AI agent.
OptionaldescriptionTrue when description was AI-generated (via generateMetadataFieldDescriptions). Undefined or
false means the description was authored by a human or has not been generated yet; it clears back to
false once a human edits the description.
OptionalllmWhether the LLM extraction pass may fill this field when it is missing. Defaults to true when
undefined, so knowledge bases created before this flag existed keep their behavior. Set to false
for fields whose value the caller supplies deterministically (e.g. folder-derived path metadata): a
missing value then stays absent rather than being guessed from the document.
The name of field.
Indicates if the field is required for knowledge base entries.
Represents a field in an AI knowledge base metadata.