Stable identifier for the document (chunk IDs will derive from this).
Raw text that will be chunked and embedded.
Optional dataOptional override for which data source / collection to push this document into.
Optional sourceOriginal source pointer (URL, file path, API, etc.).
Optional metadataArbitrary metadata stored alongside chunks; values must be vector-store friendly.
Optional languageISO language tag for the content.
Optional timestampISO timestamp describing when this content was produced/updated.
Optional embeddingOptional pre-computed embedding vector.
Optional embeddingIdentifier of the embedding model used when embedding is supplied.
Optional tenantTenant the document belongs to. Used for multi-tenant isolation.
Optional aclACL groups allowed to retrieve this document. Empty/undefined means the document inherits the data source's default ACL (typically unrestricted within the tenant).
Optional classificationSensitivity classification. Drives default redaction policy and audit obligations.
Optional statusLifecycle status. active chunks are returned by default;
draft/archived/deprecated are excluded unless the retrieval
options opt them in.
Optional effectiveISO timestamp: when this document became authoritative.
Optional expiresISO timestamp: when this document stops being authoritative.
Represents raw document content provided for ingestion.