Amazon S3 annotations attach up to 1 GB of mutable context per object

Amazon Web Services has introduced annotations, a new metadata feature for its S3 object storage that lets customers attach rich business context directly to stored items.

S3 annotations support up to 1,000 named entries per object. Individual entries can each measure 1 MB, for a combined maximum of 1 GB. Supported formats include JSON, XML, YAML and plain text. Any annotation can be edited or removed without altering the original object.

The capability targets AI agents and autonomous workflows that locate, interpret and operate on data independently. Annotations remain synchronized as information changes, extend across petabyte-scale collections and stay directly queryable without external stores.

Annotations travel with objects across operations. Context automatically follows during copy, replication and cross-region transfers. The service deletes annotations upon object removal.

Enabling S3 Metadata routes annotations into managed tables. These can be queried through Amazon Athena or comparable analytics tools. The feature functions with every storage tier, reportedly including Amazon S3 Glacier classes, without restoration or retrieval fees.

Use cases span media, finance, and life sciences. Media teams apply separate annotations for transcripts, moderation outcomes, subtitles and licensing details on video files instead of aligning multiple asset systems. Financial groups link AI-produced investment summaries and sentiment scores to documents so agents can surface relevant sets via natural-language searches. Life-sciences users tag trial records with compliance status, cohort information and approval histories, accelerating audits on archived data.

Annotations overcome limits of existing S3 metadata options. System-defined fields remain immutable and cover only intrinsic attributes such as size or class. User-defined metadata cannot exceed 2 KB and must be supplied at creation. Tags are restricted to 10 per object with tight character caps and primarily support access rules or lifecycle policies.

By contrast, annotations deliver mutable, large-scale context reaching 1 GB per object. Previously, comparable detail lived in separate databases or companion files, generating synchronization costs that sometimes surpassed storage itself. The S3 Tables MCP server supplies a uniform interface allowing AI models to query annotations through natural language.