Google Cloud on November 19, 2026 introduced object enrichments in preview, a Cloud Storage feature that runs AI models over files in a bucket and writes the results back to the objects themselves. The announcement was written by product managers Himanshu Kohli and Manjul Sahay. Its premise is that most enterprise information sits in object storage as images, PDFs, video and audio that an agent cannot reason over, and that reading raw files at query time is too slow and costly. Pre-extracting metadata, the post argues, gives agents grounded context at runtime.
A team defines which buckets to process and where, picks one or more of five first-party Cloud Storage models for images and PDFs, and grants IAM permissions to the service. Google says it then handles inference, response parsing, retries and throttling, and can scale to millions or billions of objects, with no queues or workers to manage. The image models cover object detection, face detection, image labeling and explicit content detection, and a PDF model sorts documents into business categories such as Finance, Games, News and Health. Setup is available in the Cloud Storage console or through the Cloud Storage API.
Results are saved as object context, a Cloud Storage metadata construct that holds key-value pairs next to the object. Google says this context follows an object through copy, move and replication, disappears when the object is deleted, and is governed by IAM, so no separate metadata store is needed. Applications can filter object listings by context, and the data can flow into BigQuery through Storage Insights datasets for SQL analysis. The post also points to the Cloud Storage MCP server as a way to give agents IAM-scoped access to objects and their context.
On price, Google cites published pricing of $700 to enrich 10 million images or PDFs and describes savings of 75 percent on image annotation against similar services from other cloud providers. These are Google's own comparisons, and the post does not name the competing services or show the calculation. Enrichment with Gemini for summarisation, entity extraction, transcription and translation is not yet generally open: customers are asked to email [email protected] to request access to what Google calls upcoming functionality.
The post offers several customer statements. L'Oreal says it can label its creative asset library 100 times faster at a fraction of the cost. Hyperscience reports a lift of more than 50 percent in document classification accuracy with two to three times lower inference cost, and Vivint says the combination with BigQuery cut AI development from six months to one. Google lists use cases including creative asset search for retrieval-augmented generation, content moderation, training data curation for autonomous driving and robotics, and multimodal analytics.
Those figures come from Google and its customers, not from independent testing, and the feature is in preview. The most advanced model option, Gemini, is not yet available without a request, so the offer today is mainly image labeling and PDF classification at scale.