AWS Bedrock Knowledge Bases add Marengo 3.0 for video and image search

Amazon made TwelveLabs Marengo Embed 3.0 generally available in Bedrock Knowledge Bases, enabling fully managed natural-language semantic search across video, image, and audio content. AWS published a walkthrough showing how to build a knowledge base and run semantic queries against media — extending retrieval-augmented generation beyond text into the multimodal content most enterprises actually store.

Mechanically, Marengo Embed 3.0 generates embeddings for media that let users query 'find the clip where the CEO discusses Q3 revenue' in natural language, without manual tagging. Managed inside Bedrock Knowledge Bases, it removes the infrastructure burden of building multimodal search pipelines, handling embedding generation, storage, and retrieval as a service.

Competitively, multimodal search is an underserved enterprise need — vast archives of video and audio remain effectively unsearchable. By integrating TwelveLabs' specialized embedding models into Bedrock's managed RAG stack, AWS extends its enterprise-AI moat into media-heavy verticals like broadcast, media & entertainment, security, and training. It complements NVIDIA's IBC broadcast-AI push the same week, signaling that media understanding is a hot 2026 category.

For developers, the value is removing the integration tax: rather than stitching together embedding models, vector stores, and retrieval logic, they get managed semantic media search through familiar Bedrock APIs. The caveat is lock-in — deeply integrating with Bedrock Knowledge Bases ties workloads to AWS. Watch adoption in media and enterprise-knowledge use cases and whether competing clouds match the managed multimodal-RAG capability.