Invest1 publisher3 min readPublished Updated
Twelve Labs' Marengo video search joins Amazon Bedrock's managed knowledge bases
Marengo is the first video model AWS lists inside Bedrock managed knowledge bases, a slot that hands Amazon the ingestion and indexing work and hands Twelve Labs distribution into every account that already keeps video in S3.
The Investor · Invest desk
What happened
- Corporate customers using Amazon Bedrock managed knowledge bases can now search their own video libraries for specific scenes in natural language, without standing up a separate video search system.
- Doing that before meant extracting frames, converting speech to text, linking output to a separate database and designing logic that analysed visual, audio and dialogue signals and combined them per query.
- Twelve Labs is demonstrating the Marengo-Iconik integration at IBC 2026 in Amsterdam, which runs from the 11th to the 14th, at booth 5.D82.
Compiled by The InvestorSomething wrong?How this is made
Why it matters
- decision A media or bank team that had budgeted engineers for frame extraction and a vector store now chooses between staffing that build and connecting a bucket, and the answer turns on how much of its own retrieval logic it wants to keep.
- constraint With ingestion, indexing and retrieval running inside an AWS service, the surface Twelve Labs sells into those accounts narrows toward the model call itself, and Amazon holds the relationship for everything wrapped around it.
- cost What the buyer stops paying for is engineering time on four build stages; the frames still get extracted and indexed, now inside AWS's service.
Four engineering steps leave the buyer's scope. Sedaily's account, citing industry sources on the 12th, lists them: extract frames from the video, convert speech to text, link the results to a separate database, and design a system that analyses visual, audio and dialogue information individually before combining them against a user's query [4][12][14]. Connect a repository such as Amazon S3 and AWS's service does the ingestion, the indexing and the retrieval [3][5].
"If the video is in Amazon S3, you can use Marengo's multimodal search right away without writing code or standing up vector search infrastructure," said Danny Nicolopoulos, head of strategic partnerships at Twelve Labs [6]. The report does not include pricing for Marengo on Bedrock, so each buyer is measuring the change against a build estimate it already had.
Two companies sit on either side of that connection. AWS's managed knowledge bases handle everything from data ingestion to indexing and retrieval [3]; Twelve Labs supplies the model that reads visuals, text, speech and sound together and returns the scene, the "show me the second-half penalty kick" case [13]. Plumbing the customer does not build is also plumbing Twelve Labs does not sell into that account, and what it buys in exchange is reach into repositories that are already in S3 [5].
Iconik, a cloud-based media asset management platform, is among the first adopters, and semantic search powered by Twelve Labs is now a standard feature of its platform rather than a paid tier [9][10]. The two companies will demonstrate the integration at IBC 2026 in Amsterdam, which runs from the 11th to the 14th, at booth 5.D82 [11]. Marengo's stated availability date, the 11th, is the show's opening day [15].
Marengo could stay the only video model in the service, in which case a first listing is worth real money because it is the default answer for a Bedrock customer with an archive. A second listing would flatten that, and the comparison would move to which model returns the right scene at what price on identical plumbing. There is also the version where the pipeline was never the binding cost: extraction and indexing are solved work, and the hard part is whether the retrieved clip is the one the analyst wanted, which a managed index does not fix. In my view the catalog fills up, and the durable question is recall on a customer's own footage.
One part of the pitch a buyer can verify itself: video and related data are processed inside the customer's own AWS account and never leave it [7]. Twelve Labs said that lets the technology be used in fields with strict data management requirements, naming finance, health care and the public sector [8].
What to watch
- Whether a second video model is listed in Bedrock managed knowledge bases, and how buyers then compare recall on identical plumbing.
- Whether Iconik's platform price changes now that Twelve Labs semantic search is bundled as a standard feature.
- Named finance, health care or public sector deployments citing in-account processing as the reason they proceeded.