Docker AI Catalog : The Future of Curated AI Models and LLMs

TL ; DR

• AI is evolving so fast it’s hard for anyone to keep up, let alone enterprises.
• Companies don’t know which AI tools are secure or fit their needs.
• There’s a big trust issue when it comes to picking AI tools for enterprise use.
• Curated AI catalogs simplify things by offering pre-vetted, reliable options.
• Docker’s AI Catalog solves this by providing a trusted library of AI models and tools.
• It gives businesses a consistent way to deploy AI without guesswork.
• Verified tags and usage stats help build confidence in what’s safe to use.
• The catalog isn’t just for models—it includes frameworks, databases, and orchestration tools.
• It’s easily accessible through Docker Hub and Docker Desktop.
• By making AI easier to trust and use, Docker’s catalog helps companies innovate faster.

Context

AI, and specifically GenAI is progressing at a tremendous pace; it's almost impossible to keep track of it, even for someone who is full-time dedicated to that ecosystem.? The AI newsletters are brimming with details about the next new LLMs, SLMs, AI orchestrators, AI gateways, and so on.

This innovation in AI is extremely promising, but presents a new challenge to most enterprises :

how to process all this information, and decide what is the correct course of action for them.

Enterprises cannot afford to experiment with every single LLM, or library that is out there - they need someone to guide them as to what is trusted, secured and allowed to be used, within the context of an enterprise ecosystem.

The AI Corundum : To Much to Choose From

As part of the Internal Platform Adoption team at EY, when I speak to my enterprise customers, most of the questions around GenAI revolve around the topic of :

"Is this LLM allowed to be used by my company ?"

"Can I download LLMs from HuggingFace, is that allowed ?"

"How can I be sure this LLM does not have any vulnerabilities, can I use my client's data with it ?"

Obviously, every single solution team will need to do their due diligence on what is secure, or allowed within their usage context, but there is a need within enterprises for some generic guidance to help them choose between what is allowed, and what is blocked, within the confines of that enterprise itself.

Basically, they need someone to filter this ever - increasing catalog of AI models, LLMs, SLMs, AI Orchestrators, AI Gateways... and tell them which :

• Are secure, or has non-critical vulnerabilities
• Are supported by the respective vendors
• Provide a standardized way of deploying

The need for Curation within the AI Ecosystem

The AI Ecosystem is too fractured now, for any single person to know what to do. Let alone individuals, entire portfolios, or teams are having a difficult time in identifying what to use, and what not to.

In the vast and growing landscape of AI, not all AI models are created equal. Businesses often struggle to identify which AI models meet their needs, ensure compatibility with existing systems, and maintain ethical and reliable performance.

Curated AI models solve this problem by offering pre-vetted, high-quality solutions that:

• Enhance reliability: Users can trust that the models meet industry standards and have been tested for real-world applications.
• Save time: Developers avoid the lengthy process of searching for, validating, and configuring models.
• Ensure ethical use: Curated models often come with documentation detailing data sources and intended use cases, reducing the risk of bias or misuse.

From my perspective, this is where AI Catalogs have a huge role to play.

They offer a single pane of glass within an enterprise context, for consumers to choose from the curated subset of models / LLMs that are supposed to be more secure, and compliant than others. This provides the necessary confidence that engagement teams need to get started on a piece of work.

Docker AI Catalog

Docker is a trusted name in the industry, and for long the Docker Hub has been a gold-standard for teams to find docker images, helm charts, based on the use-case. With the addition of the GenAI catalog, teams can also search within a curated list of offerings that they can then use confidently for the entire GenAI application development lifecycle.

Docker AI Catalog is a curated repository of AI and machine learning (ML) models, and other associated artifacts, designed to simplify the process of integrating advanced AI capabilities into applications.

Let's take an example :

Say, an enterprise user wants to start using Llama, by deploying it on their company's private cluster setup, but don't know where to get started, what is the latest secure LLM image they can use, and so on.

They come to Docker Hub, and can now search for Llama

Once they find what they are looking for, they can get more details :

As you can see there is a lot to unpack here :

• Verified Publisher tag : this provides the necessary confidence that this image was pushed by Meta, and not by someone else posing as Meta. In an enterprise setting, this is critical to consider, as while going for InfoSec reviews before production deployment, all of these aspects will be critically analyzed.
• We also see that this artifact was last updated 2 months ago. This also provides a good deal of confidence that you are no consuming a legacy version of the LLM, which is not being updated, might have vulnerabilities, and so on.
• You can also see how many users have already downloaded it, and how many have starred the artifact - again increasing the confidence of the enterprise user to consume it.
• Finally, you have a set of instructions to start working with the LLM, by downloading and deploying the image. Another interesting point here is that no matter the publisher of the LLM, the consumption patter is fixed - this means as a developer you can rely on how the image will be downloaded, and deployed; this consistency in consumption approach helps provide stability to the enterprise, rather than having to memorize different techniques, etc.

Docker AI Catalog : Different artifact types

Docker has been in this business of managing, verifying and offering containerized workloads for a long time now, and Docker Hub to this day is still the de-facto image repository for most of us.

So, with the GenAI ecosystem in play now, it's only normal for Docker to put their skin in the game, and leverage all the existing investments they have in this space, to provide a similar sort of experience for GenAI assets.

But, it's not only images of LLM models that they are providing a standard consumption pattern for, you can also search and consume other artifacts involved in building a GenAI stack like :

• Model Deployment & Serving like ollama, localai, cuda
• Orchestration offerings kubeflow, langchain, mlflow
• ML Frameworks like pytorch, tensorflow
• Databases like Mongo Atlas, Neo4J, Redis, Milvus

So, essentially it's a one-stop catalog comprised of a curated, and verified collection of components that one might need to create a GenAI stack starting from the models, to the orchestration layer, to the eventual deployment engine.

Where can I find Docker GenAI Catalog

You can access the Docker GenAI Catalog right from the homepage of Docker Hub ( https://hub.docker.com/ ) :

Recently, it has also been integrated within the UI of Docker Desktop, so you don't need to navigate outside of Docker Desktop, and go to Docker Hub to find the GenAI related artifacts - you can find them within the Desktop UI itself.

Conclusion

Docker, Inc has been around for a long time now, so has Docker Hub, and other associated offerings from them.

Till now, they were providing an easy-to-comprehend way of searching, and consuming container artifacts like docker images, helm charts, etc. Now, Docker has taken it up a notch, and extended their framework to support GenAI assets ( LLM models, orchestrators, etc. ) as well, while following the same consumption pattern.

By offering curated AI models and LLMs in a user-friendly, secure, and scalable format, it paves the way for businesses to harness the full potential of artificial intelligence. As the industry continues to embrace AI, platforms like Docker AI Catalog will undoubtedly become integral to the next wave of innovation.

Anybody who has spent time in an enterprise setting knows how difficult it is to agree on something. Most teams just want to consume the latest and greatest, but don't want to invest any effort in researching different options, etc. For such use-cases, having a trusted repository like Docker AI Catalog makes a lot of difference.

If the enterprise already has a enterprise agreement with Docker, Inc , it becomes even less of a discussion. In those cases, the security team has already vetted docker and approved it's usage within the firm, in which case using the catalog just becomes an extension of the same process, and becomes even easier.

In any case, when it comes to AI - there is a gap in the industry today, that is not a technology gap, rather a trust gap.

With so many companies coming up with the next big orchestrator technology, or vector database, or LLM(s), SLM(s0, etc. nobody really knows who they can trust with their client data, and who not. Any enterprise-grade company who tries to solve this issue, by providing an authenticity to the consumption process will always have my support.

References

[1] https://hub.docker.com/

[2] https://www.docker.com/blog/docker-desktop-4-37/

[3] https://www.docker.com/resources/2024-12-09-docker-ai-catalog/

[4] https://www.docker.com/blog/accelerating-ai-development-with-the-docker-ai-catalog/