Swapnil Bhartiya: AI is not just being used as a chatbot these days, it is writing code, it is taking action, the whole agenting workflow on people’s behalf and it is writing exploit faster than most teams can patch them. And then if you look at models like Mythos or glasswing, they are churning out vulnerabilities at a pace that teams cannot keep up. There are some actually open source projects, they actually totally rejected any contribution, the code that is written using AI. But the problem the teams face is that there are protections, but every login wall and every procurement cycle between a developer and secure image, it widens that gap even more. Minimus has finally decided to tear that wall down. It’s opening its entire catalog of container images to everyone for free, without any signup needed. And today we have with us once again John Morello, co founder and CTO at Minimus to break it all down for us. First of all, John, it’s great to have you back on the show.

swapnil: Yeah, thanks for having me. I appreciate the opportunity to talk with you.

Swapnil Bhartiya: It’s my pleasure. Of course we are going to talk about why you decided to kind of embrace this approach. Before that, let’s just talk about security, open source and AI. When we put all these three in the mix, what problem it creates for developers, for maintainer, for security teams, because it’s like a double edged sword. You can use it for both sides, but what challenge it’s creating and what led you to make this decision? So let’s understand the problem area first.

swapnil: Yeah, I mean, the problem is one that’s existed for a long time. It’s not a new thing that AI has created. It’s just that the application of AI for offensive research is increasing the volume and the frequency of vulnerabilities that are being discovered. But even before that, many organizations really struggle just to keep up with the pace of vulnerabilities that existed, you know, two years ago, five years ago, 10 years ago. I mean, it’s really, again, it’s not a new problem and it’s not all the fault of the end user of the software. I mean, in a lot of cases the problem really starts even further to the left, if you will, in the sense that the images that people pull to run their applications and infrastructure, even the latest official versions of many of those images, have a lot of vulnerabilities. So not to pick on Python, but just as an example, because it’s so popular, if you have someone, a developer and an enterprise organization, whatever, that wants to build an application on Python and they try to do the right thing and they pull the very latest version of the official Python image. They’re going to get something that has more than 400 vulnerabilities, 70 or so of which are high or critical. And it’s really difficult for any organization to make meaningful progress towards a secure image. In the end, if the thing that you start with has that many problems and then you combine that existing situation with now applications like Mythos and Glasswing and so forth, but just the general usage of AI to discover new vulnerabilities, if those vulnerabilities increase further, it’s that much harder for organizations just to keep pace with them. So what we’ve done in our product and what we’ve had on the market now for a bit more than a year is we basically run a pipeline that is continuously building tens of thousands of individual open source components directly from source every time there are new versions upstream and then assembling those into the container image artifacts that people need to build on and to run their infrastructure. So for example, if you think about like our Python image as an example, every component in that image, every day we are checking to see is there a new, more up to date version of it available. If so, we build it, we put into that image. And so the image itself is being maintained literally on a continuous basis. But also so we’re really thoughtful about what we put in that image in the first place. One of the problems and things that lead to all these vulnerabilities is oftentimes images are constructed just for convenience. You know, people might put a shell in there because they want to be able to log into it to do debugging. They put in some debugging tools, those have their own second and third order dependencies. And suddenly you have these images that have lots and lots of software in them. You know, all of which just is accruing vulnerabilities over time. So the other thing that we do is we make these images very minimalistic. We build everything on a distrust base and all these images only include the components that are really necessary to run the application. So the effective end result for you as a user is it’s the same python software or nginx or golang or any of the other 1500 plus images that we have. The same software that you’re already familiar with, it’s just built in a very minimalistic way, so it has less stuff to start with, and it’s being made continuously over time so that it always stays at that 0 CVE baseline and that’s a huge value for the user because now all that work of how do I know when there’s updates, how do I build them and test them and create images from them? We do that one time. You know, it’s hard to do that at scale, but we’ve built our product to be able to do that at scale. And the whole business model for us is there’s enough value in that for people that they’re willing to pay for that, for that no longer to be their problem. We can track, actually guarantee that we’ll keep these images up to date for our enterprise customers. They have dedicated support, they have integration capabilities and so forth. And basically what this ables to do is all they have to simply do is pull and use the images and they can immediately go from maybe having 400 images and 70 high and critical to literally having zero vulnerabilities to deal with without them having to do any work. And what we just announced is that whole library of images we’re making available for the larger community freely, without authentication, without signup, without any kind of gates. And that’s really important because so far the small number of images that have been made available freely require people to sign up for accounts and they were throttled or there were limits on how many you could have, or there were other things that made it harder for people to get them. We’ve just very much democratized, if you will, that whole scenario where you can now get these images, including even our FIPS images and images that are ready for FedRAMP scenarios, all freely available in our community Edition. And then Enterprise Edition continues to be the way that we sell support and SLAs and the integrations we have and the customization of images and self hosting capabilities, the kinds of things that we’ve already seen in enterprises value and are willing to pay for.

Swapnil Bhartiya: First of all, thank you not only for talking about this, also you know, making these images available. And there is so much to unpack here and I’m kind of struggling where to start because there are so many nice things to talk about. Let’s start with why you decided to just create a free tier. As you say, even no authentication is required versus you could have just created a freemium model and you know, you use this and then so you have some control over what was the driver behind this specific decision.

swapnil: It’s a good question and really a lot of the driver behind it is AI and more specifically developer usage of agents for building software. You know, if you have things behind some kind of signup process. It is harder for an agent to find out that those things exist, find out how to use them, actually pull them and make use of them. I mean, it’s not impossible, of course. Maybe you sign up for an account separately and you get some kind of token or password or whatever and you can provide that to the agent. But it’s just more speed bumps, more barriers to actually using them. The fact that these are available without any kind of authentication wall, without any kind of paywall, means that it’s extremely easy for agents to be able to find them and to use them, and thus to be able to really build software on these images much more rapidly than you might have been able to do before. Like, you know, one of the things that I had done personally after we shipped Community Edition as like a demo is, you know, I found an open source project that I was interested in creating a container image for something called Echo ip. It just shows, you know, shows you your IP address information and some other details about it. Just a really simple thing. But it was not in a container. It just existed as a golang, you know, repo that had a binary basically in it. And I never configured this environment to be able to, you know, to have to know anything about a credential or password or anything to use Minimus images. I just simply told Claude in this environment that I was testing in, you know, go and build me a containerized version of this application, but use Minimus images so it has zero CVEs. I literally just said something as high level as that. And because there’s no gating, because there’s no barriers to the images being discovered and pulled and utilized, Claude was able to actually take that Golang project, to take our static base layer, to be able to combine those two things together, to be able to create an image from that. And because of the way that we’re building those images, he can run that job, or Claude can run that job at any time and recreate the most up to date version of that image, even if it’s something that’s completely custom. And so one of the other things that’s really powerful, helpful here is not just building a new app, a very simple case that I just described there, but imagine like a real world scenario where maybe you’ve got a lot of different repos, you’ve got many different applications, many different images that compose them. Some are deployed with Kubernetes manifest, some might be you’ve got Docker compose for whatever it may be. You can literally now tell your agent go to these repositories, discovery what images are currently being used. Go to Minimus using the skills that we publish for agents already. Go to Minimus and figure out what are the replacement images for the things I already have and then rebuild those manifest or those compose files or the docker files to utilize the Minimus images and then immediately drop my vulnerabilities by 98, 99, 100%. And that’s really something that, you know, the fact that there is no gating here makes it much more possible for people to be able to do that. Because all the rich information about how those images work and how to use them and so forth, all that’s available directly without any kind of barriers.

Swapnil Bhartiya: Europe’s CRA is almost here. How does CRA impact Minimus? When you are making everything available, you know, without any gates. Does that kind of free you from CRA as well or you are still bound by their requirements, that hey, if you are distributing software, you’re still responsible for it.

swapnil: Well, you know, one of the things that’s really powerful with Enterprise Edition that many of our customers get a lot of benefit from is our integrations, what we call actions. And actions are really simple ways that you can trigger notifications and activities, you know, drive things like webhooks and GitHub Actions, or send emails or Slack messages. You can do that based on these really granular thresholds that you define about when you want to know about something or when you want to trigger action. So as an example relative to CRA, we already have Threat intel integrated into the platform and we obviously know the vulnerability state of all of our images at all times. And so we have a very straightforward way that you can just in the product say whenever that CRA threshold is met, something’s a critical vulnerability and it’s being actively exploited. When those things occur, I want you to send a notification to this team or send an email or run a GitHub action and rebuild the application and test it. You really have the ability now because of that integration of the threat intel that we have the ability for us to send signals to other systems outside of Minimus. You have the ability again to notify whatever teams might need to be aware of things when that CRA threshold is met, as well as to drive automation to take corrective action like pulling a new version of an image that includes the fix. So, you know, that’s one of the great things we think that the customers see value in with our Enterprise Edition product.

Swapnil Bhartiya: Excellent thank you. Now let’s talk about Mythos and Glasswing, they are finding vulnerabilities faster than ever. I am not in support of blocking these models, you know, because security by obscurity doesn’t make sense because they will find all those things. But talk a bit, how is it changing the game for developers and why can’t traditional remediation keep pace with them?

swapnil: Well, I mean, I think it’s a complex situation first because there’s this general idea that there’s all these vulnerabilities that are being discovered by AI, offensive research, and there is truth to that, but there’s also a lot of hype and a lot of noise and a lot of frankly garbage that’s being created with that too. You know, it’s hard to separate the hype with AI from reality sometimes right now. But you know, I look at the maintainer of Curl, very well known, trustworthy, well established person who’s been in the open source world for a long time that builds a piece of what I would consider, I think many would consider to be critical infrastructure. You know, Curl is present across all kinds of devices and services and so forth. And he wrote a piece a few weeks ago that I thought was really good about this topic, saying that while they get a lot more junk reports in, they also have things that are truly legitimate. And you know, Curl is a small compact project that’s been really focused on security from the beginning. It’s very well audited. So it shows that even things that have been built to be secure by default, there still are threats and vulnerabilities that offensive research may be able to find in them. And historically the gating factor for whether or not that offensive research occurred was, you know, was there enough human beings with the skill and talent, experience to do that, that had the motivation to also do that? Obviously with AI you take away a lot of those barriers there. And now anyone, be it a benevolent person who’s trying to find and fix problems, or be it somebody who’s malicious, who’s looking to find things that can be exploited, anybody has these abilities now to audit these code bases and find more vulnerabilities. And so I think the challenge for everybody is how do you respond when there are more vulnerabilities being discovered and Minimus helps with part of that. I certainly would not describe us as being the full solution there because part of it is how do you effectively triage all the stuff that’s coming in? If historically you might have gotten one or two security bug reports a month and now you’re getting 70, that’s something that’s going to be really hard for you to just figure out what’s legitimate and what’s not. And that’s a real situation that’s occurring for a lot of open source projects today. So that’s part of it. A second part of it is once you know that there are these issues that are out there, how can you fix them and distribute the fixes to them as rapidly as possible. One of the other reasons we want community edition not to have gates and barriers around it is it allows projects that might want to build on top of Python or just a static glibc image or that ship a model, a multi tiered app that has nginx and postgres and a caching service and so forth. It gives them an ability to say, look, we’re going to use these images that are zero CVE and we’re going to really reduce the amount of stuff we have to worry about. Not just because they don’t have any vulnerabilities to start with, but because they’re being maintained to continuously stay in that state without vulnerabilities. And that’s really, I think the only viable way most organizations can stay in front of this is to rely on somebody who does that at scale.

Swapnil Bhartiya: Once again, thank you. Now if I look at you folks, you’re not looking at AI agents or AI automation as kind of an afterthought. Most people are not looking at that. You have actually come out with something called minicli. If I’m not wrong, it is also kind of built for and is agent ready and built for AI agents. Is that correct?

swapnil: Yeah, minicli is just our command line utility for interacting with the product. You know, you’ve seen a lot of different iterations of how people wanted to expose data to AI. You know, there’s obviously just like consuming text. There have been MCP servers where there was more of like an API based way to do it. But I think what you’ve seen as a lot of these agents have become more mature and more capable is if you provide a consistent, well designed CLI interface, that’s really the best approach because it provides them an ability to not only discover the data they need, but also take action in a very structured way as well. And what minicli does is it basically first exposes that whole library of images we have and detailed information about them. So an agent can say like, do you have an nginx image? Yes, we do. Okay, what is the UID and GID mapping for that image? Well, it’s this, what volume mount points are present, you know, what port does it listen on, etc. All that information can easily be discovered through this structured command line interface. Secondly, once that has been discovered, the agent might say, the Minimus version of this image is a great fit for what we need, let’s build on it. So you can actually call minicli to create a Dockerfile that includes our image to start with, or even to help with the migration of an existing Dockerfile to that image. And the agent can rely on our command line to do a lot of that work so the results will be more consistent from it. And then for customers that are on Enterprise Edition, another thing that minicli can do is to evaluate the current images that you’re using, determine maybe there’s some components that you use in that image that are not typically required. Like again, maybe you have an nginx image, but in your environment that nginx image also needs to have curl in it because you use that for some kind of health check. Well, our default nginx image does not have curl because again, we build things to be minimalistic. minicli can then go interact with our custom image creator and say give me a new version of the Minimus nginx image that’s just for me that you’re going to build the same way you build your normal one, but also include curl in it. And so an agent can now through minicli discover what we already have, how does it work, to be able to request to build a net new application off of images in the gallery, and those are all things that even the Community Edition users can do freely. And then if you’re an Enterprise Edition customer, you can then further have that agent call minicli to be able to request and modify custom images that are built and maintained by us under our same SLAs, but for your specific needs and use cases.

Swapnil Bhartiya: Excellent, thank you. Now this is something I want to learn more about. How do you folks keep up with it? If you look at your images, there are near zero CVEs and only aligned with a lot of frameworks like NIST. How do you folks keep up with it?

swapnil: Well, I mean, we really focus heavily, as you might imagine, on durable automation for all these images. We are not manually updating anything. We in fact in many cases do not even manually build the initial image for the recipe entirely by hand. We have built a whole agent framework and skills and prompts that make it easy for us to be able to say build an image that includes this software and then to have Claude basically go and construct that recipe for our pipeline as completely as it can and then to have one of our engineers go and validate that and do any kind of additional polishing that might be required there. But from that point, once that recipe has been built, it is going to be maintained automatically by that pipeline over time. And so for us there is just not feasible. There is no way we could be a viable product if we did not have that extreme view of automation and really all the software that we are building.

Swapnil Bhartiya: Since you have made the whole catalog free, are there any restrictions for corporate users or does it really not matter whether you are an open source developer or an open source project?

swapnil: No, it’s freely available to everyone. It’s not limited, you know, you cannot use it for commercial purposes or anything like that. Anybody can use the images. Our whole bet, our business model with this is, you know, a common situation you have seen with many other successful open source companies which is having the software available freely is great, but there is a lot of value that people are willing to pay for, for support for contractually backed SLAs, for how quickly we remediate vulnerabilities for things like the custom image creator so that you can give us your own recipes and we can build and maintain those things for you, for the ability to self host all of it, for the integrations like as we talked about for CRA. But many other scenarios where you want to drive automation when images are published or vulnerabilities are fixed, those are all things that only exist in the enterprise version of the product. And from what we have heard from our existing customers and ones that we are talking to as prospects now, people understand the value of that and it is indeed something that they are willing to pay for. So certainly I am sure there will be people that will happily utilize the free images and never consider paying us. And that is completely okay. I mean that is part of the deal when you do stuff like this in open source, we understand that this is not a trial, it is not something that we are going to end in a year. We went into this with our eyes very wide open to that. And we believe that still the overall good, the selfish good for us as a company is still positive here. It is still something that is going to be beneficial to us in addition to being beneficial to all the users of these images.

Swapnil Bhartiya: How should teams look at this now? Open source, as important as it is to make things freely available, it does not mean free as in beer, but how do you help organizations understand that they can use it, but they can also move to enterprise? What benefit will they get by moving to enterprise?

swapnil: Well, by moving to enterprise there are a number of benefits that they get. First is contractually backed SLAs for vulnerability remediation. We have the best SLA in the industry for how rapidly we fix high and critical vulnerabilities. So you do not get a contractual guarantee of course unless we have a business agreement in place. And so that is one thing. Secondly, you get 24/7 support from the same team of people that created Twistlock and really started the whole container security space, something we know very intimately. So if you have issues, we are contractually obligated to resolve those issues. Third, just from a product feature standpoint I think is the image creator capability. It is one of the more popular parts of the product where you might have a need to add a set of environment variables or upload your own internal PKI certificates or add some specific packages to create an image that is still very minimalistic and still continuously maintained, but reflects a little bit more closely the exact things that you as a customer might need. And that is one of the things that only exists in Enterprise Edition. The integrations, what we call Actions, enable you to integrate with things like Jira and GitHub and Slack and webhooks so that if a vulnerability is found and it is critical and it is likely to be exploited in this nginx image, then you want to call this webhook and drive some redeployment with the new version of the image or you want to create a Jira ticket, assign it to that team or send a Slack message. Those are all things that are uniquely available in Enterprise Edition. We also have the self hosting capability, including in fully air gapped environments. You can take all these images and you can put them into whatever registries you may already be using. That is another piece with Enterprise Edition. And then probably the final big piece is minicli, as we have described, does a lot of things for you. One of the things that we provide only in Enterprise Edition though goes beyond just that binary. We actually provide a set of configurations and prompts that you can literally copy and paste out of our product UI and put into your developer desktops, into the configuration for Claude or Codex or Cursor, that tells those agents automatically to get images from Minimus, how to do migrations, where to find minicli. Basically you could take a desktop that a developer had never touched before, you could put all of our configuration into that desktop using whatever management tool you prefer. It is just a standard config file, JSON, they all vary a little bit, but it is basically just a text file that you are going to put on that endpoint. And then your developer, without even knowing about Minimus, could just simply say build me an application that does this with Python. And that agent will automatically know to prefer the Minimus images, where to find them, how to pull them, how to get information, how to build on them. And that is a really important thing because if you think about your typical Fortune 1000 organization, they may have thousands of developers, they are not all necessarily going to know about the nuances of container image vulnerabilities or Minimus or where to find these things. It is really powerful if you can make it easy for your customers to configure their entire environment to find and prefer and use these images, it just makes adoption much more simple.

Swapnil Bhartiya: Excellent, John, thank you so much for not only of course making it available for free without any authentication, but also thanks for talking with me and as John mentioned, folks who are watching, please go and check out Minimus to see what they are doing there. Once again, John, thank you and I look forward to chatting with you again.