NATS Graduates as a Messaging Backbone for AI Agents
NATS has graduated from the CNCF after roughly 15 years of development. The messaging project quietly underpins thousands of distributed systems. Many developers still have never heard of it.
Synadia’s Andrew Connolly and Waldemar Quevedo examine that milestone with Alan Shimel at KubeCon + CloudNativeCon in Salt Lake City. Connolly is head of marketing, and Quevedo is a core NATS maintainer.
A Swiss Army Knife for Messaging
Connolly describes the project as what you would get if Redis and Kafka had a baby. It combines messaging, streaming, persistence and a key-value store. All of it ships in a single, lightweight binary with no external dependencies.
It runs the same on a Raspberry Pi, a laptop or a large cloud server. Quevedo first used it in Tokyo as an early Cloud Foundry adopter. He went on to write the first book about it.
What Graduation Signals
For the team, graduation validates years of work by many contributors. More than 2,000 companies use the project, from Nvidia and Lovable to hobbyists.
Quevedo says it supplies both the dial tone and the voicemail for distributed systems. Low-latency messaging handles live traffic, while persistence keeps data available when components go offline.
Built for the Agentic Era
Simple building blocks have held up well as AI workloads arrive. An agent is still a workload that needs to talk to other components. Coding assistants also write NATS code well because they were trained on so much of it.
AI has brought new pressure too. Maintainers have handled a wave of AI-discovered vulnerabilities since early this year. They are also reviewing a growing volume of AI-generated contributions.
To keep up, the team invests in advanced testing platforms such as Antithesis. The goal is proving new code is correct, not just accepting more of it.
Synadia offers enterprise support and day-two tooling for security, monitoring and observability. Open source users can start at nats.io, while commercial details live at synadia.com.
Transcript
Hey everyone. We're back here at KubeCon in Salt Lake City. Just like yesterday, man, the floor opens up and you can start feeling the pulse, right- Absolutely ...
of the show this year, and it's really dynamic, really exciting. Good stuff. My next two guests are from a project called NATS.
If you've not heard of NATS or you're not familiar, don't worry, we're going to get you there. But first, let me introduce you to Wally here on my immediate left, and Andrew from Philadelphia on my far left. Andrew, I'm going to ask you to go first, introduce yourself- Yeah ...
and give people a kind of sense of kind of who you are. Thank you, Alan. Excited to be here.
Yeah, my name is Andrew Connolly. Like you said, I'm based in Philadelphia. I'm actually a marketer, but a technologist as well.
And so I have worked at a bunch of startups over the years, and today I work at Synadia, who is kind of the primary maintainer of NATS. Excellent. Mm-hmm.
Marketer and technologist are not mutually exclusive. Some of the best marketing people I know- Agreed ... actually have engineering degrees.
That's right. So, and I always like to remind people of that. Wally, how about you, man?
Yeah. So, I'm Wally. I'm one of the core maintainers from NATS.
I've been using NATS based systems for over a decade. I first started using NATS when I was still in Japan in Tokyo, working for Rakuten. One of the very early adopters of Cloud Foundry.
NATS was a major component of that architecture, so that's when I started using NATS. Eventually became maintainer of their Go implementation, so I've just seen the NATS project grow from a very niche project to this very mature, and now graduated project from the CNCF. Wally also wrote the first book on NATS.
I wrote the first book on NATS- Ah, okay ... as well. That's the marketer there for you.
You owe him. Thank you. And Andrew.
Yeah. So let's, Wally, first of all, what's your last name? Quevedo.
I love the way you say that, Wally. I can never say that, but good. Yeah.
And then where are you working now? I'm based in Berkeley. If you're allowed to say.
Oh, yeah. I'm in the Bay Area. Yeah, all the NATS team is remote.
Uh-huh. Yeah, so we have open source projects. It's very dynamic, and a lot of the maintainers are in Europe, with also some presence in the Bay and US.
Very cool. " Mm-hmm. Well, how would you describe what NATS does?
Sure. NATS is like if Redis and Kafka had a baby. Okay.
It's a Swiss army knife of messaging, streaming and persistence, key value store, and it does all these different messaging patterns or data patterns in a single 20 meg binary with no external dependencies. So super lightweight, runs anywhere, runs the same on a Raspberry Pi or on a laptop as it does on a big beefy server up in the cloud. And yeah, we sometimes say it's the internet's best kept secret.
It's really powerful. Well, look, I know when a project graduates, it has to have a certain amount of downloads, a certain amount of maintainers, a certain amount of code contribution. Mm-hmm.
So I don't know if it's truly a secret secret- ... but this is the beauty of the CNCF, right? You got, I forget what it is, maybe 240 or 270- Yep.
Projects, yeah ... projects in CNCF right now. Not all of them are Kubernetes in hotel.
Mm-hmm. There are a lot of projects that are the guts of how we do software today- Mm-hmm ... of how we do applications.
That if they weren't there, everything would come crashing to a screeching halt. Yep. Right?
And NATS is in that. I had a friend of mine who was a VC. He used to say he loved investing in those kinds of products.
Sure. The ones that no one really knows, but the whole world depends on them kind of thing, right? Mm-hmm.
So that's out there with NATS. Now, Wally, you mentioned you guys graduated. Mm-hmm.
Right? Again, not everyone in our audience understands what it means when a project graduates. It's not like we have a ceremony with a cap and they move the tassel over.
I wish. It would be nice. But graduating from the CNCF ecosystem is a big deal.
As I said earlier, you got to have a certain amount of community- Mm-hmm ... downloads, maintainers, contributors of code, etc. It was announced yesterday, was that the story?
Yeah. Yep. Mm-hmm.
So what does that mean to you now? Yeah. I would say it's sort of a validation of a lot of hard work and investment by many people over the last 15 years.
And sort of confirms what we kind of already felt and knew. NATS is battle tested, more than production ready. I think in the press release yesterday we talked about NATS is used by over 2,000 end user companies around the globe.
Some really cool logos, Nvidia, Lovable. But it's used by small freelancers and hobbyists, too, right? It's for everybody.
Mm-hmm. I'm sure Wally can attest over the years, all the different types of organizations that you've interacted with that use NATS. It's an incredible spectrum.
Yeah, like Andrew mentioned, you can see companies like Lovable, they just have a really good post, like writing about the use, like the utmost ones feature set from NATS. But you also have companies like Nvidia, where NATS and the persistence module there is a significant component of the cloud functions architecture, right? So, we like to say that NATS allows you to provide both the dial tone and the voicemail capability within your distributed system.
Yes. So yeah, it has been very cool. Yeah.
It is, yeah. Very cool. Mm-hmm.
All right I want to talk a little bit about AI, because how can you not? Yeah. A theme I've been hitting on all week is that the cloud native stack is the AI stack.
Mm-hmm. Right? Had we not had this whole Kubernetes stack, including NATS, available to build our AI on top of, AI would be a very...
I don't know if it would be anything what it is today, right, in terms of how far we've come with it. But I think it's dishonest to say that AI just fits right on top of what we had before, like a hand in a glove. There are changes.
There are things we've had to do. We've got to adapt to scale. Mm-hmm.
Security and governance. Mm-hmm. Right?
Agents, non-deterministic AI in a deterministic world. Yep. So every project, every piece of this stack is being pushed and pulled and stretched- Mm-hmm ...
by how we're using AI. How's that happened to NATS? Well, the key thing about NATS is that we always try to keep things simple, right?
So NATS at its core is a very simple way of low latency communication among components within architecture. So those building blocks you can still use to build AI distributed systems as well. So yeah, same like an AI, or it's a workload, it's still a workload, it needs to communicate with other components.
So those requests upon the same semantics that you can do with NATS, they can just leverage all of that. Which is really cool, because the history of NATS, all the LLMs, the AI coding agents, they can craft NATS code super well. They're all trained on that.
So yeah, it just fits really well within the agentic paradigm. You mentioned some things changing and some things not. Yes.
What have you got to change? Well, one change, and then it's really fallen more on Wally and the server maintainers team, AI discovering vulnerabilities in open source software. That's been a big focus for the maintainers this year, is how to respond to that effectively.
There's a whole post-Mitos glass swing, right? Yeah, exactly. You're getting inundated with new vulnerability CVEs and so forth.
That has been a major source of work for the NATS team since around March. We had the first release where we had addressing around 20 CVEs, and we have been reducing them, and we fared well during the whole Black October incident. And yeah, but that is a big factor now for an open source project.
They say that now open source, how it's going to evolve, it's going to be more like a proof of work kind of investment, making sure that you invest so much that the project is secure. So yeah. That's one of the things that we really care about, of this- Sure ...
AI. Yep. So when we talk about AI and these open source projects, I think the increased vulnerabilities is one- Yep ...
kind of stress level on the maintainers especially. I think another one is the amount of code being contributed, and it's probably AI, not probably, it is. Yeah.
It's AI generated code. Yeah. Doesn't make it bad.
Mm. But now when you had this many code contributions and now you got that many code contributions, again, it's a scale issue. How do you keep up with it?
Yeah. Have you seen that? And yes, definitely, and a big factor on that is how to prove that the work is correct.
So I think that's where things are heading, by construction, making sure that... So you have all this code, but how much can I trust this code? So that's why we invest a lot in platforms like Antithesis and these other projects that are more advanced testing.
So to make sure that, okay, so we have these new codes, but we are also covering all of these new edge cases. There are no new edge cases. So that is where we invest.
On making more secure, but also more robust, the open source project. So they can trust the community to keep it down. Agreed.
I agree with you there. So I imagine when it first started happening, it was sort of a rush, right? " Yeah.
I think we've gotten our arms around it a little bit, and the team has a new operating rhythm for how they handle that stuff. Which is exciting because then we can focus more on the end users and the use cases- Right ... who are doing the cool stuff with NATS and AI.
And I think we've been fortunate that NATS was, in a lot of ways, really well positioned for this AI wave. A lot of NATS original ideas, like the simple building blocks that Wally mentioned, have held up really well. So NATS is and was great for microservices, and a lot of those same characteristics carry over to agentic workloads.
And so it makes us a great fit. Last question for you, Andrew. You mentioned the company you work for is kind of the principal maintainer- Yeah ...
or very involved. Yeah. Name of the company?
Sinedia. And I imagine Sinedia offers some commercial... Yeah, absolutely.
Sinedia offers enterprise support and services. So if you're a big enterprise and you need a vendor to give you a little more, to be your backstop when you're running open source, we do that. And we're also offering a lot of day two operations kind of software for making security monitoring observability dead simple.
We find that a lot of big companies, one or two folks were the NATS champion, they bring it in, but then the broader team isn't a NATS expert. Yep. And they want to buy software that kind of gives them that NATS expertise.
And that's where we step in. Excellent. The website?
com for the commercial. Excellent. Mm-hmm.
Guys, thank you so much. Yep. Thank you so much for having us.
Wally, you're a pleasure. I appreciate what you do. Thank you.
Thank you. Maintainers are underpaid and overworked for the most part, so thank you. Yep.
No, we appreciate it. Hey, we're going to take a break. We're here at KubeCon.
We've got more. You're watching Techstrong TV.