Cloudflare TV

💻 What Launched Today - Monday, May 15

Presented by Ricky Robinett, Rita Kozlov, Sam Rhea, Celso Martinho
Originally aired on 

Welcome to Cloudflare Developer Week 2023!

Cloudflare Developer Week May 15-19, our week-long series of new product announcements and events dedicated to enhancing the developer experience to fuel productivity!

Tune in all week for more news, announcements, and thought-provoking discussions!

Read the blog post:

Visit the Developer Week Hub for every announcement and CFTV episode — check back all week for more!

Developer Week

Transcript (Beta)

All right, what's up, everybody? My name is Ricky and I lead up the Developer Relations and Community Team here at Cloudflare.

And I'm very stoked to be on Cloudflare TV with some amazing people to talk about all the stuff we've launched today as part of Developer Week.

So we've got so much to cover. I'm going to jump right in with one of my favorite announcements of the week, Constellation AI.

So Rita and Celso, you want to talk about what Constellation AI is? I can give a quick overview.

So if you've used basically any product recently, ever, you've probably noticed that things are getting a little bit smarter, right?

Like even our Google Chat, for example, will give us a little summary now of like, recently, Ricky said this and Sam also objected to it and Celso agreed with Ricky or whatever.

And so what we wanted to do is bring that machine learning aspect and that intelligence to the super cloud stack, right?

Since it's something that every developer needs to have access to now.

And so what Constellation is going to enable our developers to do, and I guess enables them to do starting today, and we've already started to let people into the data, is upload models to our edge and run inferencing directly on Cloudflare's network.

So as a request comes in, you can have the model kind of predict what the next output is going to be and use for a whole bunch of use cases.

Awesome. Celso, go ahead. What do you have to add? I think Rita explained it really well.

I think the workers ecosystem is now in a place where it's so rich and so powerful that we're actually using our own ecosystem to build new functionality.

And this is how Constellation was born. With the power of WebAssembly and using other APIs, like using R2 for storing the models and a few other things that the internal teams helped us with, we were able to run a machine learning inference engine on top of workers.

And now all these APIs are available to our developers and customers.

We've been learning a lot over the last weeks.

We try to build a platform that can evolve and can become much better in the future.

One of the things we try to do is follow the standards and the good patterns that the data science and machine learning communities appreciate.

Right now, we're supporting one of the most popular machine learning runtimes called ONNX, but we're already researching and testing other runtimes and the plan is to support as many frameworks and runtimes as possible in the future.

That's super rad. And can you tell us a little bit about the classification demo that we launched today?

I saw it in Rita's tweet about the announcement. What does that do? Yeah.

So, you know, every time we do an announcement of a product that is highly technological, we try to put out some demos to better explain what people can do with it.

And we've chosen a few demos today. One of them is an image classification app.

It's a pretty simple app, uses a very small machine learning model running on top of Constellation.

It was trained on top of one of the largest open images databases called ImageNet.

And basically, if you upload any PNG image to that classification app, it will give you the most probable tag, the most probable description of that image.

And it works pretty well. It's pretty impressive, actually, for such a small model to be able to predict what an image is about.

The other interesting demo that we announced is this email sentiment analysis worker.

And the interesting thing about it is that it orchestrates a few Cloudflare APIs.

And it shows that if you're clever, if you're creative, you can actually do things like receiving an email, doing sentiment analysis on the body of the email, and adding a special adder to the image with the score of the sentiment.

And you can actually use that to separate, for instance, angry emails from polite emails in your inbox if you want to.

So yeah, we are trying to open source all of these demos and codes.

Everything's available in the developer documentation, as well as in the blog.

So really looking forward to see what people can do with Constellation in the future.

Yeah, Rita.

And I saw you come off mute for a second. So I feel like you have something to add.

Salsa was talking about how we're running one of the largest trained models for image classification.

And for me, one of the things that Salsa touched on in the beginning that's been really cool about working on this is that it's all built on workers.

It's all built on Wasm. And I think people have a bit of an association still with, oh, it's edge compute.

It must be pretty lightweight. But it's actually pretty hefty.

It's pretty heavyweight. And we can support so many models on Constellation now.

But yeah, that's been really exciting. I think it's not only a great product in and of itself, but also a great demonstration of just the power of the platform.

Yes, I love that. And let me ask one last question, and then I'll put Sam in the hot seat.

So if I'm a developer, and I'm watching this, and I'm like, oh, my gosh, how do I get my hands on this?

Where does the developer go to sign up?

Oh, I know, I know. Um, so if you go to the Cloudflare dashboard, and of course, you're going to navigate to workers, because if you're a developer, that's probably kind of your autopilot that you're on.

And what you're going to notice is that in the sidebar now, there's a new little tab that says Constellation, and it has a little beta tag next to it.

And right in there, there's a button that says request access.

So you can request access in there. And we're really trying to stay on top of letting folks in.

So click away. Awesome. Rita, with the quick, detailed answer.

I like that. All right, Sam, why don't you tell us a little bit about the Zero Trust Security for your AI toolkit that we launched today?


So, sorry, I'll back our noise. So when I think probably everyone remembers the kind of day or days plural, when suddenly everyone was using chat GPT, right.

And, and a lot of folks obviously have been tinkering with it for months, and maybe even years prior.

But I think it was maybe November, when there was one day in a particular at work where I literally could not do anything else.

I canceled a couple of my meetings, because all I wanted to do in the entire world was was play with this tool.

And if anyone goes and kind of crawls my wiki posts inside of Cloudflare from that week, you'll think this is a lunatic who can only who is only tinkering around with this, this idea, this tool.

And that's wonderful and exciting.

And I think that has been pretty common for any number of folks in any number of companies.

And I think that's really exciting, because like what Celso and Rita have been describing today, there are really, really neat ways that folks can use these tools, not just open AI, but generally AI style tools and LLM style tools that go well beyond just, hey, I want a funny photo of, you know, a Kermit the Frog astronaut riding bicycle, and into really powerful solutions for how they do their work.

And yet, inevitably, with any new tool, there is this concern from security teams, and very valid concern of, hey, wait a second, we are now relying on what tool to understand and be part of some of our most sensitive workflows in our organization, whether that is writing unit tests for code that is unreleased future product features, or it is summarizing some notes from a customer conversation about a sensitive customer and with their names and PII involved.

And so a lot of these tools, the way in which you get value out of it, is you provide some input into it, right, whether that is training material, or in the case of some more conversational chat GPT AIs, input about what you're working on that you want output on, and sometimes that input should never leave the organization.

So we started to hear from customers that said, I need a way to safely allow my team to use this.

Because in the same way, if every person at every desk had kind of a consultant sitting right next to them, that wasn't under NDA, that wasn't under contract, but we were all talking to it all day about the solutions to our problems, that would freak out the legal and security teams of any organization.

And that's kind of akin to what's been happening here is, wait a second, you're telling me what CSV got uploaded so that we could figure out how to format some of the data to make it a little more friendly?

And so what we've been thinking about is how do we take Koppler's set of Zero Trust tools, which are our security solutions that give enterprises of really any size the ability to, for lack of a better term, safely use the Internet.

How do we bring that to this problem?

And it's pretty multi-layered. We talked to some organizations that say, you know what, we have a contract relationship with this service, this tool, and we want to let anybody upload anything to that service and tool, but we want to block all other tools.

That's easy. That's a simple DNS filtering or secure web gateway rule. Or maybe they're earlier in that conversation and they say, hey, we've heard that this is really popular, but we have no idea what tools our employees are using.

Are they using BARD? Are they using ChatGPT? Are they downloading something else?

So we also give CIOs and CISOs immediate visibility to say, hey, you've got 400 people who are regularly using ChatGPT, and the goal there isn't, hey, you should go have a conversation with them.

Maybe the goal can be instead, you should probably think about an enterprise license with some agreements around how your data is structured.

And then finally, we also have brought our data loss scanning tool into this workflow as well, where if customers want to allow employees to upload data to these tools, but they want to make sure it's not too much data, that can be scanned as it's uploaded in flight for things like social security numbers, or if you have very sensitive tags inside of your organization or Microsoft Information Protection Scheme labels, you can bring all of those to scans that happen as that data moves through Cloudflare's network to make sure that the only data being uploaded is approved and safe for that organization so that the organization can respect and honor its commitments to its own end, customers about data control and data sovereignty, while giving their team members the ability to take advantage of these tools that last November, I literally could not stop thinking about, and I'm still somewhat guilty of that.

Yeah, Sam, I feel like you had access to my dolly logs, because I definitely made the Kermit the Frog astronaut riding a bison.

I swear. It looks good in a spacesuit.

What can you say? I love this. I feel like just the data around how many folks within an organization are using ChatGPT on its own is super fascinating.

That's awesome.

And I'll say, I was talking to some developers last week, and we're all agreeing that before all this stuff, the fear was Skynet and AI, and that was what you talked about.

And now that it's here, it's like, oh, we're actually scared of PII leaking and proprietary information.

And then that's the stuff that people are actually afraid of right now.

So I love how you're addressing that.

How can people use all of this? How do they get started with it? That's the fun part.

And maybe my favorite part about my job is customers with 50 or fewer users can use effectively all of the Cloudflare Zero Trust set, or suite, I guess both words apply here, at no cost.

And that's not a freemium model. That's not like a three months, and then we charge you a model, no cost.

And that includes the understanding about shadow IT, what tools are being used, the ability to block certain tools and allow other tools.

The DLP scanning is part of our contract plan.

We're working on ways to bring that down into the PAYGO plan. But regardless of what size you are as an organization, you can sign up either for free or at a pretty reasonable cost.

We consider ourselves, because of the advantage of our network, the most cost efficient Zero Trust security solution.

You can go sign up at and get started in just a matter of minutes.

It takes probably 10 minutes to set up DNS filtering in an office network. It can take, depending on your MDM deployment, an hour or two to set up outbound filtering from devices wherever they are.

But if you go to dash.Cloudflare .com and look for the Zero Trust button at no cost, even if you just want to tinker with it under 50 users and then think about adopting it on a contract plan, you can get set up and exploring these tools today.

Awesome. And can I get you to tease any of what's next?

Yeah. I remember the first time I ever talked publicly about Cloudflare, I came from a place that was not as transparent about Cloudflare.

Not that we hid nefarious things, but we were just very cagey about what was on the roadmap and things like that.

And so when someone asked me that question five years ago when I was speaking publicly about Cloudflare, I was very, very tight-lipped about it.

And so now I'm really excited because my boss at the time told me, wait a second, no, no, no, that's not who we are.

You can be open and transparent about what's coming on the roadmap.

So I'm always happy to answer that question. One thing that we have available to customers is our API-driven CASB.

And what that does is it takes the APIs of your popular SAS tools, like Microsoft 365 or Salesforce or Dropbox, and it scans it for you, kind of overnight while you sleep or during the day, whenever.

And it says, hey, wait a second, you've got 57 public folders that say compensation in Dropbox.

Do you really mean that? Or hang on a second, this group of employees is sharing a bunch of information about these customers with what seems to be their personal email address.

Did you mean for that to happen? And so kind of the alerts, whether they are just mistakes in configuration or something maybe more malicious, we go through and make sure everything, every SAS app that you're using has these kinds of guard rails and guide you through implementing them if they're not already there.

And what we want to do is bring that same experience to these popular AI tools.

So that in addition to this kind of scanning that we're thinking about in transit, you can also take a look and make sure that, all right, these team members have access to these plans, or these team members have access to these plugins that use maybe sensitive internal proprietary information as part of kind of a closed beta.

A lot of these AI tools, as I know everyone here is observing, are going to evolve a lot.

I think they're going to only become more robust and more full -featured in the same way that the popular SAS tools I just mentioned, maybe Salesforce was kind of just a cloud-based CRM, and now can run an entire revenue operation.

And so as these tools evolve, the feature set and the functionality is going to get both more powerful and become more risky for any individual user to accidentally click the wrong button, right?

And so we want to make sure that we give our API-driven CASB experience to folks who are using those tools as well.

That's great. And Sam, you couldn't have teed it up for me more perfectly.

We didn't rehearse this, but speaking of AI tools evolving, I think I've been personally really excited to play with chat GPT plugins.

And we saw the announcement last Friday that they're opening that up to folks and very stoked that we have a plugin out there with Cloudflare Radar.

Celso, you want to talk a little bit about that plugin and how we built it?

Sure. So we've been obviously very curious about how we can use LLMs and how we can use tools like chat GPT.

And when chat GPT opened the plugins option, we thought about Radar instantly because Radar is one of those places where we have lots of insight, lots of data.

We have public APIs for all of that. And we're trying to tell the people what's happening on the Internet in the most human possible language.

So it seemed like the perfect fit for something like chat GPT. And so as soon as we got access to the developers APIs for chat GPT, we began working on a plugin and we built it entirely on top of workers, obviously using some of our uh components that we actually used for building Radar 2.0 a few months ago.

And in a few weeks we were able to launch the plugin.

It was approved last week. Anyone with access to plugins can go to chat GPT right now.

It should appear on the store and you can install it and you can start talking to the plugin in normal language.

It recognizes a bunch of stuff and we'll make sure to keep improving the plugin with more and more data.

Yes, it's super rad. Rita, go ahead. I was just gonna make a plug while we're talking about plugins, which is that if, and I'm not sure if you were gonna go there too, but if you want to build your own plugin, which everyone seems to be doing right now, we also published a tutorial on Friday as to how you can do that yourself on workers.

Yes, Rita. And as I was talking in a session on Cloudflare TV about an hour ago, I was using that tutorial.

Christian on our team wrote it and I was amazed that I had a plugin up and running in less than 30 minutes.

So it was really, really exciting. And that's not all. We also have a Cloudflare docs chat GPT plugin, which is a very robust example built on top of workers, KV as a vector store, which is really interesting.

Using queues, durable objects, Tron triggers.

So a lot of great stuff in this plugin. And what this does is it has an index of what's in our developer documentation and lets you get the most update updated information from our doc.

So we know chat GPT only knows things up to September of 2021.

So if you think of all the things we have launched since then we can now access those with this plugin.

So this is an example plugin that you can both use to see a robust plugin build, but also can run yourself and use to help you build Cloudflare applications.

So very, very stoked about this one.

And I want to give Michael Hart on our team a huge shout out for building this.

And also just, I remember I got plugins access and I dropped it in our chat.

And Michael was like, I already built a plugin. I don't have access. So I don't know if it works, which I think is just like a testament of like, you know, the engineers here at Cloudflare of like when they see something, they're like, I'm going to be ready that the exact moment we can use it, we can do it.

And Michael did phenomenal work here.

And that also ties into cursor, which is our AI assistant on the docs.

Rita, do you want to talk a little bit both about cursor, but also like the vision around interfaces for AI and building applications?

Yes. So I think like Sam on that day in November, whenever it was, I also went kind of crazy and was like, well, what is this thing?

It is so powerful. And it was really impressive, right?

Because you can kind of ask it anything and it will give you a very impressive answer.

And one of the first things that I was wondering was like, oh, can it build a worker for me?

And so I started asking things that I thought were pretty ridiculous, right?

Like, oh, can you tell me how to build a worker with Next.js?

And also, I don't know, let's say neon database. And it was like, here you go.

And, you know, the accuracy like varies to a certain degree, but the immediate thought I had was like, wow, we can generate so many different tutorials based off of this, right?

And I think that's kind of the direction that a lot of people are going.

But I think that that's still pretty limiting in that you have to think of all those use cases ahead of time.

And then you're stuck with the burden of, okay, now I have like 500 tutorials to maintain and update constantly.

And so I think the next kind of natural iteration of this is like, well, why would we pre-generate them if we can just let someone generate them on the fly or bring that assistant to the user?

And so, you know, if you're using BARD or chat GPT as your helper, then you have the plugins and that's really great.

If you are spelunking the docs, then maybe rather than using search, the assistant can meet you right there.

And I think that that's just the very beginning still of the interfaces that developers are going to see in terms of AI.

And we're seeing so much happening right now that I think AI is going to become a very primary interface for how people are building things.

We see tools like Copilot helping a lot, but we also started exploring, you know, what are other ways that developers are going to interact with us?

So we're already thinking about how do we integrate this into our dashboard?

How do we allow developers to start prompting chat GPT and generating code?

And there's a really, really cool interactive example in our blog post that Harley, who's a super talented engineer on our team built where you can be like, hey, you know, what's the best practice for building this type of worker?

And you can generate more than one example, right? Because sometimes even as you're writing code, you're not going to get it necessarily right on the first try.

Or maybe you want to see which algorithm is going to work better for a particular use case.

And you can start kind of start building out this like decision tree of your architecture, but instead of spending like 15 minutes on each piece of code, it's auto-generated, right?

So the world is kind of your oyster. So today we introduced cursor into our docs and we're excited to see where that goes.

And we're already getting a whole bunch of feedback.

I think it's going to be really, really helpful to many users, but I definitely see that as just the very beginning.

Yes. Thank you, Rita.

And if y'all get a chance, definitely read the blog posts on cursor.

Rita laid out a lot of this vision in that blog post, and I think did a very eloquent and phenomenal job of casting a vision there.

And with just a few minutes left, I will say this is my first developer week.

I'm so stoked to be here. The thing that gets me excited being here in the planning for this is the focus on how we help you all as developers build better.

And that's really what all of this is about.

Constellation, the chat GPT plugins, the Zero Trust. We want you to be able to build really anything you can envision on workers.

And so we had one other announcement today.

I want to make sure to plug, which is a bunch of free Cloudflare services.

We use a raft of free Cloudflare services in the blog post, which I think means a whole lot for AI startups.

So if you're a developer and you're building an AI startup, we want to make sure you have all of what we do in your hands to be able to pull it off.

And before, if you've looked at our startup program, you had to be part of an accelerator, have some requirements.

And for AI startups, we've pulled that away so that you can get in.

So if you've been wanting to get in and haven't been able to before, now you can.

Before we wrap, anyone, Sam, Rita, Salso, have anything else you want to add?

Just how excited we are in Zero Trust to be first customers of what Rita and Salso and the entire developer-oriented world is building.

Generally, Cloudflare Zero Trust is already workers customers.

But I think about what Rita was describing with tutorials.

We used to, every Monday or Tuesday, publish a new tutorial in the Zero Trust world, which was a lot of value, but also a tremendous amount of work.

And on one of those experiments back in November, I had to write a tutorial for Cloudflare Zero Trust.

And I think that might've been the magical moment where I thought, oh my gosh, I can describe a very complex environment and it can make really prescriptive recommendations about how to use Cloudflare products.

This is the future.

And so in Zero Trust, the kind of improvements won't just stop with some of these new features that are specific to the security products, but also we're really excited to kind of, like I think everyone watching be users and adopters and customers of what Rita and Salso and the whole career are building.

So stay tuned for ways that makes Zero Trust better beyond just the new features.

We're excited for you guys to try Constellation too.

I know you guys use a bunch of AI already to make smart Zero Trust decisions, but go request that access, Sam.

Yeah, exactly. Yeah, we'll put you in the waiting list, Sam.

Well, last in line. Well, let me close out and just say, come hang out with us on Discord, follow the blog, hang out on the Cloudflare devs with us on Twitter, like wherever you're at, please come be part of the conversation, not just this week, but all the time.

And we can't wait to see what you all build with this new stuff we're putting in your hands.

All right. Bye, everybody. Thank you, Ricky. Thanks.

Thumbnail image for video "Developer Week"

Developer Week
All the building blocks you need to create & deploy full-stack applications on Cloudflare. Tune in all week for exciting new product announcements and more!
Watch more episodes