Full Transcript

·YouTLDR

Artifacts: Versioned storage that speaks Git

21:054,941 words · ~25 min readEnglishTranscribed Apr 27, 2026
AI Summary

Cloudflare Artifacts is a new versioned storage system designed for AI agents that provides an on-demand, scalable Git server running inside Durable Objects. It allows for the creation of millions of isolated, disposable repositories that agents can interact with using standard Git protocols without the overhead of human-centric platforms like GitHub.

As AI agent activity causes a 14x increase in Git traffic, developers need a way to provide agents with isolated, high-performance environments that prevent merge conflicts and scale programmatically beyond human-centric infrastructure.

Section summaries

0:00-1:00

Introductions

skip

Standard greetings and introductions of the speakers.

1:00-5:00

The 'Why' for AI Agents

watch

Explains the scalability issues with GitHub/standard Git and why agents need a new approach.

5:00-9:00

Code Examples & SDKs

watch

Crucial details on how to programmatically interact with Artifacts via SDKs or Git CLI.

9:00-14:00

Use Cases (Config & Workflows)

optional

Explores using the system for non-code assets like configuration files.

16:00-19:00

Technical Deep Dive

watch

Explains the implementation using Zig, WASM, and Durable Objects.

Key points

  • Git for Agents vs. Humans — Existing platforms like GitHub are built for social coding and human collaboration (pull requests, UI), whereas agents need high-volume, isolated, and programmatically disposable repositories.
  • Durable Objects Architecture — Artifacts are built by running a Zig-based Git server compiled to WASM inside Cloudflare Durable Objects, utilizing their embedded SQLite state and global distribution.
  • Beyond Code: Versioned File Systems — While it speaks Git, Artifacts can be used as a versioned storage backend for user configurations, notebooks, or web assets, allowing for easy rollbacks and diffs without exposing Git to the end user.
Everything we built today for version control hasn't really scaled for agents, it was all built for humans. Matt Silverlock
The agent doesn't have to know anything other than git, which is huge because agents really do know git. It's in the training sets. Matt Silverlock

AI-generated from the transcript. May contain errors.

0:09

Hello everybody and welcome to this

0:12

today in agents week. Uh I am here to

0:16

talk about something that's very

0:17

exciting that has been uh redacted in

0:19

the past. We've been doing a lot of

0:20

stuff about this, but uh with me before

0:22

we even get started, I want to talk to

0:24

one of the blog authors of what we are

0:26

looking at today, a thing called

0:28

artifacts. But but uh first a person who

0:31

doesn't need any introduction at all,

0:32

but I'm going to make him do it anyway.

0:33

Matt Silverlock, can you introduce

0:35

yourself?

0:35

>> Of course. Uh thanks Greg. Um super

0:37

excited obviously to talk through what's

0:39

shipping today um and artifacts. Um I'm

0:41

Matt Silverlock. I lead product for

0:43

storage and databases and a bunch of

0:44

other stuff at Cloudflare. Um been here

0:46

many years. Yeah, super excited to kind

0:47

of talk about and whether you've read

0:49

the blog or not to talk about uh

0:51

artifacts and what it actually means and

0:52

what it is.

0:54

>> Yeah, totally. We've been we've been

0:55

pushing this for a while. We've been

0:57

saying redacted for a while. I kind of

0:58

like that the social media buzz that's

1:00

happening. It's nice to finally be able

1:01

to talk about this without it being all

1:03

all blocked out. Uh um I was thinking so

1:07

we have the blog post. Great blog post.

1:08

Read the blog post. Uh you're not the

1:10

only author. Let's give a little shout

1:11

out at the start. Who who are the other

1:12

authors on this?

1:13

>> Um it takes a village. So the authors

1:15

are even only a slight reflection of

1:16

everyone that's put the work in. But I

1:18

spent a lot of time with Matt Kerry and

1:20

uh Dylan on our team really working on

1:22

the blog and working a lot of what

1:23

Artifacts is today, but there's been a

1:25

lot of people behind the scenes. But um

1:27

yeah, huge shout out to Matt Matt and

1:28

Dylan who's been kind of grinding away

1:30

helping us get this over the line.

1:32

>> Awesome. And I wanted to share, if this

1:34

is okay, I'm going to share the new

1:35

product page that uh just launched

1:37

today. So this is super awesome and I

1:41

was thinking that we could kind of use

1:42

this as walk through this a little bit.

1:43

Man, I feel like I I am I'm just now

1:45

breathing in. It's been redacted to me.

1:47

So, I'm breathing this in. So, I've got

1:49

some questions about what's going on

1:50

here. So, let's just go from the top.

1:52

Why are why are we building this? What

1:54

What is this gorgeous page talking

1:56

about?

1:56

>> Um the page is one thing. So, you know,

1:58

I think it's probably no surprise to

1:59

anyone as we have all of our coding

2:02

agents, code review agents, sandboxes,

2:05

right? Um all of our harnesses, right?

2:08

um a lot of them rely on git repos for

2:12

you know actually managing and and

2:14

obviously committing the code and

2:15

sharing it with others right um even

2:17

just managing state before that sandbox

2:19

shuts down um and I think the challenge

2:22

is is like everything we built today for

2:24

version control hasn't really scaled for

2:26

agents it was all built for humans right

2:28

every kind of

2:29

>> social code network at GitHub or anybody

2:31

else right um

2:33

>> kind of seeing this like unprecedented

2:35

scale I think the uh the CO of GitHub um

2:38

posted about a week ago actually that

2:40

they said a 14 times year-over-year

2:44

increase in uh like traffic volume um

2:47

like git operations I think it was

2:49

>> on the network right and that's not like

2:51

>> 14 from a base of like a small startup

2:53

that's been around a week where like you

2:55

just scale up a few more VMs right this

2:56

is like 14x from a company already

2:59

operating at internet scale so

3:01

>> we kind of wondered

3:02

>> yeah I think there's something we can go

3:04

and solve here if we do it differently

3:06

for agents it might actually work

3:08

um what what is why is it a hard thing

3:12

for agents with git like using git just

3:14

at the if if we're just using standard

3:16

git what what is what's the problem

3:17

there

3:18

>> so I mean the good thing is actually

3:20

sort of lesser problem is like agents

3:21

are really good at git

3:24

>> um but

3:26

we're running tens of agents maybe in

3:28

the background right we get a bunch of

3:29

open code or codeex or cl code sessions

3:31

right you've got a bunch of sub aents

3:33

right

3:33

>> right

3:34

>> maybe you actually want to commit more

3:35

so you can always roll back so the

3:37

agents It's not like blowing away a

3:39

bunch of work. We've all seen the horror

3:40

story of someone like burning, you know,

3:43

a ton of tokens for an hour and

3:44

forgetting to commit, right? It's great

3:45

if you can get away from that. Um, but

3:48

also kind of sucks if you go to commit

3:49

and push and like, you know, uh, the

3:51

upstream is down and not available.

3:53

We've all been seeing that as well. And

3:54

so,

3:55

>> again, we're just not ready for the

3:57

volume um, that we have, but also maybe

4:00

the volume we actually want is we want

4:01

to commit more often. We want to push

4:02

more often. um how do you give every

4:04

agent you have like an isolated repo

4:06

that it can act on without crushing the

4:09

other one that's a shared infrastructure

4:11

um

4:12

>> right

4:12

>> at this kind of scale

4:13

>> avoiding avoiding merge conflicts and

4:15

all that right

4:17

>> in many cases you may not even want that

4:18

that social element right you may want

4:20

to actually have the

4:21

>> agent pull down an independent copy of

4:23

that repo say for code review go and

4:26

review that right isolate it act on that

4:29

right in its own right and then maybe it

4:32

post some comments up to a centralized

4:34

platform, right? Maybe your internal CI

4:36

system, things like that, right? But

4:38

like if you can isolate its actions

4:40

while it's actually working as much as

4:41

possible, um it ends up being really

4:43

great, right? Um if you don't have to

4:46

kind of have every agent clone your repo

4:48

from GitHub or get blocked if things are

4:50

down, that obviously makes your team

4:51

more effective as well.

4:53

>> That's awesome. So it's it's changing a

4:55

little bit, but thinking about how it

4:56

works. What does it Oh, I guess I was

4:58

just going to ask what does it look like

4:59

in code here? So, so should we should we

5:02

walk this a little bit about what it

5:03

feels like?

5:03

>> Yeah. So, um we thought you know again

5:07

you take away the human element. We're

5:08

not building

5:11

you know a social network for code.

5:12

We're not thinking about pull requests.

5:14

Right.

5:14

>> Right.

5:15

>> But agents again are really good at git.

5:18

Um

5:20

so there's sort of three ways we think

5:21

about this. Right. There's the

5:23

programmatic control plane. How do I

5:24

spin up one 10 millions of repos at

5:28

scale? one for every agent on the fly as

5:30

they need them or clone something from

5:32

GitHub on the fly, right? Um so what we

5:35

call artifacts, a few lines of code,

5:37

like millions of repos, right? You can

5:38

create that through workers. Um you can

5:42

issue read tokens, write tokens, you

5:43

say, "Hey, straight to my git client,

5:45

straight to my agent harness. Here's a

5:48

regular git remote. I just clone it. It

5:50

works on it. It pushes from it, pulls

5:52

from it. The agent doesn't have to know

5:53

anything other than git, which is huge

5:55

because agents really do know git. It's

5:56

in the training sets, right? you're not

5:58

teaching it a bunch of skills or hoping

6:00

that it like understands this magic new

6:02

API, right? It just talks git and your

6:04

orchestration can take care of that. If

6:06

you kind of look at the next example,

6:07

well, yeah, there's still a bunch of

6:09

stuff that you might want to pull down

6:11

and seed your agents, but then have them

6:13

work independently, right? Like again,

6:15

for code review or for snapshotting

6:18

something or maybe a template, right?

6:20

>> Yeah, we can still pull something down

6:22

and kind of clone from say GitHub. Say

6:25

you want to go and work on like the

6:26

workers SDK which is our Wrangler tool

6:28

chain. Maybe work on some stuff there.

6:29

Isolate and have your agent do a bunch

6:31

of work that commits it and persists it,

6:33

right? It's in a sandbox. So you don't

6:35

want on your local disc where it can get

6:36

lost. You don't want to push it up to a

6:38

pull request yet because you're not

6:40

ready to do that, right? Do you want to

6:41

work in isolation? Um artifacts let you

6:44

go and fork from something upstream as

6:45

well or even clone itself and have you

6:48

know 5,000 copies of the same repo

6:50

inside artifacts. Um, and then with the

6:53

last example, I was like, well,

6:55

>> it's kind of like a more more

6:56

disposable, right? It feels like you get

6:58

>> Exactly. Yeah.

6:59

>> Yeah.

6:59

>> Yeah.

7:00

>> Yeah. And like maybe it's disposable,

7:02

maybe it's not, but you kind of get to

7:03

make decision as you go.

7:04

>> Yeah. Yeah. Did it Did it get it? Oh, it

7:06

did. This isn't it. This isn't it. This

7:07

isn't it. This one's it. Yeah. And

7:09

you're Yeah. Yeah. Cool. Correct me if

7:12

I'm wrong. We don't need to just run

7:13

this in workers.

7:14

>> Yes. Exactly. Like right if you do, but

7:17

you might run your control plane

7:18

somewhere else somewhere else, right? In

7:19

another cloud, right? you might want to

7:20

still orchestrate creating these repos

7:23

using artifacts before we've even used

7:24

workers, right? Um we expose this over a

7:27

regular HTB API. Um we have language

7:30

specific SDKs, you know, TypeScript, Go

7:32

uh and Python. Um you obviously can

7:35

co-generate these day particularly in

7:36

the world of agents, right? Uh language

7:38

specific SDK if you have something in

7:39

Rust or Elixia um or any other sort of

7:42

language that you you know that your

7:43

team might be using, right? Um and use

7:46

that to manage artifacts. like we want

7:48

this to be a case where the control

7:50

plate side can be operated from anywhere

7:53

and then the part that I'll sort of talk

7:55

through now is also the actual like git

7:58

side right like git is a really really

8:00

powerful protocol

8:02

>> for managing like versioning like again

8:04

why teach agents anything else if git is

8:07

right there it's very very good they're

8:08

very good at it

8:09

>> um but it's also kind of nice if your

8:11

environment say maybe you actually are

8:12

in a worker or maybe you're in a nodejs

8:16

somewhere you can host nodejs test or a

8:17

Python application. Um what if you don't

8:21

want to run a full git or spawn out to

8:22

git um in that process, right? It's kind

8:24

of heavy weight. It doesn't really work

8:25

in a lot of service environments in a

8:26

way that you might expect. Well, with

8:29

artifacts, we also let you interrupt uh

8:31

interoperate with it um via language

8:33

case and via those APIs too. So you can

8:35

do some you can commit files, you can

8:38

clone those repos down inside that

8:41

language and have that as like you know

8:42

say a JavaScript object for those files

8:44

as well. Um there's a lot of those sort

8:46

of sandbox or lightweight use cases say

8:48

like with dynamic workers which we just

8:50

dish up recently right to sort of have

8:51

sort of sort of lightweight workers that

8:53

you sort of explore on the fly they

8:54

still might want some gitl like concepts

8:56

where they're acting on files and

8:57

committing files just without the the

8:59

full git client embedded and so we've

9:01

sort of taken two approaches there.

9:02

>> Super cool. So walk walk me through that

9:05

real quick. Let's let walk me through

9:06

the we've got a lot of lot of new

9:08

concepts coming in. So so walk me

9:10

through a dynamic worker that has this

9:11

git or is making use of this git. walk

9:13

me through that really quick through

9:14

this artifacts.

9:15

>> Yeah. Um, great question. So, you know,

9:18

I've got uh an artifacts namespace,

9:21

right? I can create as many repositories

9:22

inside that as I want. Could be

9:24

thousands to millions or more.

9:25

>> Um, say I'm spinning up a dynamic worker

9:28

that I want to go and execute some code

9:30

or tool as part of an agent harness.

9:33

>> Um, but I want to give it a place to

9:35

maybe persist its output. Um, persist

9:37

some files, right? Um maybe it I wanted

9:39

to take some files that it's already

9:40

worked on in the past in that dynamic

9:42

worker and that artifacts repository

9:45

could potentially be tied to the same

9:46

customer or agent right that have the

9:48

same ID. I I can address that how

9:50

however I want. When I spawn that

9:52

dynamic worker I can pass in a reference

9:54

we call a binding um to that repository

9:57

and that dynamic worker can then act on

9:59

just that repository on the fly. I if it

10:02

wasn't created before, I can create it

10:04

at the same time I create dynamic

10:05

worker. If I just need to persist those

10:08

files and then pull them out outside of

10:10

the user code, then I have access to

10:12

that in my control plane.

10:13

>> And I've got again not just git, but

10:15

I've got this like versioned file

10:16

system.

10:17

>> I was talking to a friend who's a

10:19

founder about this and he's like

10:20

>> the get stuff's really cool, but I

10:22

actually have this use case where like

10:24

my customers build workflows and I have

10:27

these config files and these workflow

10:28

definitions and they want to be able to

10:30

roll back to them in different cases,

10:31

right? for like part of their product

10:32

surface and it's like a version file

10:34

system that I can go and do that with is

10:37

really kind of powerful, right? And get

10:39

semantics just make it kind of easy to

10:40

understand.

10:42

>> Yeah. Yeah. Yeah. But and for the humans

10:43

as well, I think it's also it's readable

10:45

for us humans and also for the agents. I

10:47

had I hadn't thought about that. I think

10:48

that's neat. So like a dynamic worker

10:50

could be versioned, you know, I've been

10:52

I've been playing build some stuff and

10:53

like I've been wanting to like find the

10:55

right way to show code and then let them

10:56

go and edit that. And so that's that's a

10:58

super that's how to do that. I feel like

11:01

yeah yeah yeah iterate inside of a dyn

11:03

dynamic worker on on building cod that's

11:06

awesome super cool now that totally

11:07

landed for me thank you for that okay so

11:11

um

11:13

uh here's some use cases that we got so

11:15

agent workspaces we've been talking

11:17

about that one let's talk about config

11:19

versioning here a little bit give me

11:20

give me walk walk me one of those

11:22

>> yeah so I sort of partially touched on

11:23

that but I think we go a little bit

11:24

deeper like you know if I'm talking

11:25

about you know um again a startup that

11:29

like wants subversion like customer data

11:31

but like again there's a lot of cases

11:32

like where if you think about git what's

11:34

git really good at git is really good at

11:36

storing lossful objects in versioning

11:39

them and letting you version control

11:42

those versions but also like figure them

11:44

out chronologically you can roll back

11:45

you can revert right you can go and

11:47

check out a previous commit and so kind

11:50

of turns out to be particularly powerful

11:51

if you have like again as part of your

11:53

product surface right um maybe user

11:55

generated configuration right for their

11:57

products or their product surface

12:00

You could spin up a artifacts repo for

12:01

every user on your platform. They never

12:03

have to know about the git. You're not

12:04

exposing them git, but you're using it

12:06

to version the artifacts or the files or

12:08

the objects that they're operating them

12:10

on. You can diff them like code if you

12:11

want to dip them like code. You can roll

12:13

back. You can expose that roll back

12:14

capability to users. And now you have

12:15

this way for users to maybe even flip

12:17

between versions of config and snapshot

12:20

and make your platform a little bit more

12:21

programmable. As a result, we've even

12:23

been talking about ways we want to use

12:24

this inside cler for where there's kind

12:27

of git semantics and the way artifacts

12:29

work which is distributed

12:31

>> um for some of this for our internal

12:32

services where it'd be great if I have

12:34

this like gitl like place where I can

12:36

roll back to a commit that's yesterday

12:38

for known working um that is config. I'm

12:42

not even thinking about git. It's just

12:43

again a really good way to express this

12:45

kind of versioning system.

12:46

>> Yeah, I think like I we've all built

12:48

those, right? We've all gone and built

12:50

like a change log sort of thing, right?

12:52

And I guess it would totally make sense

12:54

if it was in Git. That that's awesome. I

12:57

I my You got me going. My My brain is my

12:59

brain's going like crazy right now on

13:00

what we could do. And then of course

13:02

platform uh manage repos is the other

13:04

one here. Uh

13:06

>> Oh yeah, that's true. Right. So like so

13:09

you're going to go you're going to go do

13:10

your Terraform stuff. That's what that's

13:12

what we're talking about here, right?

13:12

>> Yeah. Exactly.

13:13

>> Yeah. Again, you've got could be like it

13:15

could be notebooks, could be

13:17

infrastructure as code use cases, right?

13:19

Um content that's generated like web

13:21

assets or JavaScript CSS assets, right?

13:23

As like part of maybe like a web

13:24

platform you're building, right? All of

13:25

those things

13:27

>> really work when they're on like version

13:29

control. Again, get just a really

13:30

expressive way to sort of solve that.

13:32

But it's really nice. You don't have to

13:33

run the get infrastructure. You have to

13:34

secure it. You don't have to scale it.

13:35

You don't have to think about how you

13:37

spawn them at runtime and clean them up,

13:39

right? Or make them durable. I think it

13:40

obviously if um you know a particular

13:43

user or uh agent is dependent on a

13:46

particular repo well like you want that

13:48

to be like distributed you want that to

13:49

be highly available um if you're then

13:52

having to replicate it and deal with all

13:53

the storage constraints and do that like

13:55

you know that is a lot of toil and work

13:57

on your teams like our job is to just

13:58

make that work all the time. If you want

14:00

to go and check that repo out it should

14:01

be available no matter where that

14:02

agent's running in the world.

14:04

>> Awesome. I I I love it. What are some of

14:07

the unlocks that this gave you that you

14:09

didn't see coming? Like uh cuz I'm

14:12

feeling it. You're telling me these

14:13

unlocks and and you're you're I didn't

14:15

think about that.

14:15

>> Yeah. So, one thing we've been starting

14:17

to use this for uh internally and

14:20

actually talked to a bunch of customers

14:21

about is in our sandbox use case, right?

14:23

And so, um you know, we've got this

14:26

internal sort of like what called

14:27

background agents inside cloud players

14:28

that can be driven from chat and you can

14:30

say, "Hey, go work on this ticket,

14:31

right? Go and solve this problem. don't

14:32

grab these logs from like our elastic

14:34

search stack and then go figure out give

14:36

me a first pass at like a diagnosis on

14:38

this problem that I'm trying to solve.

14:39

Really, really powerful, right? Kind of

14:40

these SR type pieces sort of like you

14:42

know ad hoc engineering task you can go

14:44

solve. There's a lot of cases where I

14:46

want to go and share a link to somebody

14:48

else of like here's a session I was just

14:49

running. I kind of got stuck or like

14:51

here's my thinking on this like what is

14:53

your thought?

14:55

>> And you know I'll get to this point in a

14:56

moment you're like okay cool great like

14:58

I've seen that before. How cool would it

14:59

be if I could just from that link go

15:02

fork and get an isolated clone of that

15:04

whole session and all the working files,

15:06

right? All the files that

15:08

>> that it was working on, all of the

15:09

potentially the config or the notes that

15:11

it's take taken or any other files from

15:14

other upstream repos at that exact point

15:16

in time and go, hey Ben, here's a here's

15:18

a version of this. Can you go like work

15:19

on this instead of this like four call

15:20

this multiplayer type environment? Turns

15:22

out artifacts is really really powerful

15:23

there because artifacts can just clone

15:25

its own repos. You can just call clone

15:27

as many times as you want on any

15:30

independent repo. You get a fully

15:32

isolated copy. And now I or say Ben can

15:35

go and work on the thing that I was

15:36

working on without touching or messing

15:38

with my workspace, right? With his own

15:40

set of privileges. He can go work on

15:41

that and go, "Hey, actually by the way,

15:43

I was working on this overnight or I was

15:44

in a different time zone and I think

15:45

I've like kind of cracked this problem

15:46

and like show me back the results,

15:48

right?" That's really really powerful.

15:50

So we can kind of separate this stuff

15:51

out. And so it's not just about I said

15:53

not just about the git part. That's just

15:55

a really really powerful primitive and I

15:56

think you know being able to speak git

15:58

is sort of the language of agents in

16:00

many ways but you end up being able to

16:01

again like if you can just arbitrarily

16:03

copy these on the fly without having to

16:05

like pull all the like bigger repo

16:07

constraints or things like that it can

16:08

just fork fork fork fork um yeah now

16:12

I've got this kind of like multiplayer

16:13

like concept right and I think that's

16:15

comes back to how we built this out on

16:16

durable objects which has always sort of

16:18

been inherently this sort of multiplayer

16:19

concept of of coordination

16:21

>> right and so I was that that was my

16:24

question. It's all durable objects all

16:25

the way down. This is a durable object.

16:27

>> It always is. Yeah.

16:28

>> Yeah, it always is. Can you tell me a

16:30

little bit more about the implementation

16:31

details? So, how do how do we make it

16:33

work with Git? Let's get a little nerdy

16:34

here.

16:34

>> Yeah. So, this is kind of actually how

16:36

we kind of brought this together really

16:37

really fast. You know, we've got dribble

16:39

objects, our sort of, you know, best way

16:41

to describe it is sort of like stateful

16:42

workers or stateful serverless

16:43

functions. I think we found that's the

16:45

the best way to resonate. It's it's not

16:46

quite comparable to anything else, but

16:48

they have an embedded SQLite database.

16:49

They have state. They can spin up in any

16:51

place in the world. They're It's in the

16:54

name. They are durable, right? Like

16:55

they're they're the storage is

16:56

replicated under the hood. Great. Today

16:58

they speak, you know, workers. They

17:00

speak HTTP. You can speak RPC to them.

17:03

Like gets another protocol. Git works

17:05

over HTTP. Git has its git server is is

17:08

non-trivial, but we can also run wasn't

17:11

inside Drupal objects. And so the way

17:13

this came together is that um one of the

17:15

folks on team Matt actually had a zigg

17:18

implementation of a get server and we

17:21

said what if we ran that inside Drupal

17:23

objects so that you get

17:26

>> as much git API coverage as possible. So

17:28

again it's so any git client just works

17:29

you're not thinking about like does this

17:31

particular git feature work does this

17:32

other not work like we have almost

17:34

complete coverage as much as possible

17:36

right we use a git client real git

17:37

clients to test our API service which is

17:39

awesome too so that it's fully end to

17:41

end tested but it's fundamentally like a

17:44

git server running inside a durable

17:45

object um and every repo is represented

17:48

by a durable object the cool thing is

17:50

gets hyperefficient in terms of storage

17:52

so even like some of the larger repos

17:53

like say like our own like workerd or or

17:56

workers SDK for sort of 200 to 500 B.

17:59

Something like Next.js is about 2 and a

18:00

half gigs. Um, so they fit really well

18:02

within sort of like the 10 gigs at a

18:03

boundary of a durable object. Again, the

18:05

cloning semantics is powerful if you

18:06

want to sort of clone them on the fly.

18:08

That works really well with durable

18:09

objects. And we've already kind of

18:11

proven that this can scale to, you know,

18:13

tens of millions to hundreds of millions

18:14

of objects under the hood. And so we

18:17

didn't have to go and rethink this whole

18:18

architecture in this world of agents. We

18:20

had this primitive that really, really

18:22

works and that's really efficient. Um,

18:24

there's a lot of optimization we need to

18:26

do still on the WASM layer and make that

18:27

kind of work, right? We're launching in

18:28

beta today, but we spent a lot of time

18:30

kind of testing this instead of banging

18:31

against it. It turns out this kind of

18:33

primitive where how can I create this

18:35

thing on the fly almost instantly and

18:39

then address it from any part in the

18:40

world. That could be a durable object, a

18:43

multiplayer server, right? It could be a

18:46

git repository that any agent in my

18:48

fleet can talk to and pull as it needs

18:50

or my 10,000 agents can go and have

18:53

10,000 repos on the fly. Whichever way I

18:55

want to sort of split that and so it's

18:57

really really powerful for us of not

18:58

having to worry about these

18:59

infrastructure concerns. I think my last

19:01

thought there is really all the

19:02

parameters we've been building as sort

19:03

of part of the val platform at Fler and

19:05

part of workers right let us go and do

19:07

these things really really quickly which

19:08

is great. We're not spending two years

19:10

building the foundations on the

19:11

infrastructure to go and ship this. We

19:12

say, "Hey, actually, I think there's a

19:13

problem we can go solve for customers."

19:15

>> Nice. So awesome. And I think people are

19:17

going to build on top of it as people

19:18

do, right? When we get we give them the

19:20

primitive, they start building on top of

19:21

it. They figure out what the use case is

19:23

even before we That's a great use case

19:24

for it. So super excited.

19:26

>> Yeah. I' love to see someone build like,

19:28

you know, something in the social code

19:30

sharing space or more in the multiplayer

19:31

side, right? I think there's going to be

19:32

a tons of ideas that come out of this

19:34

and hopefully we've seated a few of them

19:35

in the docs and in the blog posts and in

19:37

some of the examples we're working on.

19:38

We'll show some more of this stuff over

19:40

the coming weeks as well. But I'm really

19:41

really excited to see what people build.

19:43

>> Cool. Awesome. Same. Same. Thanks so

19:46

much for jumping on here and thanks

19:47

everybody for watching from home. Any

19:49

last things to drop here, Matt, before

19:52

anything you want want them to do? The

19:53

people watching this?

19:54

>> Yeah, go read the blog if you haven't

19:56

read the blog. That's probably the most

19:57

important blog. U today it's one of the

19:59

feature blogs. You can go to

20:00

developerscloud.com/artifacts

20:03

uh for the docs and just go get started.

20:05

Um if you're wondering how do I get

20:07

started? Is there a wait list? No wait

20:09

list. It's public beta. You can go start

20:11

today. You just need a workers paid

20:12

plan. That's the only restriction right

20:13

now. And it eventually it'll come to

20:14

free as we work through the beta as

20:16

well.

20:16

>> Let's talk about workers paid plan

20:18

really quick. What does that mean?

20:19

>> Workers play. Uh great question. $5 a

20:21

month gives you access to workers, D1,

20:23

durable objects, KV, pretty much

20:25

everything on the workers platform. Tons

20:26

of included usage up front before you

20:29

even start paying you about that $5. Um

20:30

and that does include artifacts as well.

20:32

Um and so that's really the only area to

20:35

get started and not again not going to

20:36

put you on a wait list to make you wait.

20:37

Just want you to get started and start

20:38

building. So,

20:39

>> I know that people have been asking for

20:40

this and trying to solve problems that

20:42

this is probably going to solve for

20:44

them. So, I'm I'm super excited and I'm

20:46

so glad that it's out of redacted time

20:47

and we can talk about it. We can start

20:48

getting the conversation happening

20:50

around this. Uh, thanks everybody for

20:51

hanging out uh during agents week again.

20:53

More stuff about artifacts and all sorts

20:55

of stuff that we announced today coming

20:57

out real soon. So, uh thanks for hanging

20:59

out. Thank you Matt for being here with

21:00

us and uh we'll see you next time.

More transcripts

Explore other videos transcribed with YouTLDR.

Get the TLDR of any YouTube video

Transcribe, summarize, and repurpose videos in 125+ languages — free, no signup required.

Try YouTLDR Free