0:00

Welcome to Syntax. Today,

0:02

CJ and I sat down

0:05

with Paul Koppelstone, the co-founder

0:07

of Superbase. We

0:09

dive deep into the tech that

0:11

powers Superbase. We talk all about

0:14

Paul's outlook on open source and

0:17

the origins of Superbase, including why

0:19

it's super and not super. So

0:22

without further ado, here's Paul. So

0:31

welcome to Syntax, Paul. If you want to

0:33

give us a quick introduction about maybe who

0:35

you are and what you do at Superbase,

0:37

and then we can get into some of

0:39

the things a little bit deeper about Superbase

0:41

as a product. Yeah,

0:45

thanks for inviting me. My name is

0:48

Paul Koppelstone. I'm the CEO and one

0:50

of the co-founders of Superbase. We are

0:52

basically a open source Firebase

0:54

alternative is how we position ourselves early on.

0:56

But you can think of us really as

0:59

a Postgres hosting company. Nice. Yeah.

1:02

And as a Postgres hosting

1:04

company, or even an alternative there, I've

1:07

always been impressed with

1:09

UI of Superbase, how it makes things feel

1:12

nice and easy for the

1:14

user to get up and running. It's not like you're

1:16

managing a database, right? So with

1:19

that said, here's a quick question for

1:21

my five-year-old. Why super and

1:23

not super? Yeah,

1:26

because it's very hard to

1:28

purchase the domain name Super.

1:31

That was literally it. Domain

1:34

driven naming. That's pretty

1:36

much how I do all my side projects. Yeah,

1:38

nice. Which turns out to be both, by the

1:40

way, a blessing and a curse.

1:43

The blessing is that it's very easy

1:45

to find mentions of Superbase

1:47

across the internet. And

1:50

people have started putting Super on everything.

1:53

The curse is that it doesn't feel

1:55

that enterprisey. It's

1:58

kind of like a bit of a meme, which is also. by

2:00

the way, where the name came from, like,

2:03

and to myself like to meme a lot. And

2:05

so, Super Bass obviously is

2:07

a Nicki Minaj song and we thought

2:09

it was funny that it was called

2:11

Super Bass and that

2:13

somehow stuck. Yeah, but here

2:16

we are. It's working. Yeah.

2:19

Yeah. I always thought, I always think of that there's

2:21

an MF Doom song where he says he's super in

2:23

it. And I always think about that when I hear

2:25

Super Bass. So do you want to maybe give a,

2:28

some background into, how did Super Bass

2:30

start? Even from like

2:32

technical foundations to, you know, why? Yeah.

2:36

So timeline first is

2:38

we started in January

2:40

2020. So really just

2:43

for four years old, four and a half years

2:45

old now, we went to YC

2:47

in that summer batch as well, and

2:49

have just kind of been building fully

2:51

remotely. So started during COVID and building

2:54

the company remotely. The sort of genesis

2:56

was that in my previous startup, I

2:58

was building using Postgres, but

3:00

one part of the tool was a chat

3:02

application and I was using Firebase. I actually

3:05

really love Firebase. I think the developer experience

3:07

is amazing, but we

3:09

hit some technical issues in terms

3:11

of like the performance around sending

3:14

the messages. And so

3:16

I looked at migrating it to Postgres

3:18

and I did it quite naively at

3:20

the start. Postgres has this listen notify

3:22

system where you can push messages out

3:25

of it. And I thought that was

3:27

good until I ran into

3:29

another technical limitation there on the

3:32

Postgres side. And so I had to build

3:34

a real time engine kind of

3:36

on top of Postgres. And I did that

3:38

with Phoenix Elixir, which

3:41

kind of was the right tool for the job. You

3:43

can connect lots of clients to it. It's good for

3:45

WebSockets. And we did it

3:48

by kind of listening to the

3:50

database replication stream, you know, and

3:52

then you convert those messages into,

3:54

yeah, literally JSON payloads, and you

3:56

can blast them out to the

3:58

WebSockets. So is it cool? technical

4:00

solution and I'm a big hack and use reader

4:02

and I put it on show hack news and

4:04

it got a lot of interest

4:07

at that point and so I Knew

4:10

I was going to do a dev tool startup

4:12

after my last startup And so that

4:14

seemed like the right time and I tapped my co-founder on

4:16

the shoulder and said this is what I want to do

4:18

I want to go to I see Should

4:22

we do it and yeah somehow convinced

4:24

him to leave his startup for to

4:26

help me build Superbase? Wow,

4:29

so how long after? Like, you

4:31

know first lines are written Did

4:33

you end up joining YC? Oh? Yeah,

4:37

so probably I think the hack and use

4:39

launch that I did that one was on

4:42

For just that real-time engine was maybe

4:44

in like September 2019 We

4:46

started the company in January 2020 and

4:49

then we sort of got accepted. Maybe I think

4:51

it was March or April of 2020

4:54

no and we didn't have

4:56

Wow pretty fast quite on But

4:59

we're both three-time founders. We

5:02

were shipping fast. We tried to

5:05

Check off all the things that why

5:07

see like like to like yeah. Yeah, that makes

5:09

sense So yeah, I like I like the fact

5:11

that super base is open source and

5:14

can be self-hosted Was that a decision you made early

5:16

on that you were gonna make this thing and keep

5:18

it open source? Yeah,

5:20

honestly, it wasn't even really a conversation

5:22

at the start I think just and

5:24

and I kind of philosophically love open

5:26

source I wouldn't want to work on

5:28

a dev tools if it was

5:31

closed source. That's kind of the truth. So

5:35

Yeah, that's that's about all the thinking that went into

5:37

it and then we try to support Like

5:40

our philosophy on open source I think is

5:42

a little bit unique in that

5:44

you know a lot of the tools that make

5:46

up the super base stack Are

5:49

all existing open source tools and that was

5:51

really important to us as well It felt

5:53

like open source is a lot of like,

5:56

you know, I'm gonna do some New

5:58

way of doing this and it's got that'd

6:00

be my own, but we actually like the model

6:02

where you can try to support the

6:05

tools that exist. They might not be perfect,

6:07

but we can help to try and make

6:09

them perfect is the idea. So that's kind

6:12

of our philosophy around open source,

6:14

one of the key philosophies. It's

6:17

one of the things that first drew me to Superbase. I've built

6:19

a few projects with it. And the simple

6:21

fact that it's built on top of Postgres, like I

6:23

know a lot of times when you're choosing a new

6:25

tech for a project, you worry about

6:27

like vendor lock-in or like, am I

6:29

going to have to keep using this if I don't like

6:31

it or something like that? But to me, it was super

6:34

cool to see that it's built on top of Postgres. I

6:36

can use everything I know about Postgres. And

6:39

if I really ever wanted to stop using Superbase, I

6:41

could, I could just literally take my database with me,

6:44

which was nice. Which has happened

6:46

in the past. We have like in

6:48

our docs, one of the core principles

6:50

is portability. You should be able to

6:52

take anything you build, just

6:55

basically PG dump and take it to

6:57

your favorite Postgres provider. And there might

6:59

be some things depending on how far

7:01

you lean into like some of the

7:03

tooling that we've got. But in theory,

7:05

like you can extract that. So we

7:07

try to make everything very composable. So

7:09

even though it's like a full back

7:11

end, you can just use the database

7:13

and stack other tools. Clerks very common.

7:15

You can use Superbase Auth or you

7:18

can use Clerk with Superbase. That's kind

7:20

of how we want it to be.

7:22

Do you have any stats on like how many people self-host to keep track

7:24

of that kind of thing? No, we

7:26

have no telemetry. Another thing that we think is

7:29

like if you want to self-host, that's

7:31

great. Anecdotally, I see a

7:34

lot of people do it. So we're

7:36

happy. And actually, yeah, some of the

7:38

larger customers do it as well. So

7:40

that's nice. But because I'm a developer

7:42

and I don't like tracking, then I

7:44

can't give you any stats. Same.

7:47

Yeah, that's awesome. I think it fits very well

7:49

into, you know, I don't want to talk too

7:52

much about Sentry because this isn't an ad for

7:54

Sentry. But if it's very well and with what

7:56

we're doing here at Sentry as well, you know,

7:58

open source tool. We

12:00

build our credibility around Postgres. We build

12:02

a lot of tooling for Postgres. We

12:04

build a lot of scalability into Postgres.

12:07

And we're having to do this, contributing

12:09

up, employing open source

12:11

maintainers, making sure that

12:13

the tool will go

12:16

beyond anything that they'll even see on

12:18

say AWS or on

12:21

one of the big three clouds. We wanna make

12:23

sure that we can really

12:25

fit that vision so that when we go

12:27

to enterprises, we've got 10 years

12:29

of credibility and scalability and logos

12:32

to prove it. So

12:35

looking at the docs, I see that there are

12:37

quite a few client libraries, not just

12:40

JavaScript. So I see things for like Flutter

12:42

and then in the recent GA announcement, you

12:44

announced the Swift client, but then

12:46

there are community packages for like Python and

12:48

C-sharp and Kotlin. So how do you decide

12:50

which client libraries are gonna be official versus

12:52

like keeping them more like community maintained? Yeah,

12:57

not easily. Yeah. So

13:00

we developed first our

13:02

JavaScript libraries and TypeScript libraries,

13:05

because that was the thing that was

13:07

really missing when we started was there

13:09

was no backend for the likes for

13:11

the jam stack. That's

13:14

how I'll identify. And so we really

13:16

focused in on that use case first.

13:18

So JavaScript was the natural thing. How

13:20

do you as

13:23

well, like this type of front end

13:25

developer, I would say is not so

13:27

familiar with the nuts and bolts of

13:29

Postgres. And so it's a really easy

13:31

win. We can just abstract

13:34

that away and make it incredibly easy for you

13:36

versus say maybe like a PHP

13:40

developer, popular now, probably

13:42

know a bit more of the nuts and bolts

13:45

setting up ORMs and things and they're gonna use

13:47

their Laravel and it might handle it for you.

13:49

So that was the use case

13:51

that we focused on. Then

13:53

over time, just of course, a lot of

13:56

people started piling in and requesting

13:58

all sorts. mobile became very popular

14:01

because of Firebase was very popular

14:03

and people wanted to migrate to

14:06

Superbase. And so they

14:08

wanted mobile. We ended up doing Flutter

14:11

first because it was kind of like a

14:13

new system and that actually

14:15

came from the community. This is like kind of

14:17

our hiring strategy. We had

14:20

this person contributing. They

14:22

did the open source Flutter libs.

14:24

It started getting popularity. And

14:27

so we just hired him to work

14:29

on the full time. And he's a dev role.

14:31

That's a good lesson. Yep. Yep.

14:34

You might say Tyler is a Japanese

14:36

dev role actually. So he is based

14:38

in Japan. Funny enough working at HubSpot

14:41

as a like just an engineer at

14:43

HubSpot, nothing to do at dev role.

14:45

It turns out he's also just very

14:47

good at this type of stuff.

14:49

So it was cool. We've

14:52

done a lot of that by the way, like people

14:54

who contribute to our community and they're

14:56

kind of too good to ignore that we're just going

14:59

to hire them. Yeah. Nice.

15:02

Yeah. So I think that's

15:04

kind of how it happens. Like we just

15:06

have people contributing and then eventually something becomes

15:09

popular enough that the community are kind of

15:11

shouting that, hey, can you make this official?

15:13

Then we pull the trigger. That's

15:16

probably a good strategy. Yeah. I

15:18

guess I should also mention for anyone else that's

15:20

listening that isn't familiar, you don't necessarily need a

15:22

client library though. I think that's one thing about

15:24

Superbase is you could connect directly to the database

15:26

and just use your preferred ORM or you could

15:29

just use your favorite REST

15:31

API client and talk to the REST API. So

15:34

it's important to note you don't need these client libraries, but it is

15:36

cool that you have that support as well. Yeah.

15:38

Thanks for shouting that out because like,

15:40

yeah, it's actually one thing that we

15:42

did too well at the start was

15:45

positioning like some of the Superbase tools,

15:47

the additional tools that at

15:49

the start, now we're having

15:52

to change the narrative that you don't actually have

15:54

to use all those tools. You can just use

15:56

the database and like I'm a

15:58

big fan of Drizzle, for example, becoming very

16:00

popular with Superbase.

16:02

But Prisma is popular, Keesley,

16:05

all of these, whatever you want, you

16:08

can plug your

16:10

Ruby and anything

16:12

you want on top of the Postgres database

16:14

and we're happy. That's where we want to

16:16

be. Yeah, and you could take

16:19

advantage of the Superbase platform itself.

16:21

I don't have to worry about

16:23

the client's outline. We've got a

16:25

ton of cool database stuff like

16:27

index advisors, security advisors, and scaling

16:30

out, rep replicas, just making sure that it

16:33

is easy on the database side.

16:35

The additional stuff is there if

16:37

you need to reach for it.

16:39

Great for prototyping and the Postgres

16:41

maxis, like myself, love the additional

16:43

tools. But yeah, if

16:46

they're not for you, then that's kind of fine.

16:49

You guys also have edge functions, right? Is

16:52

the idea with edge functions there

16:54

to kind of remove the need

16:56

for any kind of backend server

16:58

component to your app? Superbase completely

17:00

would become the backend of your

17:02

application. You can just use an

17:04

edge function when you need server

17:06

side functionality. Actually,

17:08

the reason... So we

17:12

started doing these things, launch weeks, I think they're

17:14

pretty popular now. And every single

17:16

launch week would say, yeah, guess

17:18

what? We're shipping people will

17:21

be like, functions. We'd be like,

17:23

no, no, we're not. And

17:25

then after like four launch weeks, people

17:27

will win the edge of each of

17:29

the functions. So we

17:31

had to scout around to find

17:33

an open source runtime for running

17:35

functions. And Dino, of course,

17:38

sort of made the cut. But actually,

17:41

the way I like to position

17:43

our functions is because they serve

17:45

a very important purpose with databases.

17:47

It's not to like, for

17:49

example, if you're using Next.js, I actually recommend

17:51

you just bake all of that sort of

17:53

routing logic in there. But where

17:55

it's useful on the database side, I

17:58

like to refer to them more as a

18:00

database. like background workers. So

18:03

as an example, we're becoming very popular for AI applications because

18:06

of PG vector. You can

18:08

store your embeddings in Postgres, do RAG, all these sort of things.

18:11

But creating the embeddings itself,

18:14

you wouldn't really want to do it inside the

18:16

database because it's a lot of compute and you

18:18

choose your compute size. So what you want to

18:20

do is maybe someone insert

18:22

some content and then you need to

18:24

shove out this workload to somewhere else,

18:27

like a background worker, and now we're

18:29

starting to bake a lot of this

18:31

embeddings, AI tooling into the edge functions

18:34

itself. So you can now insert a

18:36

million rows of content

18:38

and it will trigger out a million

18:40

of these workers, edge functions. They'll create

18:42

the embeddings and stuff them back into

18:44

the database. We call them database web

18:47

hooks. That's the way to offload

18:49

the compute. Gotcha.

18:52

I was going to just bring

18:54

up, now that you mentioned the

18:56

PG vector and

18:58

that technical bit, is that

19:00

nice because you're built on Postgres that you can

19:03

just utilize something like PG vector? And

19:05

likewise, was that a

19:07

tough technical challenge to implement it to

19:09

an existing product? Yeah,

19:12

actually, no. There's

19:14

another funny open source story that I just want

19:16

to touch on here. So the

19:18

way that PG vector came about was

19:20

because this guy came in, he emailed

19:23

me directly and he's like, I'm building

19:25

this, this is probably February 2023 or

19:27

something like that. He's

19:30

like, I'm building this tool and I need to

19:32

store embeddings and there's this extension

19:34

called PG vector and no one's using it,

19:36

like none of the Postgres hosts. Can you

19:39

please merge it? I've created a PR here

19:41

and I looked at the PR, it's just

19:43

very well structured and everything. And

19:46

I jumped on a call with him and I

19:48

said, what are you doing?

19:50

And he talked me through it and I said, okay, I'll

19:52

merge it as long as you create

19:55

like a chat with your

19:57

docs inside our docs.

20:00

which had actually not

20:03

been done before. And so we

20:05

called it Clippy. You can like search for

20:07

Superbase Clippy as a joke because we're really

20:09

worried that it would be so shit that

20:11

it would give the wrong answers and everything

20:13

like that. Clippy was, of course,

20:16

terrible like that. So anyway, the guy

20:18

the guy did it on a contract and like

20:21

it was so awesome that of course

20:23

we hired him and that's

20:25

the genesis of PG

20:27

Vector. So we were the first Postgres hosting

20:29

company to offer PG Vector and we've

20:32

been working on it now. Now it's

20:34

we're not the main creators. It's a guy

20:37

called Andrew came. But we support him along

20:39

with a few other companies that,

20:41

you know, we're really grateful for as well, because

20:43

the performance of PG Vector in

20:45

the past year has rapidly

20:48

increased, which is pretty cool to see. And

20:51

now the thing that you asked

20:53

like is also a really

20:56

interesting insight, like how hard was it

20:58

to offer this new product? And

21:00

this is like where I think

21:02

our product strategy is a little bit

21:04

unique. You can think of Postgres as

21:06

like a really common substrate. And we

21:09

think of almost everything like a Postgres

21:11

problem. And we have a couple of

21:13

other primitives that make Postgres really easy

21:15

to use. Auth, yeah, the

21:17

background workers, edge functions, storage

21:20

for file storage and, yeah,

21:23

like a few extensions that can trigger

21:25

things. But if you like, these

21:27

are what I call like this sort of first

21:30

order primitives that you need to build

21:32

almost anything. But then we can

21:34

mix and match all of these to

21:37

create these second order primitives. And

21:39

our first foray into like this

21:41

product was actually the vectors.

21:44

Like we've got a whole section of our

21:46

docs called Superbase Vector. But actually, it wasn't

21:48

really a new product. It literally was, hey,

21:50

turn on this extension and here's a bunch

21:53

of docs on how to use it. And

21:55

it turns out that Postgres is so extensible

21:58

that we could do this with several things.

22:00

and probably will eventually search. You

22:03

can use Postgres for searching, queues. You

22:05

can use it for workflows. They're more

22:07

solutions. Here's

22:10

how to use our primitives to achieve

22:12

a certain solution. That's

22:15

what I really liked the most about Postgres.

22:17

Its extensibility gives us that

22:19

option. We can just think of everything

22:22

as a Postgres problem. Definitely. Nice.

22:25

On that note about documentation though, Superbase

22:27

has some of the best documentation. Even

22:29

before you added the AI stuff. I'm

22:32

just curious, did you have a strategy with that

22:34

to make sure you had good docs? And

22:36

then also, it was cool to see Superbase

22:38

did the clippy thing, and now almost every docs you

22:40

go to have AI. Yeah. So

22:44

you can freeform search, and the docs will give

22:46

you an answer for it. So yeah, who worked

22:48

on that? What's your secret? It's super cool. Yeah.

22:51

Very much shared effort over

22:54

the years. I remember docs were extremely

22:56

important at the start. I remember I

22:58

spent a week, and my goal at

23:00

the start was you should be able

23:02

to consume from start to finish everything

23:04

in a couple of hours so that

23:06

you understand what you need

23:08

to. So the conciseness

23:10

was really important, and then we were

23:13

using DocuSource. Eventually, we kind of outgrew

23:15

it. We wanted to make it more

23:17

of a product, like docs as a

23:19

product. The guy, Greg, who

23:21

merged PG-Bekta, was really responsible for a

23:24

lot of that early. The

23:27

AI-type productizing. And then

23:29

we hired someone,

23:31

Karras, who has just focused really

23:33

on this, not only the development

23:36

of the docs, but the

23:38

information architecture of making it

23:40

really concise. And also, our

23:43

goal as docs as a

23:45

product is somehow to bleed docs into

23:48

the dashboard and the dashboard into the

23:50

docs. So you'll start

23:52

seeing the keys appearing and everything.

23:55

But that's kind of the goal. That's

23:57

super cool to see, though, because in the dashboard, at

23:59

least There's documentation, but it has

24:02

my endpoints, my keys in it as well, so

24:04

that's super cool to see. Yeah.

24:06

We have also, because you touched

24:08

on the RESTful interface

24:10

that you get free of charge, it auto-scarefolds

24:12

out of your database. So it's cool, it

24:14

just has an

24:17

open API spec and we

24:19

can literally generate an entire

24:21

documentation page for you. Here's

24:23

the endpoint that you can

24:26

click and here's the Superbase

24:29

JS query that you can run and

24:31

we can embed that everywhere. That

24:34

concept actually came from Airtable. I

24:37

really like Airtable as a tool. Actually,

24:40

I think a lot of their

24:43

DX, like imagine Airtable but it

24:45

was Postgres, then you

24:47

would have this killer product. Because

24:50

they have this auto-generated API as well, but

24:55

their docs would also auto-generate as you

24:57

change the schema. That

24:59

initial version of auto-generated docs

25:02

came from my use of Airtable. Nice.

25:06

Yeah. I suppose that helps keep everything

25:08

up to date. I would imagine once

25:10

your product continues to grow and grow

25:12

and grow, it's pretty

25:14

easy for docs to get out of sync.

25:16

Next thing you know, users are having issues.

25:18

Yeah. Man, it's a

25:20

big challenge with such a large project.

25:24

Kudos for executing on it so well.

25:26

I want to talk about the product again

25:28

in terms of the underlying tech behind it.

25:31

So when you first started it,

25:34

first today, let's

25:37

talk about first the UI tech. So what's

25:40

the UI built on and then how

25:42

has that changed over time? Is it

25:44

same initial code base or is it

25:47

evolved into different platforms or what's the

25:49

story there? So

25:52

yeah, I think what we

25:54

had at the start was a next

25:58

shares for our all. dashboard

26:01

and docuSaurus for our landing page and

26:03

docs. Then we moved

26:07

everything to Next.js and

26:09

we started using a mono

26:11

repo so we could do shared

26:13

components, moved out docs in. And

26:17

that was the goal. The goal

26:19

was to get this really

26:21

consistent feel so you can

26:23

almost feel like the whole

26:25

dashboard becomes one

26:28

key product. That's where Next.js and I think TurboRepo were

26:30

an early adopter there. Just

26:39

to make sure that we could get this

26:41

very consistent feel and UI kit. You

26:45

might see him on our Twitter. He shares

26:47

sometimes some of the design systems and things.

26:50

And he's phenomenal at coming up with some

26:53

great components and things. We've

26:55

also got this very Kaizen approach to

26:57

everything. Sometimes you'll see

26:59

that things are a bit inconsistent in

27:02

our dashboard because he's trying out some

27:04

new component or a pattern that we're

27:06

going to spread out throughout everything. And

27:08

it might just get tested in one

27:11

area. Or he's fixing my

27:13

poor design decision from 2020 probably

27:16

is the other thing that

27:18

he's doing. So yeah, we've leaned

27:20

really heavily into Next.js, which has

27:22

been great, of course. You

27:25

mentioned starting naively. So

27:27

that fits in well with that. I think that's the

27:29

best way to start, though, right? You just get going

27:31

and worry about it later. You

27:34

mentioned Kaizen as well. Do you want to

27:36

just give a quick couple

27:39

of seconds what that is and what it

27:41

applies here? Yeah, it comes from

27:43

the Toyota production system. Kaizen

27:45

means continuous improvement from the

27:48

CEO to the assembly line

27:50

worker is kind of the

27:52

term. So it doesn't matter

27:55

what you're doing,

27:57

everything is a process that can be

27:59

improved. And importantly,

28:01

incrementally improved is one of the things

28:04

that we put a lot of emphasis

28:06

on. Like if people come out

28:08

with big bang projects, we just say no, it's

28:10

got to be done like in a very small

28:12

way, chunk it down to a couple of weeks

28:14

at most and do increments everywhere.

28:17

So even like our branding and

28:19

you know, people like to rebrand

28:21

their website or something like that. We

28:23

haven't done that. We've just like evolved

28:25

the branding of the website. Yeah. Yeah.

28:29

Yeah. Like iOS does that too, right?

28:31

Like, yeah, little components change here or there. Next thing

28:33

you know, it's a very different

28:35

over time. Does that come from you, the

28:37

Kaizen philosophy or did that is that is

28:39

that something that you brought to the table?

28:42

Yeah, I think that one's kind of

28:44

one of my like, the core life

28:46

values actually, I think that everything should

28:48

be approached in this small

28:50

improvement type way. Oh,

28:52

that nice. So talking about

28:55

internals, we talked about the UI tech, it's built with Next.js.

28:57

Can you talk a bit about like the

28:59

open source super based stack? So you

29:01

mentioned the initial thing was kind

29:03

of like a plug in for Postgres. What

29:06

else is involved in that stack? Yeah,

29:10

the total stack is a

29:12

few servers. Each one of them

29:14

is open source. So Postgres, we don't run a

29:16

fork or anything like that. We

29:18

have a Postgres bundle. So our

29:21

bundle has a bunch of extensions.

29:23

That's all we do. The extensions, but that

29:25

same one that you see in our open

29:28

source is the one that you could self-host

29:30

and everything. So

29:32

that's the key component. And then like the

29:35

other components are, for example, as

29:37

one that you mentioned Postgres, which

29:40

is an auto generating API on

29:42

top of Postgres. And

29:45

what you're going to find as I go through

29:47

the tools is every single one is written in

29:49

a different language, unfortunately

29:51

for us. But usually

29:53

they're the right tool for the job

29:56

for a particular reason. So Postgres, actually

29:58

I had used it. this

30:00

stack in my previous startup,

30:03

which is why I was so confident on

30:05

it and it became the early stack. But

30:08

I had also written a blog post about

30:10

it and had reached out to the

30:13

maintainer to make sure everything was okay

30:15

to say in the blog post. So

30:17

I kind of knew him. And then when we were starting Superbase,

30:19

I reached out to him to say, is it all right if

30:22

we use it in Superbase and Promoter and

30:25

anything that we can help with? And

30:27

he said, we're looking for some sponsorship

30:29

and I need $200 a month or

30:31

something like this to keep

30:33

maintaining it. And I

30:36

said, what are you doing? And

30:38

he literally was not working

30:40

on anything. And I was like, well,

30:42

there's no way you can maintain this for $200. So

30:45

we just employed him immediately to

30:47

work on it full time. So

30:50

this is like the best reoccurring story here

30:53

is that you needed something so you

30:55

employed them. Well,

30:57

and it's just such a good common

30:59

good for the world. You know, like

31:02

so many people use Postgres and it's

31:04

so hard as an open source maintainer

31:06

to make money from

31:09

sponsorships. I mean, like $250 a month. The

31:12

poor guy, I mean, he lives in Peru,

31:15

but still for

31:17

the value that the world gets, it

31:19

just seemed completely unfair. So

31:22

anyway, we have

31:24

maintained, he has maintained that

31:27

for us, but for everyone else for

31:29

the past few years. But that's been

31:31

really helpful for the Superbase community too,

31:33

because we've had to patch a

31:36

lot of things that are important for

31:38

our customers as well. But

31:40

we try to make sure that nothing's

31:43

Superbase specific. He makes choices only for

31:45

the community. OAuth

31:47

actually came from Netlify. They

31:50

had an OAuth tool, which I

31:52

guess we didn't have OAuth when we

31:55

launched, which was kind of funny because your database

31:57

wasn't safe. But everyone was asking

31:59

for it. And our promise to the YC

32:01

community, because we got launched on Hacker News,

32:04

was that we'll build auth, but we'll do

32:06

it in a very Postgres native way. And

32:08

they're like, well, good luck. And we spent

32:10

all of our YC batch trying to figure

32:13

out how to do this. And we ended

32:15

up, yeah, finding GoTrue by

32:18

Netlify, integrating that with Postgres

32:20

row-level security, which was quite a bit

32:22

of work, but we ended up getting

32:24

there. And then the end, that one,

32:26

we actually had to fork, because, yeah,

32:29

it diverged so much that we couldn't upstream

32:31

some of the stuff. So,

32:34

yeah, storage, we have a storage

32:36

engine for large files. That

32:39

one as well, we had to create ourselves.

32:41

It just sits on top of S3. So

32:44

it's a very simple ingestion engine.

32:48

It can stream, it can do

32:50

transformations, things like that. But we

32:52

do integrate with image proxy and

32:54

support and upstream a lot of

32:57

tooling there to integrate with Postgres.

32:59

So doing things like distributed locking.

33:02

So that's been a lot of fun working

33:04

with the Evil Martians team on

33:06

that side. Yeah, oh, they're great, aren't they?

33:08

Yeah, I'm a huge fan of their blog

33:11

in general. And for

33:14

the S3 thing, for self-hosters, you

33:16

just give the

33:19

config a S3 bucket or a

33:21

location, and it Superbase does the

33:23

rest. Is that the scoop? Yeah,

33:26

so on that one, so

33:28

our self-hosting, if you want the entire stack,

33:30

is just a Docker compose. So you just

33:32

put Docker composer. And inside

33:34

the Docker compose, you put your

33:36

S3 access key and secret. And

33:39

so that will route it to

33:41

your bucket. You can

33:43

actually just, each one of the components,

33:45

because of the way we've designed it,

33:47

can also be self-hosted individually. So you

33:49

could take the storage engine, for example,

33:52

chuck it on fly and then use it

33:54

with Tigris or something like that or wherever

33:56

you want. I don't

33:58

see a lot of that, funnily enough. I thought there would be a lot more. How

36:00

do you make this into

36:02

a viable business so people aren't just taking

36:04

your code and running with it? Yeah, it's

36:07

actually one of the hardest things

36:09

of open source. What I've realized

36:11

is no matter how open

36:13

source you are, you can never be open

36:15

source enough for some people. So I

36:18

have this blog post in my brain, which

36:22

I've been meaning to post. And

36:24

there's several spectrums upon which

36:26

they'll judge you. So the first is like

36:28

your license, how open source

36:31

is your license. And so we actually

36:33

check all those boxes because we have

36:35

only MIT Postgres and Apache license. The

36:38

second is how self-hostable

36:40

you are. It's actually

36:43

unrelated to open source by the

36:45

technical definition, but people will judge

36:47

you by that. Then

36:50

the third is actually are you

36:52

a commercial entity? And

36:54

if you make money, people

36:56

just not think you're open source enough.

36:58

That's what I've come to realize. And

37:01

so it's tricky. Like I

37:03

just that third one, I think, is

37:05

unfair. And I just categorically remove them

37:07

from my brain now, if I'm talking

37:09

to those people. I know that we

37:12

can't satisfy them, which I think they

37:14

do such a disservice to the open

37:16

source community because, of course, then open

37:19

source maintainers feel bad asking for money.

37:22

And of course, then they stop maintaining

37:24

things, which I

37:26

wish that all the money in

37:28

the world poured back into open

37:31

source that would be the ideal

37:33

situation in society. Yeah, yeah,

37:35

we do a big open source donation

37:37

here where Chad from

37:39

Century wrote an app that scans

37:41

our repo. So even the syntax

37:44

repo and then donates an allocated

37:46

amount to all the projects on a

37:48

consistent basis. And it's like it was

37:51

so simple for him to set up. It's a small

37:53

allotment. We're getting so much benefit out of these projects.

37:56

Like it would be trivial for

37:58

many companies to set this up yet. They're

38:00

more satisfied with you know, just taking the

38:03

code and running really. Yeah. Yeah,

38:06

I think on the self

38:08

hosting side I Get

38:10

this question a lot like I do you hobble it

38:12

so that people have to use your cloud hosted to

38:14

this The answer is no not

38:16

at all Like we don't like

38:18

intentionally try to do anything but we have

38:21

a lot of work to do Even

38:23

on our platform, right and these people are kind

38:26

of paying for it. They're the ones who will

38:28

jump up and down rightfully Like

38:30

if things are not working, they're like, I'm

38:32

paying you $25 a month. Can

38:35

you fix this thing? Versus the

38:37

ones who are saying, you know, I'm paying you

38:39

nothing. Can you fix this thing? You

38:41

of course have to focus on

38:43

the person who's giving you money I

38:46

mean, that's just out of fairness to

38:48

them so we try like it's very

38:50

hard to balance these two because I

38:53

quite obviously I hope Demonstrate

38:55

to the world that we are open source

38:57

and focused on open source, but it doesn't

38:59

mean that we can spend all of our

39:01

time focusing on on Customers

39:04

who who wanted to self host is

39:07

just not the way that we can

39:09

operate a good business Maybe as we grow

39:11

and we make more revenue We will be able to

39:13

allocate a lot more of that revenue towards that. Yeah,

39:15

but we're not at that stage yet Yeah But

39:19

you know it you mentioned that there

39:21

you can never be open source enough and

39:23

that is true You could be as

39:26

open source as possible and there's still going to

39:28

be someone on the internet complaining No

39:31

matter what so, you know, I think

39:33

like you mentioned optimizing for the paid customers

39:36

is great giving the open source options Self-hosting

39:38

options. Those are all great. So it feels

39:40

very like I don't know it feels very

39:42

open and transparent So

39:44

it would be surprising that anybody

39:46

in turn of I can imagine someone's gonna leave

39:48

a comment on this like oh This is a

39:50

great ad for super bass. It's like, okay Somebody

39:52

did that with our pocket bass episode. We talked

39:55

about pocket bass for like 20 minutes Somebody's like

39:57

this is an ad and like it's not an ad.

39:59

What are you talking about? I mean, the

40:07

good thing is they don't have a hosted option.

40:11

They can check all of those boxes. They

40:15

truly are an open source product.

40:20

If they bring out a host option, I

40:25

think it's fair. It

40:28

keeps you focused on your community. At the

40:31

end of the day, we also want to

40:33

serve them. People jumping up

40:35

and down gets our attention. That's

40:39

the truth. We want to solve

40:41

all those problems. If we had infinite resources,

40:44

we really would. It does

40:46

serve a purpose when people complain out

40:49

loud. It's one of those things that

40:51

we learned to manage, especially

40:53

me personally. They

40:56

would make personal attacks and things say

40:58

that we're devious or whatever. There

41:01

was a kernel of truth in what they were

41:03

saying. As

41:06

you said, CJ, maybe you used it when in 2020, we

41:10

didn't have the dashboard open source yet. It's

41:13

literally because I was too afraid to open

41:15

source for security reasons. It

41:18

had billing code and things like that. I

41:21

think the way we turned it to Next.js and turned it into

41:23

a turbo repo was

41:27

actually quite an exercise to get it

41:29

into the open source. We

41:32

spent a good couple of months, but

41:34

largely because of a personal attack. Now,

41:37

I've become a lot better at, I

41:39

think as you said, Scott, you

41:42

just need to be clear, well, here's the contract that we

41:44

have with you. I

41:48

think that's a good commitment to you in the open source.

41:52

Not much more. We

41:54

just need to make sure it's very clear and we're

41:56

not trying to pull the wool

41:58

over anyone's eyes open source

42:00

we are. Yeah, it seems like

42:02

you know you guys have definitely popular popularize

42:05

the the whole launch week

42:07

thing. I think you were really even some of the

42:09

first ones that I saw do that if

42:11

not the first and it seems

42:13

like you guys just launch a lot. I

42:15

know that goes along with your incremental updates

42:17

but what is the the

42:19

future of super base look like like

42:21

what's the next step here because it's

42:24

a really well developed product.

42:26

It does a

42:28

lot of things like what

42:31

else can you even do here? You're asking

42:33

for me to reveal our next

42:35

launch week is that oh

42:38

no it does not have to be anything

42:41

non-public it can be anything yeah.

42:43

No actually like we haven't really

42:45

got into the bottom

42:48

of the laundry list on almost all of

42:50

the products so like Auth still has a

42:52

few like really key features, storage has a

42:55

lot of key features like each one the

42:58

way that launch weeks work actually is

43:00

kind of more internal at this stage

43:02

where we need to organize

43:04

the team. We're very asynchronous so we don't have

43:06

meetings or anything like that so we just say

43:08

well here's the date of the next launch week

43:11

tell us what your like big project you're going

43:13

to ship and they come and tell us

43:15

oh this is what we're going to ship. So

43:18

yeah I actually don't even know what is

43:20

in launch week I have some ideas of

43:23

some things that are kind of new and

43:25

exciting but you know the

43:27

the next stages are like

43:29

are really around scale and types of

43:31

data that people want to store people

43:34

want to store a lot of their

43:37

like people are storing logs

43:39

inside their Postgres database which is just not

43:41

a good idea they shouldn't be doing that

43:43

so we need to solve that but

43:46

we do want them to store their data with

43:48

us and their logs and all that sort of

43:50

data so like you know we've got enough problems

43:52

and as well people aren't

43:54

using branching well enough they're not using

43:56

the migrations well enough

44:00

We want better integrations with the one, the

44:02

ORMs, the drizzles,

44:04

and things like this. So

44:06

there's a ton of stuff for us

44:08

to still solve in the database space.

44:12

Speaking of branching and stuff like that, congrats on the

44:14

GA. I know it was like two months ago at

44:16

this point, but if you didn't know,

44:18

Superbase was in beta for like four years until

44:20

two months ago. Can you talk

44:22

about maybe some of those features? Because I didn't even

44:24

realize you all had branching now, so that's super cool.

44:27

And maybe what are some of the other recent

44:29

things that you launched that people should be excited

44:31

about? That's a good question. Now that we've got

44:33

to, let me see

44:35

what we did for GA week. I

44:39

will say, I'm also looking at a blog post

44:41

on Superbase. It's called Top 10 Launches from Superbase

44:43

GA Week. So that's definitely something to check out.

44:46

What is our top? I even wrote that

44:48

and I can't remember all that. Oh, yeah.

44:52

I love it. There's a bit of something for everyone,

44:54

yeah. Yeah, and I hadn't

44:57

used that one yet either. There's

44:59

so much new stuff that I need to explore in

45:01

Superbase, but the first one is Bootstrap. So you literally

45:03

have a CLI that'll create a new project. That's pretty

45:05

sweet. Yeah, that one's actually cool.

45:07

Like we're going to do a bunch

45:09

of these templates to like launch a

45:11

project very fast. So you can Bootstrap

45:13

with Next.js or Svelte

45:16

or whatever it is. Yeah,

45:18

as you said. Actually, yeah, these

45:20

ones were very cool. The Security,

45:22

Performance and Index Advisor. This one's

45:25

cool because I talked

45:27

about you can use anything with Superbase.

45:29

This is one of those tools that like really

45:32

fits into that category

45:35

in that like I

45:37

saw someone was like, when

45:39

we released our Index Advisor, it actually

45:42

gives you this hypothetical index. If you

45:44

add this index, it will

45:46

reduce the latency on your

45:48

query by 90% or

45:51

something like that. And so I

45:53

literally saw like one day after we released

45:55

it, someone said, oh yeah, I turned this

45:57

on and my query is 90%.

46:00

So it's like, in the media when, yeah. And

46:05

that's cool because security as well

46:07

on databases is

46:09

incredibly hard. And we

46:11

lean into low-level security, which can be a

46:13

bit niche. So, you know, we

46:15

want to build out a lot of this advisory stuff,

46:19

which you don't think about when you're, like, you don't

46:21

want to have to learn, if

46:23

you think about a development life cycle,

46:25

you want to get started really fast.

46:28

And then you want information exposed to you

46:30

only at certain points. You don't want to

46:32

have to be thinking about, is it secure,

46:34

is it fast? You just want to

46:36

be thinking, does it work? Then

46:38

after you get it working, you want to

46:40

think, oh, is it safe? Is

46:42

it secure? Is it fast? And that's

46:45

where the advisor is kind of the

46:47

right information at the right time. After

46:49

you've done prototyping, even on your local

46:51

machine, you can then go

46:53

to the advisor, have a scan

46:55

through, like, all the, the

46:58

recommendations and, like, enable this index or

47:00

do that. And so

47:02

it's kind of like a nice workflow

47:04

pattern where you focus on building and

47:06

then focus on the secondary steps. Yeah.

47:09

And that works, that tracks too with

47:11

what you were saying about, maybe even

47:13

initial, some of those first customers being

47:15

JAMstack or even primarily non-back-end developers, right?

47:17

Like, those people are the prime candidate

47:19

for somebody who's not going to write

47:22

an index when they should. And they're

47:24

going to see potentially slow queries because

47:26

of it. So having any kind of

47:28

advisor there is a great way to

47:30

not only, like, help them, but to,

47:32

like, educate them. You know, I

47:34

think that's kind of like a step most people have to go through

47:37

when they're learning about performing database

47:39

queries or any of that stuff. One

47:42

thing on the advisors, I think, like, credit

47:45

really goes to PlanetScale on this one because I

47:47

think they do a phenomenal job on

47:50

database performance and really exposing,

47:52

like, for a developer how

47:54

to make the queries

47:56

faster. So, yeah, I think, like, we

47:59

look up to them. in that respect

48:01

in terms of building a really good

48:03

product for performance. That's cool to

48:05

hear. We host on planet scale and have

48:07

used their branching, have used their indexing

48:10

features like that. It's cool to

48:12

hear. It

48:15

sounds to me like you could even take your Postgres database to Superbase and

48:17

all of a sudden get a bunch of really good info, even

48:20

if you're not using the rest of the product. Just all of a sudden

48:22

you have a bunch of insight into your database, which is pretty cool. One

48:25

last lighthearted question for you. The Superbase

48:27

account on X is really

48:31

good at posting memes and stuff

48:33

like that. You mentioned at the

48:35

start of the interview that you like memes. Do

48:37

you run that account or is it like you

48:40

have a Slack where people like workshop meme ideas?

48:44

That's my co-founder. Who's

48:46

funny enough is our CTO who spends

48:48

a lot of time on our Twitter.

48:52

We also have an internal memes channel,

48:54

which is great for hiring around. We

48:57

literally sometimes hire people because the

49:01

reason why they get ahead is because they're good

49:03

at memes on Twitter. I

49:07

think our strategy there, no one wants

49:09

to see a boring corporate Twitter account.

49:12

We actually had never used Twitter

49:14

before this really. I think we

49:16

had accounts but had never really

49:19

done developer marketing. But

49:21

the one thing that we did think is no

49:23

one just wants to read about Superbase, this product

49:25

that they've never heard of. They

49:28

just want to hear about shipping cool shit and

49:30

doing memes. We were already sending

49:32

memes to each other and he just started

49:34

posting them on Twitter. It's

49:37

worked very well for us. If

49:39

you can do it right, it works. Sometimes

49:41

for certain companies it'll come off as cringe, but Superbase

49:44

is doing it right for sure. I always have a

49:46

chuckle every morning. That's the other

49:48

funny thing. We realized that we

49:50

actually realized this by trying

49:53

to hire someone to do this for us

49:55

instead of Ant doing it. We

49:57

realized that actually it's very hard

49:59

to do. to find someone who can do it right.

50:01

And so now we're

50:03

like, well, and that's clearly the only thing

50:06

you should be focused on. And

50:10

other companies kind of just can't pull it

50:12

off. I see sometimes like our competitors have

50:14

clearly looked at our Twitter and they're like,

50:16

oh, let's do what they're doing. They've got

50:18

a lot of followers now. We

50:20

need to do that. And I look at what they've done

50:23

and it's just very cringe, just

50:25

really off the mark. You kind of get

50:27

memes or you don't, it seems. So yeah,

50:30

a lot of them, a lot of them don't,

50:33

but yeah, get a good

50:35

chief meme officer on your on your team

50:37

and you're good. Yeah. All

50:39

right. So now is the part of the

50:41

show where we talk about sick pics and shameless plugs.

50:43

Paul, do you have a sick pic for us today?

50:46

Yeah. So I just arrived in

50:49

the US two weeks ago and

50:51

I bought to

50:53

try it out the like an Apple

50:56

Vision Pro. So I've been

50:58

testing it because I fly a

51:00

lot. I travel a lot. So the idea is then I

51:02

can work on a plane, I work

51:04

in a hotel room. So I've

51:06

been trying it out. It's so far. I think

51:08

I'm still in the honeymoon phase of

51:11

this. I don't know whether it's going

51:13

to be a regular feature in my life, but

51:15

I'm very impressed with the

51:17

technical aspects of the

51:20

Vision Pro software. Yeah. So I

51:22

think it's a bit to be not many things

51:24

to download on it at this stage, but

51:26

yeah, for a usefulness as a

51:29

as a like monitor, second monitor

51:31

or multiple monitors has been good so

51:33

far. Yeah, I have one

51:36

too. I use mine a lot

51:38

like if I'm working outside, like I'm working

51:40

at a like a table outside.

51:42

I no longer am constrained to my little laptop. Blow

51:44

it up. There's an app

51:46

that if you don't have yet split

51:48

screen, it allows you to do multiple

51:51

monitors so you could have several monitors.

51:54

And so that that's a must have for me. I use that

51:56

one a lot. But I agree

51:58

there's not a ton ton to use right

52:00

now. experience is really pretty incredible, huh? Yeah,

52:03

yeah, it's crazy. Like I

52:05

actually, in

52:08

another startup life, worked on like kind

52:10

of Metaverse and AR, VR, and

52:13

now like went deep on a lot of

52:15

the things that could be solved. And

52:18

I thought, well, the technology, the hardware, it just

52:20

isn't going to be there until like 2026, 2028,

52:22

10 years ahead. And

52:27

what they've delivered now is kind of just

52:29

pulled four and five years. It's crazy. Yeah,

52:32

it's crazy. Yeah, it feels like the software

52:34

is what's lagging, which is the

52:37

crazy bit. Yeah. Yeah.

52:40

And shameless plugs, is there anything you would like to

52:42

plug? Yeah, I mean,

52:44

of course, like if if

52:47

people want to use Superbase, I would love them to

52:49

use it. But I guess the real plug, from

52:52

my point of view, goes to Postgres.

52:55

I think it's just such a good

52:57

tool. It's really becoming better

52:59

all the time. It's one of

53:01

the most unique open source communities

53:03

that I've ever seen. And I can see

53:05

it becoming kind of just

53:07

a staple to the world. So I

53:10

think it would be hard to plug Superbase

53:13

without just directly giving a lot

53:15

of the credit to Postgres itself.

53:18

Nice. Awesome. Well, Paul, it's been amazing

53:21

having you on. It's

53:23

really great to learn about internals,

53:25

tools, all that sort of stuff. And, man,

53:27

I can't wait to see where Superbase goes

53:29

from here. So thank you

53:31

so much. Thank you. Peace. Peace.