0:00

I need need to extend my thanks to both

0:02

you and Brent for covering in my

0:04

absence last week. I had one of those

0:06

annoying one of those annoying viruses that I just lost

0:08

my voice and just felt crummy all

0:10

week. So to you you to you guys for

0:12

stepping up. maybe thought maybe he had sabotaged

0:14

you over there because it sort of was It sort

0:16

of was perfect timing with his new

0:18

hardware setup that all arrived. Seemed kind of of coincidental. well

0:20

we did hear well, we did hear from a

0:22

few people right into the show that said

0:24

now they're going to use the the to build

0:26

their own own and stuff, right? right? Yeah, you know, he's

0:28

got a pretty good you know, he's over a pretty good little

0:30

plan over there, though. I like how much thought

0:32

he's put into it. know, this is something he's been

0:34

you know, you know, he's been slowly working towards this

0:36

as a goal for a while, and it's coming together.

0:38

and it's Now we have a stacked

0:40

show to get into today, so I'm not

0:43

gonna waste too much time with the to

0:45

waste too much time with the pithapatur at the We've got We've got

0:47

the founder of of home on the show

0:49

today. show today. Yeah, Yeah, he's joined us

0:51

because the Home Assistant Assistant Voice it's actually

0:53

here, it's hardware, it's available, and we've

0:55

had our hot little hands on it. our

0:58

hot had so much to ask him about. and we

1:00

had so much to ask him about. Hey

1:02

Home Assistant onto the show,

1:04

please. Home Assistant onto the Welcome back. Hello,

1:06

Paulus, welcome back. to be here! be

1:09

here. It's your third visit to the

1:11

show, and you're here for a

1:13

special reason, this visit, visit, because there's

1:15

a new product. And it's pretty exciting

1:17

because it's not just a new product, but it's kind of

1:19

like a new era. but it's it's

1:21

definitely a new era.

1:23

it's We have launched era.

1:25

December launched last December 19 Home Assistance

1:28

Voice preview where where we've been

1:30

tinkering on voice for two years,

1:32

making small improvements left to right. left to

1:34

Now we have hardware that makes it all

1:36

it all super accessible in your house

1:38

And you can actually use it. it.

1:41

This has been something that I've been tinkering with for a

1:43

little bit. So I've had a chance to get some experience

1:45

with it. And one of the things that I thought. experience

1:47

with it. And most delightful was

1:49

the actual I thought was the of

1:51

the design was the actual just small little

1:53

square with a USB -C

1:55

port, little jack. port, a then

1:57

an almost iPod like iPod-like...

2:00

wheel. You noticed it. Yeah. Can you talk about you talk

2:02

about that? Yeah, we is inspired by the inspired

2:04

by this nice iPod. It had this nice

2:06

wheel you go to would feel it click

2:08

when you go to the next step. want

2:10

this. And so were like, we want this.

2:12

And so we were looking and then

2:14

we found this rotary encoder. And so

2:17

it's not capacitive touch because capacitive touch

2:19

is just touch is just nah. Like you feel you feel every

2:21

click. Like it's not a click wheel wheel.

2:23

there's a button in the middle, but

2:25

just, you know, the rotation feels so

2:27

good. good. Yeah I was like, oh, this.

2:29

this brings me back. so can

2:31

you It really does. a little bit about you tell us

2:34

just a little bit about some of the

2:36

changes that are happening at a software level in

2:38

Home Assistant to make something like this work?

2:40

Because when I got it out of the box.

2:42

immediately It was immediately discovered by the connected

2:44

have connected to my I also got I also got

2:46

like a prompt on my phone, can you

2:48

just talk a little bit about that? that?

2:50

You know, You know, I'm the president of

2:52

the Open Home but we

2:54

don't just make don't we make home

2:56

Home, and ESP Home, that is

2:59

the open that software to create your

3:01

own devices for the smart home.

3:03

the smart home and... our voice assistant is

3:05

based on on ESP Home. Now we've

3:07

been working on making ESP

3:09

Home easier to use and get it

3:12

to get to get so one

3:14

of these things that we've

3:16

developed as part of that

3:18

that we've this thing called of that

3:20

is this developed ourselves called open standard

3:23

that we've developed ourselves called on devices

3:25

because So it's on devices that

3:27

are generally devices that are generally the old pre- I

3:29

know, couple of years ago, ago, you can DIY

3:31

device, it has to, it a Wi -Fi access point,

3:33

you point, you -Fi access point, you set up

3:35

your Wi -Fi credentials, you hit your Wi-Fi It

3:37

has to take down your Wi -Fi access

3:39

point to connect to the Wi -Fi, so

3:41

you don't get any feedback. You don't actually

3:43

know if the device was connected. And we

3:45

And we were like, that's super annoying. And

3:48

And every device you buy off the shelves,

3:50

they have like a Bluetooth onboarding. So So we

3:52

a Bluetooth standard. standard. We added it to

3:54

Home Assistant, it's of the Bluetooth the Bluetooth proxies,

3:56

you Assistant will find it, find it's

3:58

part of their of Assistant system apps. So even

4:00

if you don't have a Bluetooth

4:02

have a Bluetooth the Home the will help

4:04

you and up the device via

4:06

Bluetooth. up the device this open standard is

4:08

actually built into actually built So ESB Home,

4:10

device that is that is like for

4:13

ESP ESB have have improv over Beely on, so that

4:15

that like the open home the open home

4:17

community, we can know, we can build

4:19

devices for one another and just

4:21

have a great onboarding experience. But

4:23

we actually went a step further

4:25

in this case for case we know

4:27

this with voice in the the years, two years.

4:30

There's many different ways to create a

4:32

voice assistant. it's when I say stack, right?

4:34

it's actually about the stack, text, we

4:36

speech to text. We have to turn

4:38

your speech into process we have to

4:40

process a command, if that command, command, that act

4:42

on that on that command. we Then we

4:44

text text response. And the text response

4:46

has to be has to be into speech

4:48

that we can play back on

4:50

a device, on your phone, or wherever

4:52

you are. are. Now, we realize that

4:54

you do you do this locally,

4:56

self and that's I'm on this podcast,

4:58

we we a a whole lot

5:00

of that about that right at We want

5:02

stuff to be local be it's

5:05

possible. it's And so we

5:07

actually open a wizard now. now. That

5:09

visit will guide you saying, okay,

5:11

if you have homelessness Homocyst and

5:13

Cloud is the way people

5:15

can support the Homocyst and

5:17

support the actually has built that actually

5:19

has -speech services that are and

5:21

but or super services that are super

5:23

accurate, super fast, high and support

5:25

super languages and dialects. 130 But

5:28

if you don't have dialects. we will actually guide

5:30

you through in cloud, we will

5:32

and installing you through Whisper is

5:34

open your eyes, is open AI's speech to

5:36

text. Piper is is our speech

5:38

-to -speech system we've built ourselves

5:40

and basically at the end you

5:42

have a working working a

5:44

working voices that can work local.

5:47

Now we ran kind of into kind

5:49

of into issues with is a Whisper

5:51

is model, not a large not a

5:53

large language model. It's a a AI

5:55

system that is. right? So we realize

5:57

that the realize that

5:59

in green... raspberry pie, that's not the

6:02

best to run it on. So we

6:04

recommend an Intel 100 and 100, like

6:06

the one that Brian the other day

6:08

with the DIY NAS guide, like recommended,

6:10

right? The most efficient Intel chip nowadays.

6:13

Sure. That one will run whisper base

6:15

model just fine and you can get

6:17

like for English, right? Like I think

6:19

that one works quite well. The cloud

6:22

is still faster and better, but of

6:24

course. you know there's a that runs

6:26

in the cloud doesn't run at home

6:28

this is home assistant after all right

6:31

like it's got to be local first

6:33

and I'm curious you added some Bluetooth

6:35

stuff to the the stack over the

6:37

last year or two with the Bluetooth

6:39

proxies and what have you There must

6:42

be other stuff you've been secretly adding

6:44

to Home Assistant with future hardware plans

6:46

in mind, right? No, I mean the

6:48

voice stuff, like all these stuff, for

6:51

example, ESP Home, we added over-the-air updates

6:53

where an ESP Home device can update

6:55

itself without you having to compile it

6:57

on the ESP Home dashboard, which allowed

6:59

us to, for example, sell a voice

7:02

assistant or Lewis to sell his everything,

7:04

presence one sensor, and people don't need

7:06

the ESP Home dashboard to... or East

7:08

home device builder as it's been renamed

7:11

to actually update the device. There's just

7:13

an update entity in Home Assistant that,

7:15

oh, boom, one click, it gets updated.

7:17

I noticed that I get my updates

7:20

for this little park thing just in

7:22

the little notifications bar. So hey, Home

7:24

Assistant Voice needs an update, press go.

7:26

Isn't it nice? Really nice. Yeah, great

7:28

job on that. No, we really looked

7:31

at all the... parts of a device

7:33

of the cell that you buy. What

7:35

would be the ultimate experience? And we

7:37

just made sure ESB Home can deliver

7:40

that experience. Yeah, it feels like a

7:42

really consumer-friendly ready for that next tranche

7:44

of users that might be looking for

7:46

something like this. I mean, it works

7:49

great for us. Yeah. And it has

7:51

a lot of perks too, because it's

7:53

a media endpoint now as well. And

7:55

you can attach a speaker to... So

7:57

the speaker in our device is really

8:00

meant just for voice responses, not for

8:02

music, but we did add an oux

8:04

jack on it so you can connect

8:06

it to an active speaker. But what's

8:09

actually really cool is that in this

8:11

device we have an ex-mos chip. And

8:13

ex-mos is his 20-year-old British company that

8:15

have been doing voice processing forever. Like

8:17

that's their deal. They make speaker phones,

8:20

and a speaker phone just wants to

8:22

capture voice, nothing else. Now, by putting

8:24

this chip in there, they clean up

8:26

the audio for us, but they also

8:29

have this thing where any audio that

8:31

we are playing through either the internal

8:33

speaker or the outjack is removed from

8:35

the input stream. So we can play

8:38

music very loud. It doesn't get on

8:40

the input stream, and we can hear

8:42

the wake word throughout the room. And

8:44

then, because we are playing the audio

8:46

through our voice preview edition, we will

8:49

lower the volume so we can hear

8:51

you speak. And then at the end

8:53

when you hear your response will bring

8:55

the volume of the music back up.

8:58

So I wanted to ask you, why

9:00

not a device with a screen? Because

9:02

there have been some ESP devices, and

9:04

I have one actually here in the

9:06

studio, that works with voice, that does

9:09

have a screen on it. Was that

9:11

considered for a bit? And then this

9:13

was decided, was a better route? So

9:15

we were playing with the ESP 32

9:18

S3 box as a development target for

9:20

a while and it really also forced

9:22

our team to think about user interface

9:24

and see what we can do and

9:27

we had like these nice pictures that

9:29

would change if it was like responding

9:31

or thinking but we realized that what

9:33

you really want from a voice assistant

9:35

is for example when you ask it

9:38

about your shopping list or the weather

9:40

you want to see that on the

9:42

screen and to do that you kind

9:44

of need like a bigger screen and

9:46

as we were discussing this we realize

9:49

okay this is going to blow up

9:51

the scope like we want to ship

9:53

some hardware like this year right and

9:55

not like next year so yeah that

9:57

but that's also it's called homelessness preview

10:00

edition definitely an upgrade of

10:02

all the previous ways you

10:04

could experience homelessness and voice We

10:06

don't believe yet that it

10:08

is for everyone for with for

10:10

everyone. mean, I mean,

10:12

you want. to remove all your

10:14

remove all your minis and have

10:17

the same have the same

10:19

experience where you can cast your music

10:21

to it, where you can ask

10:23

any question in the world from it,

10:25

from it, not there yet, yet, right? But you

10:27

you use Homocystin today. wanna follow follow

10:29

if you if you want using like self -hosted

10:31

stuff, see what's all is possible, what's all

10:33

as device is perfect. this I use

10:36

it is day, my kids use it

10:38

every day my the middle button is

10:40

like the talk, right? So you don't

10:42

have to say the wake push to talk right

10:44

so you talking to have to all day.

10:46

and they're So at what point

10:48

do we end up just literally

10:50

chatting with our house through

10:52

an just that's all local? our house through

10:54

an LLLM that's all possible today. So

10:56

we have so we have Olamma support and home

10:58

So the whole So the experience has

11:00

supported AI and AI can

11:02

call tools AI called. So call can

11:05

control it's called. So We've had that

11:07

working for the last six

11:09

months and we've been iterating on

11:11

it and improving it. So

11:13

we have support for six months. We've been

11:15

way of running and language

11:17

models. So we have works with Olama is

11:19

a local the names large language the...

11:21

Too many And Yeah, Too too

11:23

many llamas. So with Lama. works

11:26

with Lama, works with with Kwan,

11:28

a and all these things.

11:30

you need powerful hardware. Like,

11:32

this is not for the Raspberry Pi crew,

11:34

right? they need they need to sit this

11:36

one out. But if you have powerful hardware,

11:39

you can run quite some good AI

11:41

at home, and then you can start talking

11:43

to it to your house. and it

11:45

has access to everything that you have in

11:47

assistance. So we give it give it the state of

11:49

your house, of all the devices that

11:51

it can control. So you can also start

11:53

asking things like, oh, which oh, which doors are

11:55

open, or these kind of things. things. The API

11:57

The API is designed in a way

11:59

that people could... build their own custom components or

12:01

custom integrations that add more tools to

12:04

the AI. So there was one person

12:06

was playing with adding Google search to

12:08

it. So now the AI can actually

12:10

search the web for you and get

12:13

way more rich information. We also allow

12:15

people to just use scripts to inject

12:17

more data, right? So you now to

12:19

allow the AI to query your calendar

12:22

and list your events for the day

12:24

or these kind of things. It's something

12:26

we're super interested in, but it's also

12:28

something where. We know only part of

12:31

our audience could basically do something with

12:33

this, even if they have the most

12:35

powerful hardware. So we're not chasing it,

12:37

but there's, our community is just so

12:40

big that there's enough people working on

12:42

it, that we're making meaningful progress all

12:44

the time. It seems like there's a

12:46

good and a bad side to where

12:49

Home Assistance is at with voice control

12:51

right now. And I'd like to just

12:53

ask you about some of your future

12:55

plans here. I'll put it in context.

12:57

So one of the things I love

13:00

is that I can go in and

13:02

I can set aliases for just about

13:04

anything. And so then we can have

13:06

really familiar voice prompts to get the

13:09

temperature of a room or change on,

13:11

turn lights on or things like that.

13:13

I like that a lot. But it

13:15

really only works when I kind of

13:18

go in and I curate all the

13:20

names for everything. And I kind of

13:22

have to change it from the defaults.

13:24

A lot of things get named, just

13:27

kind of these long random names in

13:29

homelessness systems. What are your thoughts around

13:31

future changes there about maybe the default

13:33

names of devices or just kind of

13:36

smoothing that whole thing over in general?

13:38

So, well, one, the AI is actually

13:40

the answer, right? You can do a

13:42

lot less. organizing as us as a

13:45

human of your house because an AI

13:47

can understand that a light entity in

13:49

the living room, you can refer to

13:51

it as living room lights even if

13:53

you don't have, you know, it renamed

13:56

to be that such. But I think

13:58

in general what we want to do

14:00

is that we want to work on

14:02

our information architecture, so you bring a

14:05

device into a home assistant. We don't

14:07

know. So now we want to create

14:09

a dashboard. Now we want to offer

14:11

you an environmental state. If you have

14:14

three temperature sensors in your living room,

14:16

homelessness doesn't know if these are ambient

14:18

temperature or if this is your coffee

14:20

machine or is this the nozzle of

14:23

your 3D printer? Like we don't know.

14:25

So now we want to create a

14:27

dashboard. Now we want to offer you

14:29

an environmental state of your house. Now

14:32

we want to have AI know about

14:34

this. Currently we don't know. So information

14:36

architecture is very high on our list

14:38

of things we want to tackle and

14:41

together with information architecture, better naming, that's

14:43

also there. So if we know that

14:45

it's a decorative light for an area,

14:47

we can already provide a lot better

14:50

default names. unrate.net/self-hosted. Go and leash your

14:52

hardware and check out Unraid 7. The

14:54

release candidate is here and it is

14:56

packed full of game-changing features. and folds

14:58

ZFS capabilities. Unraid is a powerful and

15:01

easy to use operating system for any

15:03

self-host or home lab or out there.

15:05

If you've got some disks, you've been

15:07

meaning to set up a server, maybe

15:10

have a little network attached storage. This

15:12

is the time. Go make the most

15:14

out of your hardware. No matter what

15:16

you have on hand. Different size drives

15:19

or manufacturers, that's not a problem with

15:21

unraid.com. Unraid now has integrated tail scale

15:23

support. How great is that? They've got

15:25

a great video that shows you how

15:28

to use it. It's such a nice

15:30

thing to have on your server like

15:32

this. And you know when Unraid does

15:34

it, they do it right. So, they've

15:37

gone full integration. When you install it,

15:39

almost any Docker container gets the ability

15:41

to connect to your tailnet. You can

15:43

access them using valid HDBS certificates and

15:46

give them alternate routes to the internet

15:48

via exit nodes. You get to integrate

15:50

tail scale into Unraid. It's like your

15:52

peanut butter and your... jelly coming

15:55

together. Unrate is

15:57

flexible, it's easy to

15:59

use, it has

16:01

a brilliant UI, and

16:03

and 7 makes it

16:06

even better. Go try it

16:08

out try it out and finally take advantage of

16:10

that hardware you've had sitting around. had worrying

16:12

about it and just

16:15

start building it. just start slash

16:17

it. unraid.net slash self-hosted. So

16:20

2024 was the year of the voice.

16:22

I assume I assume you're gonna come

16:24

up with a moniker for a moniker for 2025.

16:26

So our year of the

16:28

voice our year of the weren't done, so we

16:30

just kept done. So we just kept going. Now know.

16:32

you just kept pushing on the voice.

16:35

on the we did more things. I

16:37

think. things. I Our branding of of Voice

16:39

was so successful successful that came

16:41

around, around, know, months ago, people people

16:43

were like... they acted like we didn't

16:45

do anything but voice even even though

16:47

we had made a lot

16:49

of progress on automations and dashboards

16:52

and these kind of things. of

16:54

So things so April, we actually

16:56

launched a roadmap and a roadmap broken down per

16:58

category down per we've been for example, we've

17:00

been revamping our there's now this new this

17:02

new sections dashboard type that is, it's

17:04

not the default yet. That's the

17:06

final step we still have to tackle

17:08

kind kind of relies on the information

17:10

in our just touched. touched but If the, if

17:12

you create a new sections dashboard

17:14

you can drag and drag you can,

17:16

it's very intuitive to build a beautiful

17:18

dashboard through just the user interface. a

17:20

user interface. been re-simplifying our and

17:22

automation editor. We've been

17:25

getting all been getting all

17:27

certified, which that's a

17:29

project that's going to a

17:31

project that's going to get into 2025. Yeah,

17:33

so there's a a lot of progress

17:35

happening all over the place. think I

17:37

think for 2025 right now, we're doing our road

17:40

map annually at April. in in April,

17:42

we're going to announce our next

17:44

roadmap. We did just update like a publish an

17:46

update on, like, know, you know, six

17:48

months in. six months it's

17:50

going pretty well like like the we

17:52

wanted to to do and the

17:54

things we're things we're doing of of lining

17:56

up, which is very nice. nice. Like I said,

17:59

Like I said, information is - is important doing a

18:01

bit more AI stuff like stuff, like one

18:03

of the things people love doing

18:05

like camera pictures of like camera pictures

18:07

of doorbells to the AI and ask of

18:09

in front of my door that should just that

18:11

that should just be easy. It should just

18:13

be built yeah yeah I love that. Yeah we

18:15

all like something if all a like if you

18:17

have a camera a you have a powerful

18:19

system that just be a toggle in a way,

18:21

right? a way right like sure well there's a there's a

18:23

bunch of stuff the commercial products do you

18:25

know, like the know like the ring example for example that

18:27

the the basics that that of little things

18:29

like having them appear in your Apple TV

18:32

is like a little is like a little -picture window

18:34

that kind of stuff and I think

18:36

all of the I think are there with Home

18:38

Assistant there with connect all these things together all

18:40

these things I don't want to say say lacking

18:42

because that sounds kind of unfair but

18:44

the kind of what's missing is that but

18:46

that last 10 % user experience to kind

18:48

of glue it all together. to kind often it all

18:50

together. Yeah often call this like where

18:52

is is we are we are making a

18:55

toolbox, But people people don't want a toolbox,

18:57

they toolbox, want solutions. They want to

18:59

have. to have. if you look at at

19:01

our energy dashboard, that's a solution. a

19:03

You just select your couple of

19:05

input sensors we we generate a whole

19:07

dashboard with graphs and like how much much energy

19:09

are you using? How much solar are

19:11

generating? and like. you know, whole overviews, and And

19:13

that's something I would love to do

19:15

more of it. it. But that's that's something

19:17

we want to build. we we have

19:19

our default we have our then we can

19:21

start building on top of that. building

19:23

on top of that. So just

19:25

air quality throughout the house. throughout

19:27

the house, humidity, looking at the looking

19:30

at the security dashboard, automatically laying out

19:32

all your cameras and maybe if doors

19:34

are open, course we already have course we already

19:36

have the energy dashboard of these of these

19:38

dashboards that kind of are cut through

19:40

your house right to give They give like a

19:42

focus on one element of your house.

19:44

of your I think Well, I think the automation

19:46

of laid the foundation for some

19:48

of that kind of some of that kind of

19:50

community fed knowledge to kind of be of be

19:52

shared with people, but I think

19:54

there's more that could be done. be

19:57

I agree. I I I agree. mean,

19:59

even on blueprints. Right? Like we should

20:01

be able, you now have to browse

20:03

this forum, I actually want to turn

20:05

it around where we say, you have

20:07

this device, we have these blueprints that

20:10

match this device. Yeah, yeah, yeah, particularly

20:12

if you own the hardware, right? If

20:14

you have the Home Assistant Voice Park,

20:16

then here's all the stuff it can

20:18

do, by the way. Yeah. I think

20:20

Paul's what strikes me is when you

20:23

first started talking about voice, you came

20:25

on the show and we chatted about

20:27

it. working with the community to develop

20:29

some of the voice modeling stuff, but

20:31

of course also with Piper and then

20:34

this device now, which is fully modifiable

20:36

by the end user, there's no way

20:38

you could describe this thing as lockdown

20:40

in any way. And I feel like

20:42

you've kind of struck that balance of

20:45

a device that end users can use,

20:47

a device tinkers can use, and yet

20:49

it's still something... I want to pay

20:51

for, I want to buy, I don't

20:53

want to build myself. That's good. I

20:56

think you guys have nailed this. Thanks,

20:58

thanks. It's really kind of completing the

21:00

loop from the first time we had

21:02

the conversation around voice and to this

21:04

end product, which I've been really enjoying,

21:06

too, with the pickup range. I thought,

21:09

oh, I'm going to buy like, you

21:11

know, a dozen of these. I think

21:13

I'll probably only need like two or

21:15

three. Yeah, no, the pick-up range is

21:17

really good and that's the ex-mos chip

21:20

that besides like cleaning up the audio

21:22

It also automatically applies the gain so

21:24

it will use if you're further away

21:26

It will increase the volume. So it's

21:28

always like a normalized output and that's

21:31

where the which is the level that

21:33

the model is trained on right so

21:35

it's a text so good. So we

21:37

what one thing we want to do

21:39

with this device is that we want

21:42

many different companies to make voice assistants,

21:44

right? Like we don't want to be

21:46

the gatekeeper of voice assistant hardware for

21:48

the home assistant community, for the open

21:50

home community. So we're going to open

21:53

source everything of it. So the schematics,

21:55

the firmware, the firmware that runs on

21:57

the ex-manship, the firmament runs. on

21:59

the ESP chip, everything

22:01

will be open and all the the onboarding that

22:03

we about about at the beginning, all all

22:05

of that will be available for every

22:07

voice assistant product that is gonna be

22:09

hooked up to be hooked up to home assistant.

22:11

That is fantastic. you for you for

22:13

joining us once again. We're gonna

22:16

have have links to to everything in the

22:18

show notes, but happy holidays to you and the

22:20

whole team. This feels like a gift to

22:22

the home a community, so. assistant it is, it is.

22:24

It's pretty awesome. It's It's gonna be It's gonna be

22:26

I think that's a great price. great know, with

22:28

that, I can afford to buy a few of

22:30

them and put them around the house and

22:32

they're not too big. They're pretty and they just

22:34

need not power. They're pretty discreet. They just

22:36

need USBC thank you so much. Paulus,

22:38

thank you so much. tailscale.com slash

22:40

self -hosted. Go try it out for

22:42

free for up to 100 devices

22:45

and and users. Go to Go to.com

22:47

slash self -hosted. That's not a limited

22:49

time deal. time deal. That's the deal.

22:51

That's the the plan. That's on. plan I'm I've

22:53

been on it for a very, very long

22:55

time. very long is the easiest way to connect

22:57

your devices and services to each other. and services

23:00

they are powered by by

23:02

WireCard. Secure remote remote access

23:04

to your databases, to to your

23:06

applications, to to your servers, to your

23:08

Kubernetes cluster, whatever it might be.

23:10

be. And tail is fast, really, really

23:12

fast. It's intuitive, to it's easy

23:14

to use, and ultimately it's unlocking

23:16

programmable network lets you lets you manage

23:18

networking like it is code. You

23:20

can deploy it, you can set

23:22

it up with rules. it's really It's

23:24

really powerful. What you're building is

23:26

a zero trust mesh mesh network that

23:28

connects each device directly to each

23:30

other so you get a flat

23:32

tail net. net. And And if you have

23:35

a a on a on a VPS you have a node

23:37

on your note on you have a mobile device,

23:39

they all exist in this virtual in this And then

23:41

in there, you have all kinds of tools

23:43

and capabilities. and can do name resolution, like you can

23:45

move files between your systems. can You can have

23:47

them talk directly to each other. You can

23:49

host applications inside this tail net. you And that's

23:51

what I've chosen to do. inside this tailnet. And

23:53

that's I run at home. to do.

23:56

Jellyfin, I'm standing up? run at of it.

23:58

I don't new that I'm standing up, all of. putting

24:00

it on the internet anymore. I

24:02

just have to put it on

24:04

my tailnet and I just connect

24:06

my nodes to the tailnet and

24:09

I can get access to everything.

24:11

No inbound ports at all. It

24:13

makes old VPN systems seem, well,

24:15

like legacy systems. It's one of

24:17

those things where once you've experienced

24:19

it, you've gone through a bit

24:21

of a networking awakening and it

24:24

really fundamentally changes the things you

24:26

can do with networking. You don't

24:28

have to. You know, you could

24:30

just use it as a VPN

24:32

if you want, but it's so

24:34

much more than that too. And

24:37

it also ties in with your

24:39

existing authentication infrastructure. So if you're

24:41

a business and you've been thinking

24:43

about giving it a go, you

24:45

can still go to tailscale.com/self-hosted, try

24:47

it out, and then tie it

24:50

in with your overall business authentication

24:52

scheme.com, slash self-hosted. Well,

24:54

I'm sat here in my lovely bonus room

24:56

with a friend of mine, Shane, who

24:58

I've worked with in the past. Welcome to

25:00

the show, Shane. Thank you, Alex, hi.

25:02

I didn't want to introduce you as this,

25:05

but you are the developer of the

25:07

Bento Box app that I mentioned a few

25:09

weeks ago. That's true, I am. But

25:11

you're also a Knicks. Addict, is that a

25:13

fair? I think I'm a recovering Knicks

25:15

addict. Yes, yes, that's what we're going to

25:17

talk about I think in the show

25:19

today, a carefully considered discussion of Knicks on

25:22

the server and why you and I

25:24

might both be falling out of love a

25:26

little bit with it, I think. I think

25:28

that's fair. I think that we gave

25:30

it a fair shake and we have a

25:33

number of things that we kind of

25:35

came to realize. Are you ready for that

25:37

Chris? Yeah, and I want you get

25:39

comfortable comfortable. If you need to lay down

25:41

first, go ahead and make yourself a

25:43

cup of tea and then let's get into

25:46

this. That's what they say on YouTube,

25:48

right? Grab a snack, grab a drink, sit

25:50

back, relax. Smash that like button. So

25:52

listening to an episode of Linux Unplugged I

25:54

believe it was, what, a year ago,

25:56

you had an idea for a project.

25:58

at this a long

26:00

time on-raid user. And so I liked

26:03

so for various aspects. One of the things for various aspects.

26:05

One of the things that I really

26:07

liked about it start ease it was

26:09

to start up applications. of the things I hated

26:11

But one of the things I hated about

26:13

it was text out these text fields. liked the I

26:15

really like the idea of a declarative HomeLab

26:17

configuration, something where I can have a folder

26:19

with all of my files, and it's all

26:21

my configuration for all of my all apps. lab

26:24

apps. And And so kind of kind of was

26:26

the obvious choice. It was something that I was

26:28

interested in. And at this time, I had

26:30

a Nick I had a Knicks server built was

26:32

using using Knicks OS which is

26:34

a feature. a feature that's built into Knicks

26:36

OS, where that the Is that stuff? Exactly,

26:38

the system The system the short of

26:40

it is you can essentially have

26:43

small system configurations for each one

26:45

of these containers. of these And so

26:47

for each so for each Home not only

26:49

could I put, could I put, say like...

26:51

Sonar in in there as a service, but

26:53

I can also have tail scale in this one

26:55

container. And this I container. in so I

26:57

really fell in love with this, was an issue there

27:00

was an issue, it which was that it was

27:02

one giant monolithic config. Not only was it a

27:04

config for a server and all of the things

27:06

you might have configured there, but then every there,

27:08

but then every have its own, its

27:10

own, child configuration. And

27:13

so so managing these containers was

27:15

a separate command, which felt

27:17

not not Nixi. And then having to

27:19

to recompile the closure became

27:21

very painful. listening to an episode of

27:23

listening to a lot was of about

27:25

was talking about containers like, well, why can't

27:27

I was like, well, why run NixOS in a container?

27:29

I'm already running it in these N -Spawn containers, but what

27:31

if I can run it in a Docker container? in a

27:34

Docker And so it And so me down

27:36

this path of building an entire platform

27:38

as a service, which is something I

27:40

called I called Nix server. And And I'm actually

27:42

still running it. It's running most of

27:44

my home lab. what it allows you to

27:46

do is write allows you to do is write configurations Knicks

27:48

you can compose where you three, four. two, three,

27:50

four different services. so for me, this was

27:52

a dream of, oh, I'm gonna build

27:55

this out and release it to the

27:57

community. the But sadly to say, we

27:59

never got there. got there. We actually even met with

28:01

a few of the movers and shakers in

28:03

the Knicks community on this one, you know,

28:05

the debt cis guys, for example, and we

28:07

just ended up finding that the flexibility of

28:09

the Knicks module system just lacked a few

28:11

knobs and dials that you need. You know,

28:14

we were talking earlier before the episode about

28:16

how if you wanted an application, for example,

28:18

that had a side car database container. If

28:20

you want to customize, was it post-gress you

28:22

were talking about? If you wanted to customize

28:24

post-gress, you have to try and fork the

28:26

upstream module, which that's easy, right? No. So

28:29

if anybody's ever tried to replace a module,

28:31

a service module, and Knicks, you'll know that

28:33

it's kind of a little bit of a

28:35

nightmare and it's pretty messy. And that is

28:37

to say, like, Knicks, there's multiple parts to

28:39

Knicks, and specifically the part that I really

28:41

gripe with. is the services side of things.

28:43

There's a lot of inconsistencies upstream. And when

28:46

building out Knicks server, I strive to do

28:48

a number of things that I was trying

28:50

to fix from my experiences on Unraid, which

28:52

was if you're writing a file inside of

28:54

a container, sometimes that user is not the

28:56

user that you're using on the server itself.

28:58

And so how often if you ran into

29:01

the situation where you're trying to read a

29:03

file on your server, low and behold, it

29:05

has a random user ID from whatever docor

29:07

container user that was set up. And so

29:09

I had systems in place to fix this.

29:11

And what I would find is that these

29:13

upstream Knicks OS modules would all implement the

29:16

user slightly different. And replacing these or trying

29:18

to jump in and modify them got really

29:20

messy, really quick. And if anybody's written services

29:22

or looked at these services, you'll know that

29:24

like they're very inconsistent. Now on the flip

29:26

side, Knicks has packages. And their package system

29:28

I actually really love, and I continue to

29:31

use Knicks to this day. I'm just starting

29:33

to shy away from using Knicks service modules.

29:35

It's it's interesting Chris. I'm curious to get

29:37

your take on like how you find the

29:39

module system to actually interface with well I

29:41

guess I guess I'd like to know how

29:43

is it hard to so I guess it

29:45

sounds like if you want to take a

29:48

module and change it, maybe you want to

29:50

swap out the version of Postgress, it sounds

29:52

like it gets hairy quickly. Could you go

29:54

into detail there? What is hairy about forking

29:56

the module? Overwriting specific configurations is definitely possible.

29:58

Like if you're layering your service modules, you're

30:00

able to override a particular configuration. So if

30:03

you want to change what port something's running

30:05

on, you can typically do that. Where it

30:07

gets a little harrier is that there's some

30:09

configurations or there's some safeguards that are in

30:11

place. Like there's these assertions inside of these

30:13

modules. And particularly with Postgress, what I was

30:15

doing was trying to change the Postgress user.

30:18

So I wanted to be able to deploy

30:20

NextCloud inside of a Docker container with Knicks

30:22

OS. And I wanted to make sure that

30:24

the Postgress user was the same world writeable.

30:26

permissions as the systems user. So if you're

30:28

using Chris as your username on your server,

30:30

that all of the postgress data matches up

30:32

with Chris's user ID. Just so I understand.

30:35

This is a darker container that already exists

30:37

or a darker container you're composing with NICS.

30:39

It's a darker container composed with NICS. So

30:41

NICS server would essentially spit out darker containers

30:43

that would boot up NICSOS inside of them.

30:45

Okay. All right. And I think the downside

30:47

is those two modules don't necessarily. they have

30:50

no context of each other whatsoever. The postgress

30:52

module is a standalone thing and the next

30:54

cloud module is a standalone thing. They're not

30:56

designed like a stack of containers might be.

30:58

They're not always the case, but there's no

31:00

logic between the two to have any relationship

31:02

for the users to match up at all.

31:05

So if you end up in a situation

31:07

like here where the database user doesn't match

31:09

the app user, you can end up in

31:11

a situation where the two are just completely...

31:13

doing their own thing. And the goal, just

31:15

again, also I'm following, the goal is to

31:17

just always be able to use the upstream

31:19

module, right? That's kind of like the end

31:22

goal, ideally. Yeah, so you can create overlays

31:24

for packages, and so you can bring in

31:26

like unstable versions of your packages, but if

31:28

you wanted to. replace a

31:30

service module like services .talescale

31:32

.enable equals true. One

31:34

of those modules, it's

31:37

a lot trickier. You

31:39

have to put it

31:41

on a denial list

31:43

essentially and then be

31:45

able to shoehorn in

31:47

your own module on

31:49

top of that. forking

31:52

it becomes a little

31:54

messy. And so really

31:56

what I ran into

31:58

is when I was

32:00

trying to customize these

32:02

modules to run inside

32:04

of a container to

32:06

have proper permissions to

32:09

essentially build out my

32:11

home lab. I ran

32:13

into situations where the

32:15

way that the upstream

32:17

modules were written a

32:19

lot of times there

32:21

would be hard -coded values

32:24

within the module itself.

32:26

So it would require

32:28

me going and posting

32:30

a pull request and

32:32

asking for it to

32:34

be turned into a

32:36

variable. Or

32:39

the way that it was just configured

32:41

and composed, there's so many assumptions in

32:43

there that it was really the problem of

32:45

inconsistency. So I reached the point. As

32:47

you see often with community credit stuff,

32:49

right, is they're kind of scratching their

32:51

own itch. Yeah. And I'd reach the point

32:53

where I'd go to deploy a new

32:55

service and I'd actually get really worried.

32:58

I'm like, okay, what am I going to

33:00

uncover here? And I think we've all

33:02

come down, we've all felt that. Like

33:04

I know that you've done the next,

33:06

I'm sorry, the next cloud module recently. And

33:08

it becomes a whole thing where it's

33:10

like, let me open up this module.

33:12

What options are there? And you have to

33:14

like learn each and every module. And

33:16

I wish there was more consistency there. You

33:19

end up instead of reducing friction for

33:21

trying out new projects, you end up

33:23

kind of increasing it. It also relies on

33:25

what's available in NICS packages or the

33:27

NICS module ecosystem because I know it exists

33:29

now, but when we were looking at

33:31

it in the summer, image wasn't in

33:33

the NICS module system. Sure. I mean, to

33:35

be fair to NICS, NICS is probably

33:38

one of the most expansive. I mean,

33:40

maybe the AUR, but NICS is probably

33:42

the second, if not the most expansive package.

33:44

No doubt. But that kind of ignores

33:46

the other elephant in the room that

33:48

dockers become the kind of standardized packaging format

33:50

for a lot of apps to ship

33:52

their software with. NICS goes off it. that

33:54

Before you go there, though, but I

33:56

agree with you. Before you go there, going

33:59

back to the module stuff. See, I actually I

34:01

actually find what you describe as

34:03

friction. that's how I learn. how I nice having all these

34:05

nice having all these options. I like being

34:07

able to understand all of this and

34:09

learn these capabilities in there and being able

34:11

to just able to this one spot and

34:13

understand everything this is capable of doing.

34:16

I find that to be remarkably remarkably So

34:18

for me, I don't find that to be

34:20

friction at all, but I do agree

34:22

with you on the. at all. But I do

34:24

agree with you thing, it's like that's

34:26

where the industry is going like,

34:28

has gone. the there, is really

34:30

distributing software in containers. containers. Doctor

34:32

ended up becoming the standard packaging

34:34

format for the Linux home server ecosystem. ecosystem.

34:36

a lot of stuff there's a lot of

34:39

stuff going on in with it, of

34:41

of course, with Kubernetes and that

34:43

kind of thing. that kind of thing. But the

34:45

thing that you touched on briefly,

34:47

on Shane, was talking about rebuilding

34:49

the entire closure every time and

34:51

how that can kind of be

34:53

an impediment to to iteration and

34:55

the just the fact that you can't

34:57

just make a one line change

34:59

and have just that one thing change. It

35:01

reevaluates the entire every time. time. Yeah, I

35:04

think that's true. We've all been We've all been

35:06

there up a you mess up a

35:08

line of a configuration and now you're

35:10

staring at a error. And it doesn't really

35:12

and it doesn't really have anything to do

35:14

with where your mental context was, but now you

35:16

have to understand Nix as a language. a language. And

35:18

it takes a certain amount of effort and

35:20

eventually you get used to reading these things used

35:22

you get the muscle memory. and you get but this

35:24

iteration cycle of like, okay, time cycle of like,

35:26

rebuild this. The way that NIC server worked

35:28

itself the way that again, it would build out an

35:30

entire closure and shove it into a Docker

35:32

image. closure and shove it of the things I started

35:34

thinking about one of the things I I jump

35:37

to next Darwin real quick I that to Nick's

35:39

Darwin configuration feels really great when I

35:41

set up a new machine. great when I set

35:43

up a new machine. taking a new machine out

35:45

of the box, the cloning my repo my repo,

35:47

running running and having everything in its place,

35:49

even my in apps. even my MacOS

35:51

apps. However, when I use it day to

35:53

day, that iteration that you're talking you're

35:55

talking about, that building the whole world,

35:57

it becomes an an impediment.

36:00

I'm sitting there and I just want to make

36:02

this one line change. I think I was working

36:04

with you today and we was update your path and

36:06

it was we minutes, your path and it to the right

36:08

place of the configuration, to making the change and then

36:10

building, having to type your password a few times.

36:13

And so there's something there where it's like,

36:15

few the so happens like when you're setting

36:17

up a new system, but then like like

36:19

you're paying every day system, but then like what feel

36:21

worth it. every day See, I feel like this is a

36:23

worth it. So I feel like this is a flipside too. in

36:25

one way, who somebody who doesn't always

36:27

know what he's doing, I appreciate

36:29

that that if it's not it's not right, it

36:31

won't build. I also can I also

36:34

can if it's I know if it's not right,

36:36

it won't build. I can just

36:38

throw something in there and let's just see how

36:40

it goes. And sometimes it works the first

36:42

try and sometimes I have to iterate a few

36:44

things. a few things. And you

36:46

when you combine... Like like the output air

36:48

messages with something like Claude I can get to

36:50

the bottom of it in of it in 30 seconds.

36:52

So I don't find that to be too particularly

36:54

to be too I do agree either I

36:56

know, agree that you know take a while to

36:59

build. take a while but that just sort

37:01

of depends on how much you're managing with Nix

37:03

on how much you're managing I don't think we should

37:05

lose sight of the fact that we should and

37:07

there's fact and you could use Nix like

37:09

you and Darwin, and you could just have it

37:11

there's packages. you could use Knicks that you wouldn't be such

37:13

a problem. It just really depends on the scope. on Darwin

37:15

And it's kind of funny because you know I think you

37:17

can go too far with it. I think I think

37:19

you can have it manage too much stuff, and then

37:21

you end up kind of in the position you're

37:23

at, in the position you're at It seems like to

37:26

me, this is a system, a system system. build This

37:28

is how developers would do it, right? This how would

37:30

development works, is it either builds or it doesn't build. is

37:32

it either builds And so this is how software developers

37:34

think about it. how But it's not really, I think,

37:36

how systems people think about it. not really I

37:38

do enjoy stuff like about you

37:40

remove a package from. about it.

37:42

your configuration that Nix will evaluate the

37:44

entire world. mean, world. you say, it's

37:46

a say, it's a flip side. it's a curse

37:48

and sometimes it's a blessing it's a

37:50

where it will automatically remove the thing

37:52

that you've removed from your configuration. you've

37:55

removed from deploy that thing

37:57

as an entire, thing know,

37:59

artifact. you know, artifact. It's just I wish there

38:01

was kind of two guess I guess

38:03

what I'm asking for is like

38:05

is like two modes like a jackalind or hide which

38:07

is like a pure mode which

38:09

is only deploying what's fully and and

38:11

and the entire closure. The other

38:13

one is a really quick like deaf mode

38:15

mode just I can just you

38:17

know I know there's all that kind that

38:19

kind of stuff like short -term

38:21

stuff But like but like Let's say I'm

38:23

working on a configuration for Starship or for ZLSH

38:25

or for NeoVEM or something or something like

38:27

that and I just you know, I know a

38:30

to hot reload the config can I put

38:32

myself into a temporary into a reload

38:34

mode where it's only looking at

38:36

a very small looking at a of the

38:38

world. portion of the world? I think another

38:40

idea is being able to mutate the

38:42

world the then have it read what that

38:45

mutation was. that I edited the config by

38:47

hand and then you can import that into

38:49

the NixOS config. So it feels a

38:51

little more natural. So it I think where I'm

38:53

at with I right now where do love it for configuration. now

38:55

I do love it for packages and I'm going to continue

38:57

to use it that way. for But actually

38:59

I spent a considerable amount of time over

39:01

the last few months building a separate

39:03

project I I've convinced myself from my home

39:05

lab I think I'm moving away from few

39:07

specifically these separate service modules convinced I'm going back

39:09

to my home lab. one of the simple reasons

39:12

of going back to Docker is it's

39:14

kind of like the Ubuntu thing thing, it

39:16

is the standard. There's always an

39:18

example for it. You can go

39:20

to official official upstream if you're a project

39:22

maintainer like Image like Image, feel a

39:24

responsibility to publish a Docker a file. file.

39:27

so with with it's kind of like this mix

39:29

of this mix of maintain it. going to maintain it.

39:31

And to the author they're going to

39:33

interpret the needs and how they're

39:35

going to implement the service module.

39:37

module. And there just is that lack of lack of

39:39

consistency. like what I really want, and

39:41

and maybe this this future exists,

39:44

I'll switch back, is if we

39:46

saw more package developers or the

39:48

service developers publishing their own are the ones

39:50

So then that were the ones controlling what

39:53

was like. interface was like. be That'd

39:55

be great, came be great if it came from

39:57

them. Like they do with Docker compose files today

39:59

docor today, yeah. yeah. Absolutely. K,

40:01

EEEB.I-O slash self-hosted, .io

40:04

up self newsletter and

40:06

take 5% Go over there, sign up for

40:08

their newsletter and take 5 % off your

40:10

first order. well-built open they specialize in beautiful, well

40:12

really keyboards. also A lot of split keyboards,

40:14

the ones I really you but they also

40:16

have those regular types for you folks that

40:19

don't want the split keyboard. Friday, On an

40:21

Friday, keep an eye out for one

40:23

of their keyboards a a their lot like

40:25

the Microsoft Natural Natural 4000. which was a great

40:27

keyboard back in the day. the day. But

40:29

even more ergonomic. lots of options,

40:31

and it's gonna be at a great price. be

40:33

at a great come fully

40:36

built fully built or... You can get a kit

40:38

and assemble it yourself. it And a lot

40:40

of the parts you're gonna have no

40:42

problem with because everything is because so there's

40:44

no is hot for so that kind of stuff.

40:46

They also have of that with like nine to

40:48

16 keys. I use that here 16 for

40:50

my OBS machine. There's all kinds of

40:53

things. You could do kinds of things. You deck with

40:55

that thing or maybe you wanna have

40:57

like a home theater PC control board or

40:59

shortcuts for home assistant. Those are really,

41:01

really slick. And of who they are, they

41:03

also have lots of DIY parts parts like

41:05

and they support open source. and they

41:07

support open parts they publish 3D

41:10

They're part of the they're part

41:12

of the core team. team. of

41:15

their keyboards use use QMK.

41:17

If if you're a keyboard person,

41:19

you know that's a big deal. a

41:21

go check out go check out Kebio. K-E-E-E-B-I-O -E -B If

41:23

you sign up for the news If you sign up

41:26

for the 5% get there and grab go there and

41:28

grab one of those Black Friday Cyber

41:30

Monday deals and get yourself a keyboard that

41:32

deserves to be touched all the time.

41:34

You're interfacing with your hardware all the time

41:36

the time like me. Why not have it

41:38

be great? have it be Why not have

41:40

it feel great have it why

41:42

not have it be something

41:45

you can repair and fix

41:47

that runs an open firmware? Support the

41:49

show and check them out. Go to k-e-e-e-e-b.io/self-hosted. That's Gebio

41:51

slash self-hosted. So what's the idea behind

41:53

where the idea behind where it headed? Is

41:56

it a directory full of compose files

41:58

that you can kind of ingest? into

42:00

into some tool what like what's going on. on?

42:02

So I So I actually started and I I kind of just

42:04

reset my brain of like, what do I want to

42:06

do in my do I What am I deploying? home lab?

42:08

What am I it came down to like a

42:10

handful of things. I have I have cron

42:12

jobs, I have shell scripts. Sometimes I'm

42:14

a I'm a go so I have binaries

42:17

that I need to run that

42:19

I'll write myself. I'll write A A of

42:21

times, darker images or files. files. And

42:23

what is a reasonable is a reasonable

42:25

way to manage all of this? And like there

42:27

isn't one system that really addresses it. it. And

42:29

And so this idea had this idea of like, well,

42:31

what if I could know, laugh if you want to, but what

42:33

you want to, but what if I could yeet

42:35

a binary and a server and it would run

42:37

as a service? I could just SCP What if

42:39

I could just it that file? and

42:42

it would automatically configure itself as a service.

42:44

And so this is where I started, and that's

42:46

kind of where I started in what I

42:48

built. I built. And where I landed on is this

42:50

project that I'm calling I'm calling yeat for fun for fun,

42:52

but you can yeat. any one of these things.

42:54

You can eat a binary or docor image

42:56

or compose file to a remote system. So

42:58

you can run this from your laptop. And

43:01

what I'm experimenting with up as these up as

43:03

when it's like a script or if it's it's like a script

43:05

a if it's a binary, and I compose if Docker

43:07

an if you're pushing an image or if

43:09

you want to just send a compose file. I

43:11

And then I stand them up inside of

43:13

their own network space. And one And one of

43:16

the goals because I I love I love tail scale I

43:18

wanted to attach because to each one of these

43:20

services. I love, this was something I didn't talk

43:22

about. a few minutes ago, but one of the ago, but one

43:24

of the things was really striving for with for

43:26

with NICS modules and what I did with

43:28

and what I did with I loved the was I loved

43:31

the composability where docor containers are are like,

43:33

well, here's your one thing that you're running.

43:35

if you want And if you want to put

43:37

there, you do it as a you do it as

43:39

a I kind of got And I kind of got

43:41

tired of writing like files files with sidecars.

43:43

I I could just do I could just do

43:45

services scale scale, .enable. dot enable. And so so what

43:47

I built into this this tool, this this yep is

43:49

the ability to not only only put

43:51

your on your land, but you can

43:53

also put them on your put automatically,

43:55

so you don't have automatically. So you don't

43:57

have to I love it. I love

44:00

it! running like on an Ubuntu system? Is

44:02

this a Debbieian or is this on Nick's

44:04

like? It's a Bunto, yeah. Okay, so then

44:06

I use Yeat on top of that? Yeah,

44:09

so you use Yeat like on your laptop,

44:11

whether that's you know Linux, Matt. Yeah. And

44:13

then you point it at an Ubuntu server

44:15

and then you point it at an Ubuntu

44:18

server and it's kind of tail net first.

44:20

And so this was an opinion I took

44:22

which is like I love the tail net

44:24

because when things are there. And so I

44:27

put every service I have is on my

44:29

tailnet. And so when you start up, it

44:31

automatically gets on your tailnet. You interface with

44:33

it remotely from the start, which means that

44:36

you have security built in. It's just using

44:38

your ankles. And you just start throwing files

44:40

up. And underneath, it's just using SSH and

44:43

SCP to transfer the files. And yeah, so

44:45

it's something that I'm curious, like what the

44:47

community thinks about, like, what if we went

44:49

back to Doc or Compose? What if you

44:52

could just create a repository of all your

44:54

Compose files and there was a tool that

44:56

let you spray them onto your servers and

44:58

manage them as services for you? I've kind

45:01

of gotten a little tired about six or

45:03

seven years ago. I wrote what I call

45:05

the doc a composed generator in Ansible. So

45:07

my workflow for the last six or seven

45:10

years has been creates a massive yamel dictionary

45:12

with a bunch of entries that I can

45:14

then specify like per container and then the

45:17

ginger two templateing iterates over that and spits

45:19

me out a doc a composed file. The

45:21

benefit of that is that I can have

45:23

secrets in an encrypted file, as long as

45:26

I don't push them to get hub like

45:28

I did this week. And then I can

45:30

have it, you know, use Ansible to talk

45:32

to multiple service over SSH. So I manage

45:35

all the JP infrastructure that way. Does yit

45:37

do to manage things like secrets and things

45:39

like that? Is it just all plain text

45:41

in the compose wall on your box and

45:44

just gets yetted over? Or like, how does

45:46

that work? Yeah, I thought about this because

45:48

when I implemented this for next server, I

45:51

did it in like a very. security focused

45:53

way. You had secure like encryption to talk

45:55

to your Knicks box and then you'd have

45:57

age encrypted secrets. I kind of went overboard

46:00

with it and then I took a step

46:02

back and I looked at what was happening

46:04

in the darker world and I'm like oh

46:06

these are just all environment variables and you

46:09

go look at get hub actions for how

46:11

infrastructure is deployed and it's a lot of

46:13

environment variables. And it's a lot of environment

46:15

variables. And sure there's vault and there's secure

46:18

ways to be able to be able to

46:20

be able to pull these things. Treat the

46:22

home lab as what it is. Like this

46:25

is your sandbox that you're playing in. It's

46:27

my pseudo production, gosh darn it. And so,

46:29

like, it's in your environment files. You can

46:31

SEP up, like, you know, secret files. Like,

46:34

today, we SEP'd up one of your Ohoth

46:36

secrets. But then the rest of it is

46:38

like, you just edit an environment file. And

46:40

you can type that in there, and it

46:43

just kind of exists in memory. I admit,

46:45

I had to look it up, I'm an

46:47

old man. I'm an old man. I'm an

46:49

old man. I'm an old man. It is

46:52

a verb, it is a verb, it is

46:54

a verb, it. to throw something with force

46:56

and without regard for the thing being thrown.

46:59

Now I get it. Yeat, you're yeating the

47:01

files. You're yeating the docor composes up onto

47:03

the system. Exactly. You know, what's funny is

47:05

Drew, our editor, is the one that educated

47:08

me on what yeat means a few years

47:10

ago. Really? I just had to look it

47:12

up. So I'm curious, like, we're talking about

47:14

yeat as if it's this publicly available. like

47:17

what you know as I onboarded Alex earlier

47:19

today and I say okay here's what I'm

47:21

doing and immediately you hit some some edge

47:23

cases so there's assumptions that we made in

47:26

building this I made it with another developer

47:28

so our plan is right now going into

47:30

the holiday season hopefully we'll have some time

47:33

to put our head down and polish it

47:35

up a bit and really I just want

47:37

to write the documentation so people can invite

47:39

themselves on to it and yeah so let's

47:42

look forward to 2025 let's say January February

47:44

I'll shoot Alex a note and we'll get

47:46

this in people's hands Excellent, well we will

47:48

update and of course when we have a

47:51

link we'll put one in the show notes

47:53

for that episode too. Sounds like a, oh

47:55

I should have looked up another, it sounds

47:57

like an exclamation of excitement. an surprise

48:00

or energy too, or energy

48:02

too, Shane. So, I guess also it's guess

48:04

also it's Australian slang for a

48:06

very strong yes. yes. Is it? Okay,

48:08

it? near to me. to me. Yeah. Well, there

48:10

you go. go. Well, so so the you're actually

48:13

here, Shane, is that you're moving house

48:15

Shane, you wanted to co house your server

48:17

in my basement. And we ran

48:19

a server experiment this morning. And You left

48:21

your house at what time? 7, 8

48:23

a .m., something like that? I unplugged

48:25

the server at 7 .30 in the morning

48:27

what we left at eight, eight. Yeah, so you so

48:29

you brought your server from Connecticut, checked

48:32

it in the belly of the

48:34

plane, just in a massive suitcase, I just,

48:36

I wrapped it in a towel because

48:38

the suitcase itself was almost the

48:40

exact dimension of the 4U case. the for

48:43

you towel take care of it is

48:45

fine. of it is fine. Yeah. With a sign saying

48:47

this is a server this top. on top.

48:49

It weighed 53 weighed the 12 pounds

48:51

it. And the 12 me with it

48:53

overage on that. And so that me with the And

48:55

on that was so that was You know,

48:57

I put the then it was the tension

48:59

Alex said, know I put the note in there so

49:01

said and I was just this afraid TSA would

49:03

open this up I'm like, I I was talking to Alex

49:06

I'm like I don't think can get through. can get

49:08

through I don't know if they know what this

49:10

is. I actually I actually overheard. I was flying out

49:12

of a small airport. I overheard the TSA

49:14

agent agent. tap one and saying, hey, we

49:16

need your help. There's something called a server. something

49:18

called a need you to look to look at it. Wow,

49:20

we added an hour by We added an hour by

49:22

stopping for lunch on the way home from the

49:24

airport. That's fair, yeah. It's a quick, what, 90

49:26

-minute flight from Connecticut down here. down here. We had it up

49:29

and it up and running by, what, three o

49:31

'clock this afternoon? So if we take an hour

49:33

off for lunch, we'll say two o 'clock. two

49:35

o'clock. So what was the theoretical bandwidth

49:37

of your plane flight today? I think

49:39

we we it. I think there's about

49:41

30 terabytes of storage space. And

49:44

so all those drives and with the so with

49:46

all those drives broke 10 the

49:48

a second, constant. 10 you go.

49:50

That's not bad. That's not bad.

49:52

There you think we still achieved

49:54

99 .99 bad. And I think we still achieved 99.99%

49:57

far as you know, no packet loss,

49:59

as you know, no packet loss. There you go. Knock

50:01

on wood. Well Shane, it's a good

50:03

adventure, but thank you for being here

50:05

today and chatting with us. I also

50:08

want to say thank you to Paulus.

50:10

You know, when he was wrapping up

50:12

with us, Alex, he mentioned that folks

50:14

are doing 3D printed cases. for these

50:17

little units. Yeah, it sounds like there's

50:19

some really fun designs because it's a

50:21

fully open source project. 3D printed models

50:23

for the case and stuff like that.

50:25

He mentioned there was a Minion design.

50:28

It was going to be pretty fun.

50:30

What was the other one? Was it

50:32

a poker ball? Yeah, a poker ball.

50:34

I could see like, you know, somebody's

50:37

going to do a Starship enterprise. You

50:39

know it. There's Star Trek geeks. They're

50:41

going to make it. Now, this is

50:43

a special early holiday episode recording, so

50:46

we won't have feedback or boost this

50:48

episode, but we don't have very many

50:50

booths. If you want to send us

50:52

a holiday boost, we will include it

50:55

in the next episode when we get

50:57

together. And of course, you can always

50:59

go to self-hosted dot-show slash contact and

51:01

use the old classic email contact form.

51:04

The old school email form, huh? I

51:06

think it still works. I think they

51:08

might be privatizing email though, so watch

51:10

out. Is that a new doge initiative

51:13

or something? That was a joke, yeah.

51:15

Good lord. Well as always, I want

51:17

to say a big thank you to

51:19

our site reliability engineers over at self-hosted

51:22

dot show slash SRE. You get an

51:24

ad free show as well as a

51:26

special post show. What are we going

51:28

to be talking about today, do you

51:31

think? It's going to be Alex's choice.

51:33

I've got a look at that. Yeah,

51:35

there's a couple ones. We should probably

51:37

have mentioned drones over New Jersey. Good

51:40

for you. Nice. I know. Look at

51:42

me. Don't forget to check out meetup.com/Jupiter

51:44

broadcasting for any upcoming last minute meetups

51:46

we might have. Oh, you know, I'll

51:49

mention. We're putting together some meetups for

51:51

LUP 600 and we're trying out a

51:53

new self-hosted. meetup alternative at colony.com.

51:55

And if it works

51:57

for LEP works I

52:00

think that's what

52:02

we'll use in the

52:04

future. that's what we'll use in the

52:06

future and I don't something and it's

52:08

kind of nice because it's kind the

52:11

listeners can create their own events,

52:13

and then a little bit after the

52:15

event wraps up, it self bit it

52:17

just cleans up after itself. up it

52:19

So we're testing and it just.com and we're

52:21

soliciting testing colony events.com and we're is

52:23

gonna be like February for LUP

52:25

or something is going to be like Can

52:27

you believe that? No,

52:30

and you code and just just hit 600

52:32

this week too. too. What's 600? Is Is

52:34

that like years? There five must be more

52:36

than that. be more it's more than

52:38

that. No, it's more than than my daughter daughter Bella.

52:40

a half years. you kidding? years,

52:42

are it's crazy, huh? Yeah, it's crazy,

52:44

huh? All right, so you can can .kateyz .me

52:47

to find me on the Internet.

52:49

Where can folks go to find

52:51

you, Shane? you, not really on the

52:53

not No, github.com internet. We'll put a put

52:55

a link to that in the

52:57

show right, you could you can find me

53:00

on Weapon you want, want, Chris, L-A-S, you

53:02

can try the Nostra the Noster thing,.com. of

53:04

course, all the course, shows the great shows

53:06

Jupiter.com. Thank you very much for

53:08

joining us. I appreciate you coming

53:10

on, Shane. you And that was self

53:13

Shane. And that was self-hosted dot show slash 139,

53:15

139. You