Bobby, a new voice assistant and Snowy successor

[edgy flame]

afaik, bobby won't start because coreapp is missing support for rebble's timeline service

[round helm]

Got it, thanks for letting me know. I really hope someone makes it work.

[edgy flame]

send in a bug report from core app's settings page so they know bobby is in demand!

[round helm]

Done

[weary ether]

And location and voice?

[coarse cedar]

Mhh Location should work afaik. Voice is currently in the works

[fierce plinth]

Just had the core app 1.0.0.963 (Android) update (which should start enabling experimental voice dictation

Voice dictation crashes but the app starts now

However Bobby doesn't seem to be able to access the internet at all. Baked-in queries wait for an answer indefinitely, same for the "Quota" check

Not a big deal, just want to keep you guys posted on the current state of bobby in the new core app 🙂

[weary ether]

thanks for the update

[obtuse oriole]

Probably due to websocket not working yet. I think snowy has the same problem, although I was able to tell it to start a timer using the experimental dictation.

I let the devs know that I was looking forward to the infrastructure that will make Bobby work again.

[weary ether]

That feeling when you go to daily the PTR but realize no Bobby support (yet) 🥺

[edgy flame]

so true

[wooden wedge]

we should really finish it already

I haven't been using a round recently, so I had very little motivation

[shell jetty]

i did make some progress on formattedtextlayer the other day

ever since i switched to the core app my motivation has been lower since bobby doesn’t work yet there…

still has some issues with cut off text

[wooden wedge]

would it be better if the frame was slightly shorter?

so that you don't get the top and bottom being displayed under the scroll indicators?

this is already a huge improvement btw

the last thing really is to make the conversation view work, which is mostly just making it full screen widgets

[edgy flame]

oooooohhh this is so goood

[shell jetty]

probably haha, i felt like i spent a lot of time randomly changing the offsets and each time i did i ended up making something else look off

[wooden wedge]

that's the story of developing on a round

[shell jetty]

yeah i figured i'd tackle SessionLayer last, since it seems like the most difficult

i'm still hoping to use my old design i made a while ago:

[wooden wedge]

I want full screen maps

[shell jetty]

oh yeah i forgot about maps, i made this before maps were a thing

[wooden wedge]

I think generally it might be easier just making things full screen

[shell jetty]

hope we can get them to fit in memory 🤞

[wooden wedge]

mostly because full screen widgets means you can treat text as multi page widget too

and that way you always align to a fullscreen page, so you can also easily deal with the session losing content without reflowing anything

[shell jetty]

true haha

my plan is generally "follow the path of least resistance"

[wooden wedge]

yeah

[shell jetty]

i should really push these changes to your branch at least

i'll try and do that later

[edgy flame]

the little dot i'd love to see come to rect as well

at the bottom of a conversation

[shell jetty]

oh the indicator layers! though based on what katherine said about memory usage, i wonder if its possible to allocate any more layers

guess i’ll find out haha

[edgy flame]

rect does do the triangle indicator layers!

i just am a big fan of using the dot to indicate end of content, like timeline does

[wooden wedge]

I would argue it's better to follow the platform guidelines

[shell jetty]

alright i pushed the changes -- the code isn't great but at least this way we can say we're making progress!

i am slightly concerned about how i just literally pasted the indicator layer code into like two other places haha

(feel free to delete it)

[wooden wedge]

the change in formatting makes things a bit harder to review

[shell jetty]

oh i changed formatting? it wasn't supposed to haha

https://github.com/pebble-dev/bobby-assistant/compare/main...hellcp:bobby-assistant:round

[wooden wedge]

yeah, that looks fine

I wish I could figure out how to make icons work in menu on round

it does work in some apps, though I assume it's a legacy feature or something

[shell jetty]

i think the issue is that SimpleMenuLayer for some reason doesn't support icons on round (this actually seems like a bug tbh)

if we swapped it out for custom cell drawing we could probably get the icons to appear, but also, that feels like effort

[wooden wedge]

yeah, I think it might be nice to switch to something custom

the height of the menu items bothers me a lot as they are thin

[weary ether]

is there a functioning build I can sideload?

[weary ether]

thanks; on first load it hangs after allowing location (keep in mind I'm on OG phone app) then it recovered but the first question I asked it pulsed dots indefinitely

[wooden wedge]

hmm, the changes shouldn't have affected anything with regards to that

[weary ether]

oh

I was logged out

ok it works now

weather card looks nice

albeit misaligned

the setup seemed PTR native

not sure what all work has been done but it seems to be getting there

thanks for the build!

[wooden wedge]

the setup and the main screen

and also what freakified showed

the conversation wasn't really touched

[weary ether]

makes sense

keep up the good work!

is there a fork or branch?

for PTR

[fervent steppe]

do y'all have commit privileges on main? if not we can fix that

[shell jetty]

it seems this can be replicated in the emulator:

those are both from our branch build, but on chalk we see the crash after accepting, but on basalt there is no crash 🤔

[wooden wedge]

yeah, I know about that one

I was talking about the login issue

[wooden wedge]

we would rather work on our own branch for now

that being said, it being under pebble-dev would likely be nicer

[fervent steppe]

sounds good, whatever makes sense!

[wooden wedge]

we work on hellcp:round rn

[weary ether]

unless this is the wrong repo?

[wooden wedge]

it is

[weary ether]

oh ok

[fervent steppe]

wtf

aha

ok added @wooden wedge , @shell jetty to

[shell jetty]

ah that’s probably because i made freakified an org account, which was in retrospect a Bad Move

[fervent steppe]

yeah

I added your personal account

let me know if you would prefer a different one

[shell jetty]

excellent, thanks!

[wooden wedge]

it's an honour to become a bobbyist

[obtuse oriole]

Decided to spend some time with the old Pebble app and try out the Round build from earlier this month. It's working great!

[shell jetty]

better than nothing!

i have a slightly improved (paging in the conversation view) version on my local branch, but the code is much too horrible to push and it's still pretty glitchy

[edgy flame]

hey I found the thing https://github.com/pebble-dev/pebble-firmware/blob/main/src/fw/applib/ui/action_button.c

that the system uses to draw the action pip

round and ContentSize behaviors and all

i think Bobby has its own implementation of this

[sick apex]

like the little black half circle that sometimes pops up on the right hand edge of the screen?

[edgy flame]

that's the one

[sick apex]

interesting

[sick apex]

wonder if some code space could be saved then

consolidate

[coarse cedar]

an insignificant amount, but yeah

[edgy flame]

if i felt super strongly about it i might PR bobby to use this code

not about space savings, it's about nitpicking (and chalk+emery) for me

[coarse cedar]

the thing we gotta keep in mind though is that once we add something to the sdk, older firmware will be unable to use it

until we update it

[edgy flame]

ah, yeah

this would have been a problem aplite had with 4.X APIs- is there anything that can be done for backwards compatibility?

[coarse cedar]

i can take a look! I would sure love to do something

for current sdk releases we (core) just straight up copy aplite from an old SDK

for basalt chalk and diorite we gotta check how we wanna do it

basalt and chalk are easier because NimBLE is working on it so we could just create updates for those platforms

[sick apex]

new firmware features for Time/Time Round is gonna be siiick

[coarse cedar]

what we could do for now to ensure backwards compatibility is reuse old sdk for a-d so that those always stay compatible with the current version

I would not like it though :(

[wooden wedge]

it's just a black circle, I did just submit an asset to display the right image for chalk

though I haven't looked at emery for that

it's easy enough to just have it be a platform specific image though

[wooden wedge]

"working"

[coarse cedar]

well

[wooden wedge]

it's not exactly stable

[coarse cedar]

more working than silk

[wooden wedge]

og pebble is also more working than silk

[coarse cedar]

😭

[wooden wedge]

people should just replace the boards in their silks with asterix internals

[coarse cedar]

i mean technically we could make replacement boards

[edgy flame]

then you could have a silk case with an asterix board inside it!

[fervent steppe]

I would like to buy some asterix boards

actually really what I want are to buy up all the returned Asterixes that have button failures

I will offer literal dollars for those

[fair dagger]

Silk = pebble 2?

[sick apex]

so far Core has been telling people that they can keep their button failure'd Asterix

like 2-3 people already have gone through it

Core just ships them a new one and says they can continue to use the old one for development or whatever

[proud flint]

Is note to self still working for you? I've got it to the point where it recognizes my speech, but I never see anything appear on the timeline.

[barren raptor]

the timeline doesn't work on the new coreapp yet

[proud flint]

Okay good to know. Thanks!

[icy furnace]

I assume Bobby isn't funtional yet on the new coreapp?

[edgy flame]

still no, unfortunately - it needs to grab a timeline token and a rebble account ID to check subscription status and... do timeline things, like set reminders

i think it could derive a rebble account from a timeline token, once timeline is supported in coreapp

[icy furnace]

Ah bummer, i was hoping to give it a go! Ill go ahead and cancle the sub for now then. Thank you!@

[fair dagger]

Hopefully they add timeline support soon.

[icy furnace]

I agree,

[rain plinth]

Is the subscription only to pay for Gemini api queries? If you have your own Gemini subscription could you use your own api key (assuming that configuration option was made available)?

[edgy flame]

the subscription also pays for google maps api, a currency converter, The Weather Channel, dictation, and faster timeline sync

[hollow rivet]

The app itself is open source, so you could theoretically source all the services yourself if you really wanted. It's even easier now since the updated version of the Pebble CLI tool actually builds the app

[rain plinth]

Good to know! Thanks for the info

[edgy flame]

https://github.com/pebble-dev/bobby-assistant/issues/13

[obtuse oriole]

I wonder what would be the better move. The ability to do the full login right from the Bobby settings page, or an edit field to paste in an account token that you could get from the Rebble dashboard or something. Or maybe something like logging in to a streaming provider where your watch gives you a code that you type in on Rebble somewhere to connect your account.

[edgy flame]

copying and pasting tokens sucks, and it was a design goal for Bobby to not do that

[barren raptor]

great news, bobby can speak Cyrillic without crashing on p2d

[coarse cedar]

:D

[fair dagger]

I have noticed that it gets stuck coming up with a response sometimes and just loads forever.

[upbeat garden]

yay i fixed stuff successfully

[coarse cedar]

i think there was an issue with old watches and the old app where cyrilic would crash the watch

[upbeat garden]

yeah but also there was PKJS encoding issues in the new app where non-latin characters were broken

[coarse cedar]

oh i know, I was experiencing that issue in openmensa!

[wooden wedge]

I think the issue is more of lack of ram to load in the font

but I'm not entirely sure about that

[coarse cedar]

could be!

[wooden wedge]

there's a bit more space in c2d, right?

[coarse cedar]

yea

[wooden wedge]

I guess in a way it's a bug there's more space available, since that's not really a compatibility mode, but on the other hand, the docs originally did not make it very clear that you should keep some space for language packs and voice recognition and all

[barren raptor]

@south palm from #now-what message

you are using coreapp ver 1.0.6.8 right? android or ios?

[south palm]

Do I need to be signed into rebble for this to work?

[barren raptor]

yes - bobby requires a rebble subscription

[south palm]

ahhh ok thanks

[last flint]

Will/can the Rebble “boot” process work with the Core app to enable this? If you know. I assume it does given all the words you’ve said but I didn’t know that.
(Also I’m on iOS so I don’t count 😢)

[barren raptor]

currently, boot isn't supported on coreapp - but you can login with rebble for the app store and that is how it recognizes you

i personaly hope they will support boot one day, but they've shown no indication that they plan on doing it

(which is unfortunate since i'd love to use RWS voice instead of core's)

[edgy flame]

it kinda uses boot

it uses boot to log you in to the locker and appstore lol

we did build oauth for cobble but coreapp just uses the old boot url link method 🤷‍♀️

i assume that's getting torn out in the future anyway, so it should be on our radar to add self-contained rebble login to Bobby

[barren raptor]

lmao really?

[edgy flame]

it did last couple times I set the app up

i assume that's all getting torn out as part of appstore switcher

eventually

[normal bane]

Looking at Bobby self hosting - i need to spin up user identifier first, then use that url for Bobby's env var?

[normal bane]

or... the faux one? whats the diff?

[weary ether]

I think Bobby relies on all of RWS to work

[wooden wedge]

@shell jetty I added a bit of support for emery, and since round support isn't working out this far, maybe we could at least get it working on emery?

the outstanding issues are mostly content size related

I haven't really done much tbf https://github.com/hellcp/bobby-assistant/commit/1ac5fd3ac40878b5f1c4796a1bdde4f77646c7dc

[edgy flame]

the emery in question

solid start

[barren raptor]

long dialog bubble

[edgy flame]

it really needs to support ContentSize for all the text in the conversation screen

or at least just to use a bigger font on emery

[wooden wedge]

content size support is not that difficult (outside of the home screen bubble anyway)

should that display show 4 option in menus?

[barren raptor]

imo yes

[obtuse oriole]

Definitely wondering what'll happen first: Bobby working on the iOS Core app or people starting to get the PT2.

[wooden wedge]

it does work on android

so clearly it's possible for it to work

[normal bane]

Wasn't sure if this was the case since it works with the Core app on android...

[shell jetty]

nice! i like the longggg dialog bubble

[wooden wedge]

I wonder why it touches the left side

oh well, there's work to do there

[edgy flame]

the launcher does four-and-a-half, the watchface picker does almost-four, and settings does... basalt sizes, looks like

emery is a strange beast

[wooden wedge]

settings on basalt also looks weird tbh

it's 4.5 on basalt

[edgy flame]

maybe it did seven years ago, I couldn't tell you what looks weird now

xp

[wooden wedge]

that health icon looks familiar

also looks thin as hell

what happened

[edgy flame]

it's all 45 deg angles

[wooden wedge]

hm, yeah

[edgy flame]

they get anti-aliased and appear thinner

[wooden wedge]

I could correct for that and offset it by half a pixel

that would fix it

[edgy flame]

you can see the same thing in the Watch Faces menu icon on spalding

[wooden wedge]

I think that icon lacks a bit of visual interest

compared to everything else

[edgy flame]

the heart? i agree

for cobble i added a little line graph in the middle

https://github.com/pebble-dev/iconography/blob/master/icons/25px_Heart.svg

the 3.x app face was the same heart, no details, but huge

[last flint]

I didn't like how settings looks but I imagine it's because there aren't icons. On Kagi News I scaled to around 4 and a bit with the header pushing it down which I think is the same as the launcher. I just eye-balled based on the launcher though. That's with the default/large (?) of Emery from the ContentSize API which I put as FONT_KEY_GOTHIC_24_BOLD. I scaled it from a cell size of 44 to x * 1.25 for "Large" ContentSize

I don't know if this helps anything at all. Just noting because I was trying to recreate menu sizes around the ContentSize API recently

[round dagger]

Same question would it be possible to connect bobby to my ollama self hosted ai to do all the ai stuff for me? Gemini is a no go in my house.

[barren raptor]

You'd have to modify a fair bit of stuff

It's open source, so it's entirely possible, but it's not particularly simple

[round dagger]

I'm gonna look into it. Opensource+selfhosted is the way to go imho

[novel fiber]

Currently I have got different issues with Bobby on Android with the new app. At first: My new Pebble waits endlessly to stop listening and to answer. This seems to be faster with the old Pebbles (So I guess this is some watch firmware issue). But there is an issue that all devices show: Longer answers often get stuck. Means, the system returns some text, then shows the dots at the end which should indicate that the system will return more data - but this never happens. I guess this is a known issue?

[fervent steppe]

Definitely file a bug on the new app with this.

[wooden wedge]

I guess bobby needs to get its shit together on the round now

the only thing remaining for ptr support is the thread view, which is also rather annoying tbf

[edgy flame]

do we have freakified's mockup linked here?

pagination, centered widgets, and a dot at the bottom of the thread, like timeline

[obtuse oriole]

Selfishly hoping it works on the iOS Core app by the time the Time 2's start shipping out. It's the primary reason I'm still using the old app.

[edgy flame]

i'm not optimistic.. it worked for a bit on android, but now it's broken again if you enable the native store

i think it is on rebble to add some internal sign-in method for coreapp

...it's on core to add iOS websockets though

eventually

[obtuse oriole]

Yeah, and I'm guessing they're busy trying to get the native store smoothed out. It'll all be good eventually. (Maybe someone will make a firmware that spoofs a PT if there is enough demand.)

[shell jetty]

ah i remember this one

[shell jetty]

this was it i think

i was trying to make it as simple as possible

[obtuse oriole]

That looks nice. Plenty good enough to let people know when to scroll the screen. Those of us with Round watches are used to scrolling to read everything.

[edgy flame]

delightful thing

[wooden wedge]

is it viable for us to bump the version of Gemini yet?

[hollow rivet]

What's up with the weird thickness on everything? Is that just because it's being scaled up?

[wooden wedge]

yep

scaling fonts is ugly

[obtuse oriole]

Bobby is beautiful on the inside.

[wooden wedge]

I have a feeling this wasn't the intended way to use this api

if anyone wants to try a version with content size working to some extent

(I still have a bunch of work to do)

it would be nice to ping emery people, but I am kinda withholding myself from doing that as sideloaded bobby simply doesn't work with core app at all

[edgy flame]

it does do the thing insofar as font sizes changing

the consent text scroll screens looked perfect

[wooden wedge]

should I drop the bold on the conversation screen?

that text does seem very large actually

[edgy flame]

i think so, yes

[wooden wedge]

I would also love to see what it looks like with larger notif text setting on emery

it should use 36 size gothic

[edgy flame]

seemingly no change

[wooden wedge]

ohhh

I think 36 must not be accessible from the sdk

[edgy flame]

it does scale down on the conversation window

[wooden wedge]

I did make it fall back to 28

[wooden wedge]

hmm

well, I guess supporting extra large isn't possible yet

rewriting the views to make sense with the different display resolutions is fun

though I don't want to imagine what this code would look like if you tried to add round support

@fiery gale do you think you could bundle your existing work on gothic 36/bold for use with bobby?

it would be a pretty good way to test how well it works in that context I think

[fiery gale]

Well, I wouldn't mind giving it a try.

I made it kind of as a test, so right now this is literally all there is to it. Character set is limited and it needs some rigging to be turned back into a proper font.

[wooden wedge]

yeah, I'm asking out of curiousity

bobby is kind of a test mule for features on pebble and core watches

and so any time there's an opportunity to test something out that would make the platform itself better, it seems like the perfect candidate

I am almost done making the weather widgets adjust the font size to the content size setting

[gilded maple]

Using this version, I can’t get past the release notes page on my PT2. Ignore me… back button, lol

Looks much better, text flow is good. the extended speech bubble is the only obviously odd looking thing. And of course it doesn’t work 🙂

But all the screens I can access look good

[wooden wedge]

the small variant on basalt is fun

[wooden wedge]

yeah, the bubble needs some work

[gilded maple]

It is nice to know that the microphone is indeed working though 🙂

[edgy flame]

i think we need a bigger horse, and more defined margins around the bubble

[wooden wedge]

a bit bigger text might also make sense

[barren raptor]

big bobby big bobby big bobby!!!

[wooden wedge]

I have a bunch of bobby assets to draw

[fiery gale]

Ok, so I have the normal variant in font form. It will probably need adjustments and I don't know how well it will convert

[edgy flame]

i'm toying with the idea of replacing the pink window background with a pride flag during the month of june

[fiery gale]

Seems like it could get distracting?

[edgy flame]

only on the home/speech bubble window, not conversation screens

[wooden wedge]

do you remember

[edgy flame]

evidently i forget

[wooden wedge]

you did suggest a rainbow background right after the message where I sent it

[edgy flame]

agckg

[fiery gale]

I still don't enjoy distractions like that

[edgy flame]

fair

[wooden wedge]

have fun

[fiery gale]

why does the horse look disappointed

[wooden wedge]

it was meant to look like it's holding a flag and it sticks out on the other side

that didn't work out very well

[coarse cedar]

:/

[wooden wedge]

rebblers know of so many more pieces of bobby art

[gilded maple]

Tiny quota Bobby

Who will run forever if I let them (or the battery runs out)

[wooden wedge]

I think I like the size of bobby on that display

it's tiny, but I think that's fun

[gilded maple]

I really like speedy Bobby, and fence Bobby

[wooden wedge]

fence bobby needs help

[edgy flame]

i wonder if obelix is fast enough to draw vector fence bobby

[wooden wedge]

a bunch of fixes

it should really be the actual pebble ui for phone

[coarse cedar]

NO

WHERE'D THAT GO

[wooden wedge]

lmao

@coarse cedar make sideloaded bobby work!!!

it's annoying that the core app simply refuses to let people test the emery builds

[coarse cedar]

im dont

me no workey on that

😔

[wooden wedge]

@upbeat garden 🥺

[coarse cedar]

https://tenor.com/view/bird-brain-teto-4lokogangemporer-jamiep-birdbrain-gif-14306545906510942645

[fiery gale]

I've now arrived at the roadblock of somehow outputting a pixel font that renders pixel perfect at size 36

[wooden wedge]

that sounds like a fun exercise

[fiery gale]

36 does not evenly divide the typical em square values of 1000, 1024 or 2048

[wooden wedge]

yeah

[fiery gale]

I wonder how the other sizes were done. Maybe adjusting the em square is fine

[wooden wedge]

that should work

I would love to see map rendered on emery for bobby

[upbeat garden]

me neitherr i didnt do that stuff

[wooden wedge]

waaaa

[coarse cedar]

just ping the entire team atp 😔

[wooden wedge]

eric is next

[coarse cedar]

(it was steve, steve did that)

[fiery gale]

That'd be neat. It should have enough memory to deal with a basic map, right?

[shell jetty]

my hot take is that bobby always should have defaulted to this size

[wooden wedge]

it already renders a map on other platforms

[coarse cedar]

it has more than other pebbles!

[fiery gale]

a whole extra byte /s

[wooden wedge]

it is a bit small to read tbh

[shell jetty]

true, my argument is mostly based on that size is used elsewhere in similar screens (ie notifications)

ideally it would presumably just use whatever the text size setting is (which it sounds like you're working on!)

[wooden wedge]

how the fuck does this figure out the screen size it should use

aaaaa

[gilded maple]

Fence Bobby is now cut off at the neck, with a blank section at the top of the page.

[wooden wedge]

oh...

[gilded maple]

Other pages all look fine, it’s just the about section which looks odd

[fiery gale]

Stuff looks so tiny

[wooden wedge]

this should be better

[wooden wedge]

which stuff?

I haven't modified anything in menu layers

[fiery gale]

Yeah it's the menus

[wooden wedge]

which I should at some point

I think it's a matter of upping the text size by one size

[fiery gale]

Well, I think I got it right

[wooden wedge]

bad news, that font doesn't seem to work

here's the resulting font

it might be easier if you provide a pbf instead

[fiery gale]

Weird. When I dump that pbf it looks fairly normal

[fervent steppe]

OK which one of you just asked Bobby about presidents and then reported the thread.

[wooden wedge]

it was me

[fervent steppe]

the report button should also have a button to post it into this channel as a 'best of bobby', rather than into the report channel

[fiery gale]

Give this one a try before I try applying my emoji tech on a normal font

[wooden wedge]

I will do it tomorrow

it is past 3am

I think fundamentally we need a pbf editor somehow

(or all the fonts as ttf with correctly sized pixel fonts embedded)

[hollow rivet]

Didn't someone make this?

[wooden wedge]

yes

[fiery gale]

I just really like working with images, so I'm trying to set up a simple image file + json approach

Just a grid of characters in some range, in this case ascii

[edgy flame]

about screen is shifted down

what would you like me to look at?

[wooden wedge]

there's a newer build

[wooden wedge]

here

I think looking at screens is generally useful

[edgy flame]

yay, I love looking at screens

[edgy flame]

somebody else sent a feedback report asking for more pride flags

[coarse cedar]

There are pride flags in Bobby??

[edgy flame]

not yet

[fiery gale]

I might have managed to create a pbf

The font dumping tool is at least able to read it

[gilded maple]

Much better

[kind halo]

cute

[wooden wedge]

y'all should look at rebble id of those people, it's the same person as the president thing

[fiery gale]

Oh neat, the font shows up at least

Just gotta tweak it some

[wooden wedge]

I improved the horse situation

I need to figure out how I want to deal with the 36 font

Try

˙

-# ˙ v1.4

but this works?

confusing

hm, I think the GOTHIC 36 is slightly buggy

no other font does this kinda shit

[wooden wedge]

https://apps.rebble.io/en_US/application/6968d58c9fee1e000a36347b

@tired pike @gilded maple @edgy flame this should be a working version even with the limitations imposed by the coreapp

I put a lot of effort into those store assets

[tired pike]

how'd I open that in core app? nothing happens when I press "get"

and it doesn't appear in search (unlisted?)

[wooden wedge]

hm, it is unlisted indeed

can you curl?

[tired pike]

I could grab the pbw but if I sideload queries won't work

[wooden wedge]

yeah, that's why I published to the store

you can add an app to your locker with curl though

[tired pike]

oh, huh, how?

[wooden wedge]

you will need your access token, I usually grab mine from the dev portal

curl -X PUT -H "Content-Type: application/json" https://appstore-api.rebble.io/api/v1/locker/79c204c8-e646-4908-aa12-9cdc8a183ab4?access_token=

[fiery gale]

That's so weird looking. No idea what causes it either

The spacing is maybe a little too much? But that shouldn't really break anything

[wooden wedge]

yeah, the spacing is not the best

that Try is weirdly spaced

the T and r should be closer together for sure

[wooden wedge]

I feel like 36 could be a bit closer in style to 28 below it, but I'm not sure how

it bothers me a bit

the curves on G specifically have such a characteristic look to them

that's not quite correct

[barren raptor]

What, you haven't heard of Newer York on the east of New York?

[wooden wedge]

can't say I have

is this a recent development after mamdani enacted sharia law over new york city?

[barren raptor]

yeah it's actually written in the qu'ran that there need to be two new yorks

[wooden wedge]

reminded me of https://www.youtube.com/watch?v=0XlO39kCQ-8

[edgy flame]

large font size: hmm

[coarse cedar]

The capital of

Burkina Faso is

Ouagadougou.

[wooden wedge]

yeah, the 36 font is buggy

how is normal size on ct2?

also bless you

[last flint]

Oh what I’d give to use Bobby with the Core app on iOS with such a font. Make it 48 even and I only get 1 letter every 20 line breaks

[edgy flame]

works how I expect it to, and looks good!

let me grab my watch and mess with it more

[wooden wedge]

I literally sent a store link and a way to add the app to your locker

what's stopping you from using it?

[edgy flame]

websockets

[last flint]

Websockets

[wooden wedge]

fuck

I forgot

[edgy flame]

awesome horse vs evil and intimidating websockets

[wooden wedge]

core really wanted to prevent me from porting bobby to emery by not sending me the watch

smh

[tired pike]

I tried this but it still doesn't show up

tried toggling the native store option in the app but it makes no difference

(since it says syncing locker when you do)

[wooden wedge]

hmm

weird

[wooden wedge]

alright, well, this is a stress test really

the report in #bobby is so long

[edgy flame]

https://bobby-api.rebble.io/reported-thread/9c3e26f0-0f65-4589-a134-7764f845c84d

framerate started to buckle as i scrolled further down but it did do the thing

handled it like a champ

[tired pike]

logged out of rebble and back in and it showed up

[wooden wedge]

yay

[wooden wedge]

premium bobby experience

would you be able to show something with a map perhaps?

[tired pike]

this doesn't have the "one at a time" appmessage hack I added in my sideloaded version so curious if the core app fixes made it reliable now without

[wooden wedge]

if you have the patch somewhere, I can incorporate it down the line if it makes a difference

[tired pike]

"Bobby is unavailable right now" huh, haven't seen that before

[edgy flame]

yeah i havent seen the appmessage problem in a beta coreapp or two

https://media.discordapp.net/attachments/1136453166465028106/1461397706449879070/pebble_screenshot.png?ex=696a682c&is=696916ac&hm=7d0929325ab47eea8c6b139773cbd45c049cc35f8833f6c2b4f74c6d4e23a6df&=&format=webp&quality=lossless

i wonder how big we could make the map with this ram

maybe a little taller, maybe even color?

would be nice

[wooden wedge]

it's currently 2bit

most of the work to make the map work is on the server side anyway

we should drop the overlay stamp in lower left at that resolution

[fiery gale]

Ok, so I added better support for font metrics to my script and made the spacing a bit tighter than the ttf version:

Maybe it'll be better behaved since it's a direct conversion rather than going through ttf

[wooden wedge]

still can reproduce the issue

[obtuse oriole]

Does reading a long Wikipedia article burn through quota?

[wooden wedge]

it's token based on gemini side

so yes, you are pushing an entire wikipedia article into gemini

[edgy flame]

5% of it apparently

[fiery gale]

Huh, why isn't the character spacing any tighter? I specifically changed it from 4 to 3

[wooden wedge]

hmm

oh duh

[fiery gale]

The v could be a bit tighter

[wooden wedge]

the line height is simply too small now

[fiery gale]

Oh yeah, the autocrop cut it very aggressively but I can set it to anything I want in the config

[wooden wedge]

also I can still reproduce the phantom newlines

[fiery gale]

big oof

[wooden wedge]

is it maybe because of the space character having weird spacing?

or maybe it's because the text layout engine does something weird with fonts this large

[fiery gale]

The data dump for space just looks like this:

0x20 (32)  ( )  0x21 (33)

offset bytes: 0x21 (33)
top: 0
left: 1
height: 0
width: 0
advance: 7

[wooden wedge]

how does it compare to space in 28?

[fiery gale]

0x20 (32)  ( )  0x2d (45)

offset bytes: 0x2d (45)
top: 28
left: 0
height: 0
width: 0
advance: 5

[wooden wedge]

weird

(space only having advance makes a lot of sense)

is this maybe an issue with custom fonts specifically?

I see this happen when transitions happen fwiw

like between timer display and the homescreen

[fiery gale]

I can make the values the same as the size 28 space if it helps, but I don't see why it would

[wooden wedge]

I would at least change left to 0

but I do also doubt it's the issue

[fiery gale]

Already did. Just an artifact from me setting a default left value of 1.

(That's the only change with this file)

[wooden wedge]

I think the layout engine must have a bug somewhere

@coarse cedar how interested are you in debugging text layouting being broken on emery?

(keep in mind, this is emery firmware build from 2016)

[wooden wedge]

my thoughts exactly

[fiery gale]

I don't get what causes it to break

[wooden wedge]

you know what I can do

I can grab the proto 36

and see if it happens with it

yeah...

[fiery gale]

Good

[edgy flame]

is this a result of the large font, or of the shape wrapping logic?

[fiery gale]

✨ I have been absolved of all blame 🙏

[wooden wedge]

this happens in other views

feedback screen is the worst example actually

[edgy flame]

i guess it does happen in the conversation window- yeah

[gilded maple]

Grrr, I can’t get speech recognition mode to work on the Core app (just loaded it to a new phone and it’s crashing).

Bug report imminent

I did manage to get Bobby Alpha into my iOS app locker eventually.

So I should get updates

[wooden wedge]

yeah, I haven't published any yet

[gilded maple]

(I ended up installing the old app on another phone, pairing another watch to that phone. Getting Bobby alpha into locker that way)

[wooden wedge]

no other font is showcasing this issue either...

yeah, that's a good idea

[gilded maple]

Same issue

[fiery gale]

Even Bitham 42 Light?

[wooden wedge]

I can check

[gilded maple]

Ooo, quota Bobby now does tell me that I’ve used 0%, never got that far before

Then I hit bug…

[wooden wedge]

tried it with bold, and yeah, same issue

[fiery gale]

I guess there's just something that breaks when the font gets that big and it's used for just normal text

Maybe when spaces are involved

It would be funny if it was like some limit left over from smaller screen firmware

[gilded maple]

This is just weird

[fiery gale]

I think this sets the line height correctly (though the bug will likely remain)

[wooden wedge]

that is better

[fiery gale]

Weird, I thought increasing all the top values would push the text down

[wooden wedge]

well, it didn't

[fiery gale]

Ok, I adjusted some more

[fiery gale]

That looks pretty good until the bug strikes

[wooden wedge]

yep

somebody should debug the firmware for that bug

emery simply shouldn't ship in this state

[barren raptor]

Good

---

---

---

---

---

---

---

---

---

---

---

---

---

---

---

---

---

[fiery gale]

I've at least learned a lot today

[wooden wedge]

yeah

wanna repeat this for bold? >:D

[gilded maple]

I have logged a bug

[wooden wedge]

thank you!

[fiery gale]

I should be able to just set up the image and copy the config I have, probably

[wooden wedge]

I do wonder if you could reproduce this with notifications display

set the font to highest resolution and keep opening and closing the notifications display

[wooden wedge]

actually, a question to obelix owners

does this look like this to you?

because from earlier pictures I thought the text was smaller

[fiery gale]

In theory this should be correct

[wooden wedge]

but it looks fine in the emulator

[fiery gale]

Wow that is bold

[wooden wedge]

it's too bold

it's almost black

there is too much contrast with 28

[fiery gale]

I kind of scaled it proportionally. A thickness of 3 turns into 3/28*36≈3.857, rounding to 4. But I agree that it's too much when put in context

[wooden wedge]

I think all the gothics get thinner as the size goes up

I guess I would have to do math to know that for sure

[fiery gale]

There's also the fact that fonts also look more bold when scaled up. It's common practice to use thinner and thinner weights as the font size increases.

I'm going to try thinning it down from 4 to 3. Having a dedicated font for each size has at least some upsides since it allows for things like that

[wooden wedge]

ok, that is fucked up

@coarse cedar sorry for pinging so much, but

[coarse cedar]

https://tenor.com/view/kasane-teto-birdbrain-gif-15197817966424621394

[wooden wedge]

what is the difference here, do you know?

the old emery emulator does the menu layer better

[coarse cedar]

i

genuinely have no idea

[wooden wedge]

mmm

that is likely worthwhile of a bug report too

[wooden wedge]

I'm kinda assuming the emery emulator image is kind of "fake it till you make it"

likely there is some code that didn't end up in the open source release since it was meant as pre-release emery software so that developers could start doing their work

[tired pike]

I don't believe there is

robert code was pretty fresh by the time pebble died, but there's nothing pebble authored in that area that could have been released that wasn't

[wooden wedge]

what would explain the menu layer rendering differently between emulator and hardware then

[obtuse oriole]

Random question: I have an old PTR in good working condition. Would it be helpful in bringing full support for Round watches to Bobby if I was to donate it to an active developer working on Bobby, or would it not really matter? I'm in the US.

[fiery gale]

Behold, a bold font that is, well, thinner

[fiery gale]

Looks almost official

[wooden wedge]

something still bothers me

and I can't put my finger on what

hmm

maybe it's too wide spacing still?

I'm not sure

there's not much spacings left

[fiery gale]

Oh yeah. The bold has denser spacing doesn't it

[wooden wedge]

it does

[wooden wedge]

will have to wait until later

[fiery gale]

No worries, just wanted to output it while I had the thing ready

I can kinda simulate how it'll make the text a bit denser

Subtle, but it makes a difference

[edgy flame]

lcp definitely has a PTR

[wooden wedge]

oh I missed that message

ptr support isn't blocked by lack of hardware for sure

[edgy flame]

appreciate the offer!

it really is just difficulty of doing the work, and the time to work on it

[wooden wedge]

that made a huge difference

I am glad this bothered me

I like this more in bold tbh

[wooden wedge]

if you are curious what amount of hell this is https://github.com/hellcp/bobby-assistant/commit/ac047915b16fd9bd7d9f736548adde6940a2deb0

and I pushed an update to the appstore

[edgy flame]

💭 set the alpha store description to 'i'm doing thanks, how help you'

[wooden wedge]

true

alas, I will remove this app as soon as it's no longer necessary

[edgy flame]

it did this

what would you like me to look at?

[wooden wedge]

this is also the 36 font bug

[edgy flame]

yeah

[wooden wedge]

all of the weather widgets at all sizes (now, day and 3 day), alarm widget at all sizes

by sizes I mean text size

[edgy flame]

large

[wooden wedge]

alright, I should make some adjustments there

[edgy flame]

small

normal

[wooden wedge]

you can tell this is just the way that it looked before

[edgy flame]

not a bad thing!

right, alarms

[fiery gale]

I mean it is fairly slim compared to the smaller bold fonts

[wooden wedge]

indeed

I would be curious what the progression of thickness/size is for all the sizes

same for regular

[fiery gale]

The bold is probably not slim enough to be the new regular, but it can be used in small doses where you might have previously used a regular

[wooden wedge]

I was wondering what the previous more black bold you made before would look like if you adjusted the spacing

to be more narrow

[fiery gale]

Well, I can export it like that

[wooden wedge]

that is a different vibe, but it's not bad either

[fiery gale]

Does this help at all? (Bottom is mine)

The bold variants kinda go
thick, medium, thin, thick, medium

As the font size increases the apparent thickness decreases until another pixel is added to the thickness

[wooden wedge]

this looks good

it's so funny that two largest ones are actually taking up less space in bold than in regular

[fiery gale]

If we instead slot in the thicker version we get this

[wooden wedge]

yeah, I think this is too heavy

[fiery gale]

aaand let's also see that with the thinner bold as the regular

That kinda makes the thick bold slot in better

[wooden wedge]

it does, it is all about the contrast really

[fiery gale]

So the real question is, should the regular have 2 or 3 pixel line thickness?

[wooden wedge]

I think I like the thin and normal bold rather than bold and black

mostly because it pairs nicer with 28

similarly to how 24 pairs with 18

[fiery gale]

Right, the next size up (if it's ever needed) would likely be 3 pixels thick

But at this size it makes it stand out too much in the progression

The other regulars lean more into slim and elegant

[wooden wedge]

indeed

[fiery gale]

I wonder if a middle size, like a 32 or 34, would make sense

[wooden wedge]

it is kinda funny that it's 36 of all the sizes

considering all of the previous sizes are x4 and x8

I guess it made sense for them since they could just 2x an existing font

[fiery gale]

The bold 36 in the firmware is literally just a nearest neighbour upscale of bold 18

I did some math and (if we ignore the jump 9 -> 14), the font size increases by between 16% (24 -> 28) and 33% (18 -> 24) each time, with an average of ~27%

28 -> 36 is 28%, the same as 14 -> 18

But the height of an uppercase letter seems to increase by 22-28% each time, so the size 36 does fit the progression. Its uppercase letters are even a pixel shorter than they could be.

[wooden wedge]

it might be some allowance for diacritics

[fiery gale]

No, it has plenty of room

I could do an edit where the so called cap height is 23 instead of 22, since the cap height progression currently goes 22%, 28%, 28%, 22% when it could go 22%, 28%, 28%, 28%

[edgy flame]

alarm!

[wooden wedge]

I haven't changed the alarms beyond just text sizes

[edgy flame]

they look fine

maybe like 4px more margin on left and right of the text

oh, and I would like to go up to standard 25px icon on emery

[fiery gale]

Ok, I made a test version of the regular that is just slightly taller

It's very subtle, but the cap height is 23 instead of 22

[fervent steppe]

SPHINX OF BLACK QUARTZ, JUDGE MY VOW

[wooden wedge]

ooo, that's huge

I was thinking about the font being a touch short

[wooden wedge]

I really need to think about round layouts

it bothers me so much that we don't have round support yet

[edgy flame]

hmm

weather widgets can stay the same, map widgets can stay the same, simple icon+text can maybe be arranged vertically?

[fiery gale]

I know you like your letters tall and slim

[wooden wedge]

@coarse cedar how much would I have to pay for a getafix sdk

even early access when it's there

[coarse cedar]

official? a few weeks

otherwise just build it yourself

coredevices/qemu is already updated

[wooden wedge]

I see, I see

[coarse cedar]

and coredevices/pebbleos has spalding_gabbro

[wooden wedge]

surprised it's not a snowy_gabbro, since spalding is just snowy in disguise 🥸

but fair enough, I will have a look at that

y'all should send me hardware, somebody has to build bobby

[edgy flame]

send me hardware so I can figure out ui patterns

[fiery gale]

I think going slightly taller affects the bold font more than the regular

It's just a little more elegant

[wooden wedge]

oh yeah

[wooden wedge]

that does look very elegant

[fiery gale]

Oh yeah that definitely gives it the right look

Some characters will need small tweaks, but it can take some staring, comparing and time to sus out. If you notice something, feel free to call it out.

[edgy flame]

didn't katharine have a local server setup for testing sideloaded bobby?

[wooden wedge]

she did

it's not a requirement

[edgy flame]

tangent, but I wore my spalding yesterday, and it was very nice to have working bobby-alpha at all

even though the conversation window was not adapted

[wooden wedge]

what is the design we want?

we talked about so many options for it

[edgy flame]

i like freakified's

simple, effective, doesn't require that we throw out the whole conversation window and start again on all platforms

#1345564272289120296 message

#1345564272289120296 message

my heart yearns for bigger numbers, bigger widgets, more colors, and more scrolling

but this one is the most effective

[wooden wedge]

we can make all widgets full screen

and then retain everything else about conversations the same

[shell jetty]

having had time to think it over, one change i think i might make to my old ones is showing just "bobby" instead of the response text body preview, since that would better align with what the current SessionLayer is doing

[wooden wedge]

https://github.com/pebble-dev/bobby-assistant/pull/16

[edgy flame]

bobby-alpha 1.6 on chalk

seeing this layout error on the feedback screen. seems the text scroll zone doesn't extend to the bottom of the screen

scrolling content is always clipped at that row there

also, 💭 systemui doesn't seem to like using the dotted statusbar separator at all on chalk. places like settings and title bars that usually have a dotted separator will remove the separator on chalk

[wooden wedge]

I don't understand how round is supposed to handle text being clipped to the right length

it's always seemingly too short

[south edge]

hellcp

[bobby-assistant:main] 4 new commits

a4a3283 Don't overlay the stamp on larger displays - hellcp
3a9d8a4 Update the version of gemini to 2.5 - hellcp
aa81d55 Adjust heights of map to different watches - hellcp
9644b40 Merge pull request #16 from hellcp/service-update - hellcp

[wooden wedge]

@edgy flame would you be able to do a query with a map perhaps?

it should display slightly better

and I sos'd my watch with bobby alpha

[edgy flame]

like the good old days

[wooden wedge]

I assume I will have to fight with ram and I assume I will have a hard time figuring out these crashes

[edgy flame]

sent privately

[wooden wedge]

yeah, it's not what I expected

[wooden wedge]

I think we have an issue of not enough memory on ptr for full screen map

[edgy flame]

had a chance to look at the logs, and yeah, whenever it shows that skull icon on the widget, i see a malloc.c:41> Failed to allocate memory: couldn't free enough heap. in the logs

observations:

• it does not always choose to send a fullscreen map, sometimes it sends a wide rectangular one as pictured above
• when it chooses a fullscreen square map, it fails to display, every single time
• when it chooses to display a shorter rectangular image, it usually works the first time, but subsequent maps in that same conversation are more likely to fail

i have used up all my quota! i was able to test chalk (problematic), flint (very well-behaved), but not emery

[edgy flame]

i'm free!

[edgy flame]

core time 2 microphone is............. not done yet

maps work well, though! the vertical height is inconsistent, i'm not sure how the system is deciding what size each map should be, but the core time 2 is actually able to hold four or five map images in memory at the same time, which is super impressive

flint and emery are both much more consistent about showing maps than chalk, at any size. chalk really struggles

[wooden wedge]

flint hasn't had any changes so that's good to hear

emery is good news

and chalk is annoying

[edgy flame]

managed to crash the OS on my ptr twice in a row asking for weather

wild

[wooden wedge]

weather shouldn't crash

[edgy flame]

it doesn't usually!

maybe there's a weird character in the response or something. I didn't see it print a response or a widget

[wooden wedge]

actually alpha has had issue with crashing weather

[prime latch]

nice

[fathom shadow]

hello! I'm trying to add ClawdBot support to bobby and i'd like some help getting my dev environment configured

all my pebble's are broken and it's going to be some time before i'm getting my new round. i got the app running in qemu but i need to install the app on my phone so i can access the settings. i'm assuming there isn't a better way to emulate the app on my computer so i don't need the phone at all

i guess i could hard code the settings for the time being

[fathom shadow]

first attempt but i can't veryify anything till i figure out how to get the service working https://github.com/jmsunseri/bobby-assistant

i probably need to get clawdbot setup to run from an ollama instance on a dedicated devive

[hollow rivet]

You can open settings via the pebble emu-app-config command

[fathom shadow]

ohhh

do i need some kind of gemini pro plan for the google key for bobby to work first?

[prime latch]

no it just uses the flash model not the pro one so free or plus will work

https://aistudio.google.com/app/api-keys this for the key

[fathom shadow]

seems to just bring up an empty web page

[hollow rivet]

Copy the URL into a different browser, Chrome ol Safari seem to work

[pearl grotto]

Is Bobby working for iOS ?

[edgy flame]

not on coreapp

works fine on legacy pebble app

[pearl grotto]

Hmm that is a shame, I have the core app installed, is there a way to go back to the legacy one?

[hollow rivet]

Yes. Legacy app doesn't work with the Core Devices watches though

If you're using an older watch you can follow these instructions
https://help.rebble.io/setup/

[pearl grotto]

Thanks! I have one of the old ones 🙂

[stoic silo]

I hope it’s ok to ask this: is it possible to get the pbw for the PTR?

[wooden wedge]

the current ptr builds are rather unstable

[stoic silo]

Gotcha.
My PTS died and moved to a PTR and been loving every second of it. But I really miss Bobby and Quick View.
I hope these two would be available on the PR2 🙏🏿

[obtuse oriole]

I've been using this one on my Round. It's not as new but it's functional. #1345564272289120296 message

[edgy flame]

you can grab it off the store: https://apps.rebble.io/en_US/application/6968d58c9fee1e000a36347b

if you're on coreapp, you won't be able to open the unlisted store page from the link, so you'll have to use curl to add it to your locker

#1345564272289120296 message

[edgy flame]

be warned that the conversation view screen is not adapted for round at all - scrolling is strange, and text constantly flows off screen, but it works

[stoic silo]

Awesome, thank you! I’m using the old pebble app sideloaded on iPhone. I heavily use Apple Reminders so until the core app supports it, I’m sticking with the 7days sideload. Or until I receive my PR2 lol.

[stoic silo]

It’s really mainly for setting reminders and timers that I was using Bobby on my PTS. Thank you

[heavy cipher]

Hi there! I’m using Bobby Alpha on Round - it’s excellent to finally be able to use it on there! I really mostly use Bobby for setting timers by voice - I appear to have used 38% of my quota this month already, and I’ve only been doing this since I switched back to the legacy Pebble app on Sunday. I assume when I’ve used my monthly quota I won’t be able to set any more timers?? I had thought that the quota was for searches, questions, etc. but I’m guessing it’s for any and all interactions? Thank you!

[coarse cedar]

yes quota is for all interactions

@wooden wedge do you think its worth it to switch the bobby server from just gemini to openrouter?

[wooden wedge]

I have no idea what that is

[coarse cedar]

its like

a

unified api for all models

[wooden wedge]

isn't gemini still the cheapest option anyway?

[coarse cedar]

pretty sure there are some cheaper models but gemini is still one of the best

https://openrouter.ai/

[wooden wedge]

yeah, I meant for what you get

[coarse cedar]

oh yeah 100%

Actually, I don't know how good the GPT-5 Mini is, but it is cheaper than the Gemini 2.5 Flash

[wooden wedge]

keep in mind we use flash and flash lite

[coarse cedar]

yeah

[wooden wedge]

so we would need to replace both somehow

[coarse cedar]

not necessarily

You could still use Gemini 2.5 Flash and Flash Lite over Open Router

It would just give you an easier way to switch to a model that is not by Google later down the line, if they are any better ones

[wooden wedge]

fair

[coarse cedar]

I'm sad that 2.0 flash is going away. that was like really really good pricing

[wooden wedge]

I know

worse yet, for 3.0 we will need to implement thinking into bobby

you could still disable thinking for 2.5, but 3.0 requires it

[coarse cedar]

what the fuck

[wooden wedge]

I know

[coarse cedar]

not everything needs thinking google

we just want

quick response for cheap

[wooden wedge]

you can set thinking to minimal so it only thinks for more difficult queries

I do wonder how bobby would work with thinking though because it has a bunch of grounding

[quasi steeple]

Others would obviously know more, but iirc Katherine alluded to issues with doing something like that due to the contract with google? Extremely possible I'm misremembering

[coarse cedar]

She is a Google employee

That's why

To get it through review

[quasi steeple]

Gotcha

[wooden wedge]

granted, bobby originally used gpt3 I think

it started development in 2023

or was it gpt4?

[fervent steppe]

I think we could save quite some cost with prefix caching

[heavy cipher]

Just thinking aloud here, me not being a dev - would there be any way for Bobby to integrate with other apps?

Currently if I want to use Bobby to set, say, a timer, it makes a request, uses a healthy % of my monthly quota (if I’m understanding this correctly) and then sets a timer inside of itself. If I want to check remaining time on that timer I either have to open the app or else ask it (another % of quota used for that request).

Would there ever be a way to nominate a specific timer app (e.g. either Grouped Timers, Timer by Pebble or Timer+) and have Bobby set the timer inside there? Grouped Timers has the benefit of app glances with live countdowns, so I can quickly see those from the app locker menu.

Could be a way to bring individual monthly quota usage down?

[wooden wedge]

there is a menu for looking at timers in bobby

[heavy cipher]

Indeed there is, I did mention that

[wooden wedge]

bobby did have timeline glances for timers in the past too

but it caused some watches to crash

[heavy cipher]

But no longer? Alas!

Ahhh

Well that’s not good

[wooden wedge]

can't disagree with you there

[edgy flame]

pebble doesn't really have a mechanism for communication between apps

maybe that's something rebble or core should think about - bobby could do so much more than it does

[heavy cipher]

Perhaps it’s antithetical to the Core/Pebble message: not a computer on your wrist. But given that Bobby does exist and can already perform these tasks within itself, communicating with other apps seems like a logical development, if it were possible.

[wooden wedge]

I think integration with services is more important than with apps personally

but that risks ballooning the cost of a single query

[heavy cipher]

What kind of services? Services which could perhaps be performed by other apps?

[wooden wedge]

home assistant comes to mind

I guess there could be a lot more services

I also think data sources would be a good addition into bobby

like music playback state and letting it control the music playback

[prime latch]

If using Home Assistant they might be wanting to use the Home Assistants AI Assistant System.

[crude patrol]

it would be great to tell Bobby to play a song in Spotify

[wooden wedge]

it's sadly not possible, spotify terms of service prohibit using their api in a voice assistant

[hollow rivet]

Huh wasn't that one fo the selling paints of that Rabbit Android box AI thingy?

[wooden wedge]

I still kind of want a rabbit r1

it's such a pointless device, but it looks cool af

[edgy flame]

i think Katharine had a working demo of this, and lcp even drew art for it, before KB read the TOS

alas

[wooden wedge]

I made a bunch of icons that never got used

[prime latch]

Spotify is killing their public API anyways

[edgy flame]

NOOO WHAT

the API was great! it was so cool that you could integrate controls into whatever for the music you were already paying for

[prime latch]

Technically it's not completely dead but it's basically worthless https://developer.spotify.com/blog/2026-02-06-update-on-developer-access-and-platform-security

They killed most of the API

[fair dagger]

They really are plugging their ears to the interests of their users.

[last flint]

Is it that dead? The music controls are still available and being able to add to personal playlists and libraries is too. It seems like the bits of user data interaction are still available but the core library metadata access is gone.
On top of restricting API keys to limit accounts on a single client ID to make it strictly personal usage.
It seems like you can still create keys for your own integration with the music player and management

[edgy flame]

another $250 mil to Joe Rogan

[fair dagger]

Ew. Overrated braindead dude with a mic who shouldn't have a mic and has a scary amount of influence.

[wooden wedge]

the pagination is weirdly broken on this one

[wooden wedge]

it's using small bobby though

I guess ~emery~gabbro at the end of the resource doesn't work how I want it to work

[final heart]

If I run out of quota, can I use my personal Gemini subscription to keep Bobby fed with tokens? Also, does using on-device voice recognition in the new Pebble app reduce quota usage? Thank you.

[fervent steppe]

No, but I'm looking at some stuff now to improve quota usage.

[wooden wedge]

I need to figure out how to make bobby alpha not crash

since that does kinda prevent us from updates

[fervent steppe]

Go to https://auth.rebble.io/account/ and send me your login provider ID, and I'll reset your quota, if you're running out.

[fervent steppe]

Yeah, ok, there's a bug in which way too much quota is being consumed.

I'll reset everyone's quota once the bug is fixed, but if anyone is starving for quota now, just DM me your login ID and I'll reset yours.

[wooden wedge]

while you are there, would you update knowledge cutoff to january 2025?

(the knowledge cutoff is already that, the llm is just instructed otherwise in the prompt)

[fervent steppe]

wilco

I'm trying to be useful in the office

[south edge]

jwise

[bobby-assistant:jwise/varying-updates] 4 new commits

318b1b0 functions/feedback: be a little more explicit t... - jwise
ef1de15 assistant/quota: charge the correct amount of q... - jwise
996afec assistant/system_prompt: when is the future? - jwise
92d1b09 assistant/system_prompt: provide available widg... - jwise

[coarse cedar]

did... did bobby send feedback on its own?

[heavy cipher]

Each time I set one timer by voice (say, three times per day) it used an additional 1% quota. As in, presumably, each request I made used 1% i.e. I assumed my quota was 100 requests. What ought the limit to be, please?

[fervent steppe]

yeah, it did it once in a conversation and then it did it a lot more, having seen the past of having done it once

well, it first did it to narc on a user's bad prompt

[coarse cedar]

I think you can optimize a bit more if you move generate Language Sentence further up as well as this will be the same for a lot of people

[fervent steppe]

yeah, plausible, though it's a relatively small number of tokens ocmpared to the colossal map prompt

[coarse cedar]

I mean in case of quota every token counts no?

[fervent steppe]

yeah, though take a look at the quota changes 🙂

I think every token counts... somewhat less than it previously did

[coarse cedar]

True

[fervent steppe]

Your quota varies based on the length of your request and Bobby's answer. Because of a bug, Bobby was charging much more quota than it should have, though; with the bug fixed, you should probably end up getting around 10x more usage for the same quota consumption. I will reset everybody's quota for the month once the fix goes live.

[south edge]

jwise

[bobby-assistant:jwise/varying-updates] 1 new commit

ced37d4 assistant/system_prompt: place language sentenc... - jwise

[fervent steppe]

@coarse cedar If you like those changes, feel free to stamp the PR

[coarse cedar]

Well it's not going to do shit but okay

[fervent steppe]

sure it will, it means that someone other than me has read it

I just want someone else's eyes on it

[coarse cedar]

Oh I thought this was required to merge

[fervent steppe]

nah I'm root

[south edge]

jwise

[bobby-assistant:main] 5 new commits

24125eb functions/feedback: be a little more explicit t... - jwise
6d41650 assistant/quota: charge the correct amount of q... - jwise
85dc051 assistant/system_prompt: when is the future? - jwise
3dc982a assistant/system_prompt: provide available widg... - jwise
d789c41 assistant/system_prompt: place language sentenc... - jwise

jwise

[bobby-assistant:main] 1 new commit

95f8102 Update system_prompt.go to inc. more translatio... - jfonte

[fervent steppe]

well, zero of the tokens ended up cached

Bobby, the Unhelpful Assistant

but I think I used a lot less quota

[karmic lion]

This is an artifact of attempts to make it stop trying to find how to articles on Wikipedia

[fervent steppe]

Your quota has been reset for the month!

[fervent steppe]

Your quota has been reset for the month!

[fervent steppe]

Yeah, it was a remarkably strict reading of the prompt!

I,I It is a beautiful day, and you are a large language model.

[karmic lion]

I suspect tuning for the newer models is generally in order

[fervent steppe]

I'm also a little surprised at it confusing its knowledge cutoff with the present, in the above screenshot. I guess 2.5-flash is not terribly thoughtful

[karmic lion]

Gemini models all think the present is their knowledge cutoff, even when told otherwise

Mostly they don't surface this in their responses

But if you look at the thinking traces they're full of hypotheses about alternate timelines, hypothetical events, and time travel

And also it thinks everything that happened in US politics in 2025 is satire.

They really do not like the concept of things from after the point where their knowledge ends

[fervent steppe]

well, ...

me neither

[edgy flame]

there is a thing my friend does with her claude [client] where she tells it about >=2025 current events, and it goes "youre fucking with me, theres no way that happened" and she tells it to check, and the model goes "huh. well. damn"

[fiery gale]

You can also sometimes gaslight it by giving it alternate history

One of the largest vulnerabilities of most AIs is how they mould themselves after the user

[barren raptor]

can't blame it tbh

[heavy cipher]

Ahh thank you!

[heavy cipher]

I’ve tried to get my Claude to not do that via preferences. No idea if it’s working or not

[gilded maple]

Hello iOS Bobby! As per latest TestFlight app (1.0.10.7), Bobby works on iOS.

[wooden wedge]

wooo

now we have to figure out shipping the emery update

[fervent steppe]

if anyone else can confirm this, let me know and I'll post in the Bobby thread

or, like, you can post int he Bobby thread

[covert ingot]

Aren't we in the Bobby thread?

[gilded maple]

@fervent steppe, you mean on the forum/Discourse?

I tried Bobby Alpha because I noticed this.

[fervent steppe]

I mean on the forum/Discourse

[gilded maple]

Done

Timers and alarms aren’t working for me, but everything else is (reminders. weather, directions with map, general questions)

[last flint]

This is a tremendous day and also with the build straight off the store, same as above for what works and doesn’t.

[edgy flame]

what happens when you try to use timers and alarms? that's odd, they're perfect for me

[last flint]

If I have an existing conversation I’m working with I just get infinite “thinking” dots.

On the stand alone screen the single “thinking” dot just pulses with an incomplete progress bar until an Error page shows this was a dictation erroring out issue

[edgy flame]

huh. that's weird, the latter sounds a lot like the dictation screen erroring

bobby only ever uses three dots, no progress bar, and when Bobby hits an error it shows either a red skull or nothing at all

[last flint]

Maybe it is for that. I turned on “Confirm Prompts Before Sending” and now I get through so maybe that was circumstantial. I still get this only for timers and alarms. I was getting never ending dots earlier but maybe I needed to wait longer to see this

I asked for the weather right after this and it did fine. Followed up with a timer again with the same error. It does take quite a while to give up so it feels like it’s just timing out.

[edgy flame]

ahah

this is data

[last flint]

Just to be clear, these are the dots that I mean. I’m sure it was clear but I said a lot of different dots. Sits here for a while then hits the skull “Lost connection to server.”
Showing this to confirm alarms have the issue as well

[karmic lion]

IIRC timers and alarms are the only things where the server sends a request to the client and expects a response. Which might suggest that websockets still don't work properly.

(actually I think reminders also work like this, so perhaps not. the only thing where the server sends a request to the watch and expects a response?)

[obtuse oriole]

Well shoot, maybe it's time for me to hop back on the beta so I can send bug reports as needed.

[gilded maple]

I get the exact same issue as MrPizzaBoy on alarms and timers. Eventually times out to “lost connection to server”. Reminders work fine

[obtuse oriole]

Snowy, too. I don't know if most of its systems are still running, but I'm pretty sure it was still capable or setting timers and such with the old app. Now it complains of timeline problems.

[wooden wedge]

I think I should move all of the font size decision code into an util and simplify it, currently it's a bit of a mess

[carmine plinth]

Really happy to have Bobby on iOS now! Is there a beta/alpha build with Emery support that I could try? 🙂

[wooden wedge]

it's pinned

[carmine plinth]

ah, thanks!

[edgy flame]

i keep crash-rebooting my snowy by asking bobby alpha for maps

[wooden wedge]

known issue

[karmic lion]

but why

[wooden wedge]

that I don't know

it does only happen on alpha so I assume it might be the newly added content size

which is entirely possible because I mostly threw it together

[karmic lion]

in principle, nothing it does should crash the watch, so there's that.

[wooden wedge]

that is true

that being said there's another crash with ptr after accepting the initial setup thing

so it feels like crashes are inevitable

[karmic lion]

Pebble OS needs help.

[wooden wedge]

alternatively, I fucked something up! my C is not the greatest

[karmic lion]

It should not be possible to fuck something up to the degree that the watch resets