i knew of each of these things individually:

• this iphone holder https://www.printables.com/model/992305-iphone-12-mini-standby-mode-dock-watch-magsafe
• this landline looking phone that plugs into your headphone jack (rip) https://www.amazon.com/dp/B08T91TGJH
• chatgpt has a realtime advanced voice api thats much better than just stt+llm+tts using livekit over webrtc https://livekit.io/ https://platform.openai.com/docs/guides/realtime#connect-with-webrtc
• ios has Live Activites which let you send realtime data to the lockscreen, i know because doordash uses it for a neat ux https://developer.apple.com/design/human-interface-guidelines/live-activities
• a saint on github already implemented the openai webrtc connection in swift 5 so i dont have to do any work https://github.com/PallavAg/VoiceModeWebRTCSwift
• imessage bubbles for good measure https://github.com/GetStream/SwiftUIMessagesUIClone
• ios has built in speech recognition. this will work for a crude, customizable wake word that works about 60% of the time and drains your battery. but c’est la vie baby its got magsafe to charge in standby mode so we chillin https://developer.apple.com/documentation/speech/https://developer.apple.com/documentation/speech/

and now we have everything to make a custom cradle to hold the handset and an iphone to make a phone that talks to… no one. well, its chatgpt, but thats not much better, considering the point of a telephone is to connect people with each other.

first was the cradle, luckily the original author included the fusion 360 source so i just removed the extra charger and kinda re-shaped it until it held the handset correctly. I actually 3D scanned the handset, but it ended up being a lot of work and I had to tweak the dimensions anyway despite it being quite accurate.

now onto the software. always plugged in because its a spare 12 mini thats too small for me to use nowadays but will now live on my desk instead of in a drawer. so the voice transcription battery drain shouldnt be an issue as long as the magsafe puck provides enough power. i like the mophie one because it draws less current than the original v1 apple puck i have, so i can plug it into a computer’s usb port if i want, which you can’t with the oem, and it still charges continuously to 100% even with the constant listening drain. The battery was already in very poor health, so i wasnt too concerned.

i experimented with different connection noises to distract from the time it takes to connect after saying the wake word. i tried dial-up. it felt insincere, as i dont remember much of it myself. i remember skype tho (rip) so i used the old old skype call init sound from internet archive https://archive.org/details/SkypeSounds/CALL_INIT.WAV

the first person i had try it who wasnt me immediately asked it for the weather. it can’t do that, one moment please. add weatherkit and corelocation, pay apple for developer nonsense, add openai client side tool calling, bingo bango bongo we got weather.

at this point the contentview had all of the non-webrtc logic and was very long, along with the frankenstein of the webrtcmanager i originally stole from PallavAg, so i refactored everything into actual features and now i can actually find things and add more tools. i probably wont though, because this isnt alexa. the user was confused.

now the LLLandline/Hotline sits on my desk and answers the same 3 neovim questions every day becuase my brain is cooked

the app repo is private for now. but don’t worry, it’s not very good anyway.

edit: my phone that i used for hotline was stolen, so the cradle is kinda useless now. i’ll update this post with images if i ever get another 12 mini to put in it. rip hotline 2025-2025