# I made a Spider-Man Android launcher with Claude and got REALLY carried away — retained record

> Retained reddit record from reddit:ClaudeCode.

Canonical: https://highsignal.app/data/records/5ea887c4fe38f7b0

[← sources](https://highsignal.app/data)


source: [reddit](https://highsignal.app/data/reddit) (reddit:ClaudeCode)  published 2026-10-21 04:11 ingested 2026-10-22 02:42

[original source ↗](https://www.reddit.com/r/ClaudeCode/comments/1xbdaov/i_made_a_spiderman_android_launcher_with_claude/)

## retained text



```
[Sick ass music sampling the cartoon by CZARFACE and DOOM](https://www.youtube.com/watch?v=mUanYFSi6Wk)

I'm about 45 days into this, a claude labor of love. Resisting web developer jokes.

First thing off the bat, I made this for myself and it isn't available anywhere nor do I think Sony (or Disney?) would be happy with my open sourcing it. It doesn't even have a settings page right now cause I just made everything to my liking. I would love to find some way to publish some version of this though. It's been a tinkering project that I simply want to show off. I hope you find it as cool as I do.

I'm a software engineer web/mobile, but I literally don't know Kotlin at all. As I've needed to use Claude more and more for my job, then playing with it on my own time, I've found Claude to be a blast for tinkering with unimportant things, cause I can focus my efforts on exactly the dumb things I want and stay high level where I don't.

The layers to this project as I unraveled them after playing Marvel's Spider-Man 2 were basically:

1. Man it would be cool if my phone were Spider-Man themed
2. I wonder if Claude could figure out an android launcher or a live wallpaper
3. It seems like android lets an app do basically anything given the permissions? let's see how deep this rabbit hole can take us
4. I've also seen apps claim a voice assistant role on Android maybe we can do like a jarvis guy
5. Huh. For my actual job I had to give Claude remote adb access to a phone. This is kinda crazy

---

This was a fun tinkering project because in my mind there's a few separate pieces both in theme and implementation, and I dive into whichever feels more fun on a given day.

## A. Spidey gimmicks

A web as a launcher. Visually so fuckin fun. I could drag this web around for hours. Rogue themes, multiverse touches, inject spider-man gimmicks into my veins. I spent days perfecting the little thwip web strand that yanks up the app drawer. 

One day I was trying to optimize something and said fuck this let's make his head a f i s h b o w l

## B. My UI/UX experimenting board

The guiding principle here is that gimmicks get tiring if they don't contribute anything, let alone if they get in the way. I tried to explore novel UX concepts with this. 

I've never used androids "shortcuts" nor have I ever scrolled alphabetically through my app drawer to find an app. So I designed around that instead of recreating every other app drawer in existence.

I've always liked the ideas of gestures needing less precision than taps too. Removing the need for close aim and allowing for gestures (both super convenient for driving tbh) was a pretty cool outcome of what started as just a fun design element.

Another thing I've always found annoying is interacting with notifications at the top of my screen because phones are so big. I had the idea of an overlay at the bottom, similar web concept, where I could give it a quick swipe without having to stretch my thumb up, to dismiss or interact.

## C. Assistant stuff

This is pretty removed from the whole Spidey gimmick and just an excuse to play with AI models, home assistant stuff, and the like. I will say Jarvis was a lot cooler in 2008 when AI wasn't in my toothbrush. That said I still find the bleeding edge of ai pretty fun to mess around with.

There are a couple fascinating pieces to the assistant feature.

Gemini has a pretty nifty [live conversational model](https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-8-live-gemini-3-8-live-extended-thinking/). Straight audio stream in both directions rather than speech to text and text to speech. You can interrupt it while it's talking. It supports function calling, so every capability on the phone is set up as a function it can call call.

Laptop, phone, and a raspberry pi all have [tailscale](https://tailscale.com/) set up for talking to each other. One of the assistant function calls can reach out to Claude via API on the laptop. Then the laptop has remote adb connection to the phone so it can iterate and install a new build back to the phone. That's the whole feedback loop. Raspberry pi is for some misc smart home things like garage door and my smart oven. The pi is also an octopi server for my 3D printer so that's definitely on the list to tinker with soon.

Assistant can watch for events, and create geofences triggered on entering or leaving places. Memory files save places and preferences among other things, Claude took inspiration from itself here. Querying memories is just another function call. 

"Open the garage every time I come home", 

"Take me to Costco, oh and open the app when I get there". 

I've got two equidistant Costco's and it recalls my preferred one. The other one is a mob scene.

Another fun integration Claude figured out is a Spotify developer account that gives API access to listening history. So when I ask it to play a genre it knows what specific playlists I've been listening to recently and will suggest a couple.

## D. The voice

The last thing I did with assistant is pretty silly. The basic Google voice models are fine, but for some reason they all miss that Jarvis/Karen/Edith/EV/assistant-in-the-ear quality. The reason finally dawned on me, and it wasn't the voice itself, its personality, or its phrasing. 

It was literally just sound design. 

AI voice models try really hard to sound human, and they do a great job. In contrast, Jarvis was voiced by a human named paul bettany that they had to make sound more like a machine. Such is life.

**The solution:** some boxy, mid focused EQ; a little chorus; a slight slapback; and some whining ring modulation. Claude made a bench to tweak these on the phone. These are applied straight to the audio stream coming back from Gemini in real time. And as you can hear in the video, the voice has a bit of synthetic texture. I actually find it less unnerving than all the assistants that try to sound like a human.

---

## The Showcase

Final cool Claude thing here was actually putting together this showcase. All the phone screen video is just screen recorded off an emulator claude was controlling (so yes all very real working stuff on my phone today). The screen capture was used as a video texture on a ThreeJS player to compose the showcase. Claude made a rough model of a Pixel phone and the other random models, I didn't go crazy. Claude wanted to jump into Blender and ray trace shit and I said get outta here I'm not even selling anything.

I think it's pretty neat.
```



document fetched 2026-10-22 00:43 · canonical https://reddit.com/r/ClaudeCode/comments/1xbdaov/i_made_a_spiderman_android_launcher_with_claude
