It still wont obey only bottom commands, aka 20% depth
1 Like
Thank you!
This is an issue I am currently tackling. At the moment the AI is simply juggling too many āRulesā lol, which can make it mess up like this, but it should be resolved in the next major release.
2 Likes
Good morning, all! 
My new, beefy GPU should arrive today (early hopefully!).
WE ARE ABOUT TO BE COOKING WITH GAS, BABY!
On another note. Iāve tried to tidy up the entire thing as there were parts of the system commands I had AI help me with that made everything too confusing. The plus side of this is that I can now manage my codebase much more easily.
The next update will include a few fixes, but, most importantly, it will split the project files into less of one giant mess of code. Itās partly the reason so many isues keep cropping up when I push updates, because itās bloody hard to maintain lol.
Iām also starting over with how the AI understands how the handy moves and how that correlates EXACTLY to massaging a dick. This will take me a while, so please keep in mind Iām just one dude. What all of this means is that when you use the app in future, the AI will ask you exactly how much depth is best for you (one time). Then it will ALWAYS make sure to stroke you EXACTLY within that range of depth. Right now it does a lot of guess work.
3 Likes
Ok. The vastly-reworked version works pretty well, but it is unstable and I donāt feel confident in sharing it just yet (no broken cocks on my conscience), but hereās everything you can expect from 1.3. Get fucking excitedā¦ 
- This is a complete, ground-up rewrite of the AIās core logic and movement system, designed to make it feel less like a program and more like a true, responsive partner.
- The app now uses a high-precision positioning system.
- New Feature: Guided Setup: walks you through a one-time calibration process where you teach the AI your personal minimum and maximum comfort limits. From now on, when the AI commands a move to ā50% depth,ā it means your 50%, making every movement tailored to your body.
- New Feature: Generative AI Movement Engine (replaces random inventive patterns) The AI now has a vocabulary of real stroke techniques (Full Strokes, Teasing, and Edging). It creatively combines these techniques with different speeds and positions to invent unique, relevant patterns on the fly in response to your messages.
- New Feature: āSmartStrokeā Interactive Auto-Mode: The old āauto modeā has been replaced. In the new SmartStroke mode, the AI has its own āmoodā that evolves throughout the session. It will change patterns and intensity on its own, creating a unique story arc from gentle, to teasing, to passionate without you having to say a word. You can still chat with it during this mode to influence its mood and actions. Unlike auto-mode which was mostly random, SmartStroke actually creates strokes based on context.
- āIntent Correctionā:. If you say āsuck the tipā and the AI mistakenly chooses a full-body stroke, the code now overrides the AIās mistake and forces it to invent the correct āteaseā pattern. This ensures your specific commands are always respected.
- āSTOP RIGHT THERE, CRIMINAL SCUM!ā: When you tell the AI to stop of press āSTOPā it will always half all actions fully until you decide what to do next.
- Persona Upgrade: The AI will now ALWAYS stick to its persona, far more deeply than it did before. It will also never magicaly grow a cock unless you want it to.
- Random fixes: You can now press Enter to send messages. The entire setup process is now a guided, step-by-step flow, making it impossible to miss crucial steps like calibration. I have completely rebuilt the application from a single, MAHOOSIVE file into a modular architecture. This change makes the app significantly more stable, faster, and easier to update in the future.
Thank you so much for all your kind works, feedback, testing, and support so far. And thank you a crazy amount to the two forum users who offered me their spare toys! Yaāll are amazing! 
8 Likes
any plans to add intiface central support?
2 Likes
I keep running into the timeout issue, and im unsure how to change it according to the specs in the post. Can anyone tell me how?
I have this issue too on my steamdeck (windows) but not on my pc. I think it might be because it takes too long to generate a response on older or lower end hardware.
i was getting close and i said dont stop. only to get ai to say stopping and losing my edge. every time i said im going to cum or im cumming it stopped and wrote lets catch breath or something.
Thanks for this.
Yeah, the AI gets quite confused right now and can sometimes stop at the absolute worst times lol. The good news is the next update should fix this entirely.
1 Like
Yeah, it definitely will not run well at all on a Steamdeck. Sorry about that.
1 Like
What are your specs?
Really interesting project. What the ai is supposed to remember between the session? Names, subject discussion, or anything else? In my_settings.json, i see a persona desc and max depth but they always reset to default ones at launch.
1 Like
This is cool. Tested once, worked really well.
One minor issue: Every time I activate any buttons or fields (including the chat) it triggers 1password to try and save it as a āloginā at the 127.0.0.1 address. Can just be ignored, but is there any way to avoid it? Either in how the web interface is coded, or on my end?
Thank you. Sadly, the memory file doesnāt work as it should right now. It should store important info, your name etc, and the general context of your previous interactions. But as you said, itās currently borked so it accidentally overwrites everything instead of instead adding to the file.
Thatās an error on my part. Iāll have it fixed in the next update. I wouldnāt recommend editing the code right now as itās a bit of an intertwined mess lol.
Thank you for the kind words!
2 Likes
v1.3 Available here!
- The app now starts with a simple onboarding guide. Set your Handy key ā Set AI persona ā Define your preferred minimum and maximum depth. You can also set up ElevenLabs (if you want to).
- The AI now knows your exact dick size, or close enough, thanks to the new guide.
- ElevenLabs voice support should be much more stable and sound more emotive.
- The AI now understands cock anatomy about 85% of the time. Yes, I actually tested it that much
. No more thinking the tip is at the bottom of the shaft. - The AI should now stop when you click āSTOP EVERYTHING.ā If it doesnāt, UNPLUG YOUR HANDY.
- The AI will now consistently stick to its personas.
- Chatting with the AI during automode can still make things janky. Itās a longer-term fix. Not fixed yet.
- Milking mode is very experimental. If the AI tries to whittle you into a makeshift goon spoon, just remember itās still learning.
- Memories should now save properly instead of just replacing old ones. (May not be 100% fixed yet.)
- I havenāt reinvented the wheel, but Iāve made it roll with way less friction.
Iāve tried getting Codex to help untangle my disaster of a codebase, so if you want to poke around, it should be easier to understand now.
Thanks for all the support!
5 Likes
Tried to command the AI to gradually increase the speed with commands like
āIncrease speed 1 % every 1 secā or
āGradually increase speed over 5 minā or
āIncrease speed continuously from 0 to 100 % over 5 minā
but it seems not to understand the commands.
The AI seems to be more or a pattern recognizer and random pattern player. 5 GB+ of downloaded AI content should be able to do more.
Thanks for the feedback
The AI runs on a small local model, so it doesnāt have timers or long-term planning. Commands like āgradually increase over 5 minutesā need features that go beyond what local LLMs can handle on their own (though cutting-edge agent models at places like OpenAI are starting to bridge that gap). Itās not random, just quite limited by design for now.
That said, this is a solo project, so thereās only so much I can pack in for now. But itās evolving.
Also worth noting for anyone curious about how this stuff works, the model isnāt custom-built by me. Itās a pre-trained open model, and Iām building the app experience around it.
3 Likes
This looks interesting
1 Like
You should read up about the size of the largest models. 5 GB is nothing basically
1 Like