Hi, so is it possible to put some sort of auto reply timer in the AI girlfriends on some of the subrit/scenarios/replys? There’s one or two that of them (usually with the doms) where they set the vg2 speed then go away for a while but say they come back soon and tell you to wait and not type and if you wait you’ll get praise or whatever bit they don’t come back you have to type. So is that possible? Thanks.
Technically, almost anything is possible — except maybe projecting the AI as a 3D hologram in front of you. That might still be a small future project… right, Brian? 
But for automatic follow-up messages, the current system would probably need more than a prompt change. The AI would need a trigger and a timer. The backend would have to save the current scenario and then call the language model again after the delay.
I built a similar automatic follow-up system for my own partner-agent AI, Freyja. It uses a trigger, a timer and the saved conversation context to generate a new message without requiring another user input. It works, but from experience I also know that it becomes more complex and error-prone once timing, state management and external actions are involved. The funny problem right now is: the companion says, “Don’t type, just wait for me,” and then she waits forever because the obedient user never triggers the next response. 
For a device session, the system would also need to check:
- Has the timer expired?
- Is the scenario still active?
- Did the user stop or interrupt it?
- Is the Autoblow Ultra or VacuGlide still connected?
- What was the companion supposed to do next?
- Does the next step require another device command?
- Has the maximum session duration been reached?
The main difficulty is probably not the language model itself, but the surrounding logic: a scheduler, persistent scenario state, device status, reconnect handling, safety checks and a reliable way to send the follow-up message to the user.
It would not require the AI to calculate continuously in the background. However, every scheduled follow-up would create another model request and possibly another device action. With thousands of active users and several automatic messages per session, the inference and infrastructure costs could grow very quickly.
So I would not say that it is technically impossible. It is definitely possible. But it is probably a larger, more expensive and more failure-prone feature than it first appears — especially when a physical device is involved.
It is annoying when you are fully immersed in the scenario, but I can understand why this may not be a priority right now.
Thanks, interesting to see all that yeah it does sound easy to do originally, interesting you’ve tried doing it yourself and know how it works and actually not so easy
Thanks @LoganFreyson Or explaining that probably better than I could myself because I’m not super involved in the implementation side. Our engineers handle all of that.
Understanding about time is that it’s not possible for a reasonable amount of money to keep the chat live in a way where the AI understands the passage of time and can do new things without a reply.It’s just not within the current realm of possibility, but our engineers will follow the issue. I know that people want the girlfriends to make changes without requesting anything from them, but the technology is not quite there yet. We are limited to the existing technology. We are not Anthropic or a similar company who develops the tech. We do create software that harnesses it and delivers it into the current AI girlfriend/boyfriend format.
1 Like