Posted October 08, 2025 by falkenbrew
A quick update on the current state of manypuli and what’s been happening behind the scenes:
I was previously using multiple AI agents with ChatGPT to handle different aspects of the game, but that system has now been deprecated. Because of this, I’ve switched to managing larger, more dynamic rule sets myself. It’s a bit more work (and more expensive), but it keeps the project alive and evolving.
I’ve also moved away from GPT-4-mini, which never quite delivered the writing quality I wanted — especially in non-English languages. The game now uses a combination of GPT-4.1 and GPT-5, which produces much richer, more natural dialogue and storytelling. The downside, of course, is higher costs… so I’m still figuring out how to make this project at least sustainable (and ideally, not a money sink).
To reduce text bloat and recurring story issues, I’ve introduced a curated set of alternate story beginnings. These are stored in a database and partially translated via DeepL, although not all supported languages have full translations yet.
I’ve temporarily trimmed the list of available tasks, avoiding some of the more experimental or “daring” ones, while I gauge how people interact with the current version.
There’s still a pesky bug that can break a story connection, showing an error. For now, the only workaround is to restart the story — but I’m aware of it and working on a fix.
Right now, I’m mainly waiting for feedback and interest to see where to take the next steps.
As always, if you’ve played or tested manypuli, your thoughts mean a lot. Every bit of feedback helps me decide how to grow this weird little experiment in manipulation and morality.