This was supposed to be a small update with a bunch of improvements and bug fixes. But then I received my DG-LAB Civet and decided to add support for it in Eudaimonia as soon as possible, and September quietly slipped into October. So let's call this one the Octember update.
Stereo v1 and the "Legacy devices" mode in EstimAI are still deprecated, and still here: their removal is postponed to the next release. If you rely on one of them, it's a good time to move to Stereo V2 and Multi-Devices.
A huge thank you to everyone supporting the development of Eudaimonia, and to everyone who took the time to report bugs, send feedback or share ideas. A good part of this update comes straight from your bug reports and from the request threads on the forum.
The scenario promised in the August notes is here, along with an alternative version of it. Both are tower-like scenarios.
For the anime aesthetics enjoyers. Every floor has been written by hand, around a classic of the genre.
The alternative version, with a more classic tease & denial flavour. No floor is written in advance: each one is generated when you enter it, which means infinite floors.
Both scenarios are huge: 15 to 20k of context.
This is the experiment of this update.
The Civet is a small Bluetooth pressure sensor made by DG-LAB. Eudaimonia can now connect to it during a session, which means that the AI actually knows how close you are, whatever you tell {{char}}.
This is still a work in progress: I have little to no experience with this kind of technology, and I could only test it on myself. So everything released here is still at a very early stage of development.
What I can say, though, is that it has worked for me in all my tests so far.
Eudaimonia doesn't use it to guess how aroused you are: it reads your fight. She pushes, you hold back, and the sensor catches two moments:
Everything happens on the new Edge Detector page: connect the sensor and run the guided protocol once to calibrate it. Then enable it in Quick Play, in the expert form, or in the in-game settings.
Read the guide: how it works, and how to play it
Two caveats you should know about:
This is an early beta, and still experimental. It has been tuned on very little data, and every body is different: expect false positives.
But you can help me improve it! Don't hesitate to run the calibration protocol and to play with the "Test" mode on the Edge Detector page: it gives me more data to tune the algorithm. And please tell me how it goes for you!
You can now create a prompt section: a block of instructions that is added to the system prompt, and that you can switch on and off. This is useful if you have the same set of instructions in multiple scenarios, or want to switch between them easily.
Make sure to reference your prompt section in the scenario's context, so that the model remembers it and uses it properly. For example:
I've also written a few common sections that I think are staples for standard Eudaimonia scenarios, such as edging_techniques, punishment_strategies, release_process, wardrobe_and_camera or learning_the_user.
The list of models was confusing. It should now be clearer and easier to understand: fewer models, a short description of how each one behaves in Eudaimonia, and a simple idea of its cost, intelligence and speed.
Too many to list, especially with the stream mode and the Coyote 3.
That's it for Octember. If something behaves weirdly, please tell me, either on the Community Space or at contact@daimonia.app.
I hope you'll like it, have fun!