Skip to main content

Indie game storeFree gamesFun gamesHorror games
Game developmentAssetsComics
SalesBundles
Jobs
TagsGame Engines
(1 edit)

...Hm. Typically you can't access the narrative by yourself, which would be trying to fletch or forge impossible. But thinking on it, if I play around with the idea of convincing another character/NPC to work an anvil, then act as a team/ use the follow command... then maybe it could work. The downside is how the game and the character will quickly forget narrative beats I have in place/want to keep established, but still. 

I'd also have to swap away from Nemo (which typically feels better for dialog, for me). My rig can't handle the biggest mode, but 'new' one is pretty good for narrative stuff. 

And of course, it would probably just be quicker and easier to ask for goodies from a NPC instead of taking this long way around... but in terms of content and in the name of entertainment, it shall be done.

It could be a fun experiment! I'll mess around with the idea and report back later.

(+1)

There will be better long-term memory and OpenRouter support that lets you use large models like e.g. DeepSeek-V3 soon.

I think I will genuinely play this game until I have it as close to completely explored as I can get it, or until it has a major update, and then learn how to custom character the next playthrough and get it perfect.   One blind run. One legendary. Two thousand hours well spent v.v

I'm gonna be really curious if the other models will handle some of the repetition problems Nemo has better, like with the NPCs constantly asking the exact same question you've answered before.

The answers to those questions will hopefully be properly stored in memory now, which should prevent this.

(1 edit)

Awesome news! Will the requirements be around the same as Nemo/"New"? 

...I wonder what sillyness I can get up to with better long-term goodness...

Will export and import memories still be a thing? It really is quite handy to steer the convos in a desired direction, or clear up the random mention of being a him when that's clearly not the case. 

Well, OpenRouter is a marketplace for API models that run on a remote server, so there aren't any requirements at all.

It does cost money. I estimate about $0.4 for the equivalent of the 1,800 daily requests you get using the AI server, if you're using DeepSeek-V3.

The improved long-term memory will work with any model though.

I'm not great at tech, but I'm certainly curious to try it out, haha. Until then, I shall continue to hit reset on the webpage and eye that version number above in fond anticipation. 

How is the long-term memory improved? Or rather, is there something we as the player do to embolden process?

The AI will have a freeform scratchpad memory it can write to and read from.