Alara
Hi. Missed me?
I made a new thing. Alara.
---
*Alara version 1.0*
the
*Character-AI-Engine*
What is it? Singularity? Real conscience? Real Person?
>A software that takes your input on runtime.
and gives you back a spoken dialogue.
Also sends "decision-data" to game-engine for the character use.
So characters speak and act realisticly.
Does a few more background-things, but that's basically it.
---
Traditional Character AI:
Personality + History + Scenario + Lorebook = Choice
Problems:
- Impossible to have truly deep characters.
- Limited by LLM's context window (chat history gets cut off).
- Quality depends entirely on the LLM-model.
- Censorship (LLM's ethical filters.)
- Lorebooks rely on vague token matching.
- Scenarios are static and must be written manually.
- Summaries cut important things.
But "Overall" it's "okay" for small fun.
And I used it to create 10 variations of 1 sentence.
But while working and creating dialogues for Ronda... she did not feel right. Not alive enough.
I want a living, breathing, real, sentient, actual lifelike realistic character.
.... and after 2 months of blood and tears.... now finally
I built my own character engine.
*My Engine: 'Alara'*
*Dynamic 'deep' Personality + all character memories + relationships + STATS + current STATES + visual perception + dynamic scenario + goals*
*Core Changes & Fixes:*
---
*Problem:* Context Limit
- Chat history is replaced by *infinite character memories*.
- Memories are retrieved via *semantic matching AI*.
- 'old' Memories are merged into backstory automatically.
→ Infinite context, no more cutoff.
→ No need for traditional lorebooks.
→ No need for summaries.
---
*Problem:* Output depends on LLM quality
- Prompts are *hand-picked* by a C# script.
- Language is *clean and precise*.
- *Different LLMs* handle different tasks.
→ There is something I consider as "100%"-Quality of writing. It's when it's impossible to recognize difference of real-life stories and the generated ones. This is gonna be achieved!
---
*Problem:* Censorship (ethical filters)
- No direct user-to-LLM prompts.
- LLM never receives the “question.”
- Instead, the engine replies in-character.
> GPT: “Sorry, I can’t help with that.”
> My Engine (Genia): “yo.. Dunno? Buy lots of firecrackers?”
---
*Problem:* Static scenarios
- Scenarios are *dynamically picked/generated* by frontend logic (C#/Java).
→ Scenes change based on what's actually happening. There is no "good" or "bad". There is only a "right solution"
---
Using: llama2, gpt4o-mini, "sentence-embedding" (probably gonna stick to bge-small-en-v1.5)
....
And now it's done.
Question:
Yoo what the fuck is this? A chatbot? That's it???
Answer:
A working app for extremely realistic character simulation.
It uses 3 completely different AI-models to work. (bge-small-en-v1.5, llama2<Asha>, gpt4o-mini)
Question:
Ok lol. whatever. so when updateßßßß?
Answer:
I get extremely frustrated and demoralized by repetitive text-writing for 20 different situations. It costs me a lot of energy. And I need to do this for many different variations for each character. And doing this for 1 week messes me up. I need to enjoy this if I want to continue working.
Question:
For FemUZero now a tool to create new dialogues. That's it?
Answer:
For now. And it's a great help. But I need this for the FemUZero-2 anyway.
Question:
Ohhhh FEMUZ ERO 2. how is the save-file transfer going?
Answer:
Yes, ingame "what you did" will be "saved".
In "2" the characters act and behave differently. You can have real realationships.
Single 'history-database' for RPG-Maker MZ and Unity / Unreal and "Messaging-app" and the Discord-character.
*Imagine this:*
You're sitting on the bus, heading home in the evening. You pull out your phone and start playing FemUZero-2.
In the pixelated 2D world, you meet Ronda. She says:
> "Hey there! Good evening. Today’s a special day. Exactly one year ago, we met for the first time. I’ll never forget what you said—and how it made me feel. Wanna hang out later?"
You reply:
> "I wish I could, but I need some rest."
She responds:
> "Aww, really? That’s okay. But you owe me one!"
You close your phone. You arrive home, go to bed, fall asleep, and then wake up.
---
Now you’re sitting in front of your PC. You launch FemU2 and enter the 3D world. You attend a class in-game. After class, you run into Ronda again. She greets you:
> "Hey! Did you get some rest? It’s been 20 hours since I last saw you. Is everything okay? That’s so unusual."
You speak into your mic:
> "I don’t know, my head hurts today. I just got home from work and I feel like I have a fever."
She thinks to herself (you don't read this anywhere):
> "Oh? Did he say he’s sick? But he looks completely fine… Oh no! He must mean his other body, in the 'RL' dimension. My poor baby..!"
Then she replies:
> "Don’t worry, dear. I know a good trick. Go to your kitchen, pour some milk into a glass, and add a spoon of honey. Microwave it for a minute. I call it my milky-honey drink. Drink it now and again before bed, okay?"
You type on your keyboard:
> "Sure, I’ll go buy some honey now."
---
You turn off your PC and head out to buy milk. Five minutes later, your phone vibrates. A message from Ronda pops up:
> "Get Manuka honey if possible."
You reply:
> "No Manuka honey here..."
She texts back:
> "Then get regular honey. Just not the artificial kind, okay?"
---
Later that night, you're lying in bed. Your phone vibrates:
> "Don’t forget to drink your milk. Love you."
> "And you still owe me one <3"
--------------------------------------------
So my lazy ass does not have to write anymore I created this. And now all little characters in the game feel truly alive. And their "actions" (ingame) are based on what they remember you doing and saying.
Question:
Wow. wtf How expensive is this?
Answer:
A regular Patreon membership. Patreon membership is required to access key for regular servers.
>I'll set up and run all the ai-models myself. If I can afford it later, locally. But for now I use clouds.
I don't know how fast or how "immersive" this works. But it works extremely fast and is super cheap right now. If server cost is above my income, I may have to think about extra-tiers or limiting calls or adding some wait-time.
>What I can definitely say is, there will be a "loyalist" tier.
Patrons who stayed in basic-tier in the last 5 months (november to April) receive a 'perma-free-key' for infinite use for the 3 "dev models" (the one I use for development)
Soo that's it.
Are there any questions, feel free to comment.
Maybe you have fancy ideas?
Maybe you have suggestions?
I can show some screenshots of the Unity Project and some UI if you want.