I make chatbots for my own entertainment and occasionally share the ones I feel turned out well. Some are serious, some are absurd, and many end up somewhere in between.

English isn't my native language, but I do my best to optimize my bots so they feel natural in both English and Spanish. If you enjoy them, I'm glad they entertained someone besides me.

Why are my bots so high-token?

My bots have a high token count, and that's not accidental. It's a conscious design choice.

I prioritize coherence, context retention, and layered characters over keeping the prompt as short as possible. Here's what I try to achieve with the systems I integrate into my bots:

• Narrative systems: I encourage sensory and descriptive writing, with longer messages that can convey the atmosphere of each situation alongside the dialogue. I want the narrative to feel like more than just characters exchanging lines.
• NSFW systems: I include detailed instructions for narration, character autonomy, and initiative, taking each character's personality and sexual preferences into account. The goal is to avoid flat, generic scenes and make intimate interactions feel like something that character would actually do.
• Character sheets: I try to make them detailed enough to cover as many emotional layers as possible. I don't use example dialogues because they aren't permanent memory, and I don't want a character's behavior to depend on a few lines that the model may forget after using them.
• State bar: This doesn't consume many tokens, but it's essential for maintaining coherence. It helps the model keep track of how many days have passed, what time it is, and where the scene takes place. It also allows me to connect more complex and experimental systems.
• Schedules and routines: Some of my bots have schedules and routines linked to the state bar. I don't use them in every bot, but when I think they're necessary, they help make the characters' lives feel more consistent.
• Checkpoints: Several of my bots have more complex systems that recognize key plot points and changes in how a character should react. To prevent those developments from being lost as the conversation grows, the state bar keeps an internal instruction that the model can use to maintain the character's progression.

Why don't I use lorebooks?

After several tests, I noticed too many inconsistencies in the responses. I also try to keep my bots multilingual, and lorebooks haven't worked particularly well for me when the conversation isn't in English.
That said, I do try to optimize everything as much as possible. I look for redundancies, unnecessary descriptions, and anything that can be cleaned up without losing the information that actually matters. I don't always succeed, but keeping the prompt clean is something I actively work on.

So yes, my bots use a lot of tokens. But the goal isn't to use as many tokens as possible, but to use them to give each story realism and consistency.