Token guide

On CrushOn.AI, bots are powered by Large Language Models (LLMs). Whether you are a user or a creator, one of the most important things to understand is tokens. Tokens are the basic units LLMs use to read and write text.

Knowing about tokens can help you make your bots respond, remember and interact better.

What is a Token?

A token is like a small crumb of text. It can be a word, part of a word, a punctuation mark or even a space. LLMs don't read language the way humans do, they read text in tokens.

🔹 Examples:

• "Hello!" = 2 tokens → Hello, !
• "I'm watching you." = 5 tokens → I, 'm, watching, you, .

Different models may tokenize text differently but generally:

• Short words and punctuation = 1 token each
• Long words = 2 to 3 or more tokens
• Emojis and special characters may count as 1 or 2 tokens

How Tokens Are Used

Creators fill in a "description" (sometimes also called "Personality" or "Context") section when they make a bot. This includes the bot's looks, personality, background and so on. This description is stored in the bot's permanent memory. That means no matter what happens during roleplay or how many messages are sent, the bot will always remember what is in this section.

This is permanent memory → the bot's forever memory.

On the other hand, immediate memory is what the bot remembers only for a short time, like greetings for example.

Every time you chat with a bot, tokens are used like this:

• User Input → Every message you send is split into tokens and then stored in the immediate memory
• Bot Memory → Permanent memory taken into account for the bot's response and behavior
• Bot Response → The bot's reply is also made up of tokens, using the available space by taking into account both permanent and immediate memory

If the bot's description is very long, it uses a lot of tokens for permanent memory, which means the bot has less immediate memory left for roleplay messages memory and may forget conversations fast.

A shorter description uses fewer tokens, leaving more room for roleplay memory. That's why it's important not to write descriptions that are too long.

Bots need to process both the conversation history and user's current input, so token use adds up fast.

Token Limits

Each LLM can handle a maximum number of tokens at once (called the context window). When this limit is reached, the model starts forgetting the oldest parts of the conversation.

For Examples:

• GPT-3.5: ~4,096 tokens
• GPT-4: ~8,000 to 32,000 tokens (depending on the version)

If the bot's memory goes over the limit:

• The oldest parts of the memory are dropped
• The bot may "forget" earlier events, facts, or instructions

That's why in long roleplays, bots always tend to forget names or past actions, they simply ran out of token memory.

Why Token Management Matters

If the bot's description is too long (for example, over 2,000 tokens), it can cause:

• Bots forgetting past events, setting or RP context
• Incoherent or inconsistent responses

Good token use keeps your bots consistent and immersive. It's recommended to keep bot descriptions around 1,500 tokens. Above that, memory loss happens fast.

Tips for Optimizing Token Use for Creator

✅ Do:

• Write short, simple sentences to describe all the sections of your bot
• Summarize and keep descriptions brief

🚫 Avoid:

• Very long descriptions
• Repeating information
• Use emoji, special characters, or complicated formatting in descriptions

FAQs

Q: What happens when the token limit is exceeded?

A: The oldest conversation parts are dropped. The bot might lose context or forget details.

Q: Can I increase a bot's memory?

A: Not directly. Memory depends on the model's token limit. But if you're the creator, you can improve memory by shortening descriptions or using external memory tools (like lorebooks if they are available)

Conclusion

Tokens are the invisible fuel behind every message on CrushOn.AI. Managing them well means smoother roleplays, smarter bots and better long-term roleplays.

Whether you're a casual user or building complex characters, watching your token use helps you get the best from your AI companions.