π§ Understanding Tokens & Context Memory (New Documentation Update)
We’ve updated our documentation to better explain something many of you ask about:
What are tokens? What is context memory? And why don’t characters always remember everything in very long chats?
If you’ve ever noticed:
- Details being forgotten in long conversations
- Older messages no longer influencing replies
- Repetition starting after hundreds (or thousands) of messages
This is directly related to how tokens and context windows work in AI models.
All models operate within a limited context window. That means they can only “see” a certain amount of recent conversation at a time. Once that limit is reached, older messages gradually fall out of context. This isn’t a bug — it’s how large language models function.
We’ve now published a clearer breakdown covering:
- What tokens actually are (in simple terms)
- How context memory works
- Why memory limitations exist
- Why repetition can happen in very long chats
- Practical tips to improve long-form roleplay quality
If you want to better understand how to get the most out of your chats — especially long roleplays — we highly recommend reading it:
📖 https://docs.spicychat.ai/advanced/tokens-and-context
The more you understand how the system works, the better your results will be.
As always, if you have questions, feel free to reach out.