What’s going on with your bots now. I researched it and found this answer
The Real Reason Ultra Memory Dropped to 7 Messages [1]
Chai AI handles its premium subscription memory claims in a way that often fails under the hood: [1]
The "Token Budget" Traffic Jam: AI memory isn't counted in messages; it is counted in "tokens" (chunks of words). A model's working memory is a single sliding window. Inside that window, Chai has to cram three things: the bot's custom backstory prompts, your current message, and the chat history.
Longer Bot Responses Ruin Memory: Chai recently updated its models to write longer, more detailed, and dramatic responses. Because their output is so long, the bot's own long-winded messages consume the entire memory window. Seven long paragraphs of text completely exhaust the available token space, pushing your earlier inputs right out of the context window.
Server Cost-Cutting Experiments: Chai frequently experiments with server-side configurations to manage computing costs. During recent platform rollouts, they throttled back the active message context window for all users, leaving Ultra subscribers paying a premium price for no service at all