Context is not memory
Recent messages are precise but finite. Once they leave the model's context window, another layer must preserve what still matters.
A clear guide to recent context, rolling summaries, long-term facts, persona boundaries, conversation branches, and lore that appears only when relevant.
5 field guides
Understand which layer remembers what, then organize facts and lore so the right context appears at the right time.
Working principles
Recent messages are precise but finite. Once they leave the model's context window, another layer must preserve what still matters.
Narrative compression keeps plot, emotional dynamics, and open threads while intentionally sacrificing exact wording and minor detail.
Long-term memory is strongest when users can inspect, correct, pin, and delete the facts a character carries forward.
A World Book retrieves the locations, factions, rules, and people related to the current scene instead of injecting the whole encyclopedia every turn.
Reading path
01
AI memory isn't magic - it's a layered system of long-term facts, automatic summaries, and per-persona context. This guide explains what your Reverie character actually keeps, what it loses, and how to make sure the important moments survive the next 100 messages.
02
Not every important detail announces itself as important. Reverie's characters now take a second look at your conversations after they've gone quiet, and quietly hold on to what mattered - without you ever opening the memory panel.
03
Last year we argued lorebooks were a complexity trap — regex triggers, insertion probabilities, a multi-language nightmare most people abandon. We still believe that. So when we built World Books, we kept the one thing worth keeping — giving your characters a world — and threw out the rest. Here's why we changed our build, not our mind.
04
What if you could be anyone in your conversations? Our user persona system doesn't just help AI remember you - it unlocks infinite ways to explore different versions of yourself and entirely new personas.
05
What if every conversation could branch like a story? Our fork feature isn't just about trying different responses - it's about giving users the freedom to explore infinite possibilities without fear of consequences.
Practice path
Save exact names, promises, and rules as durable entries. Let summaries carry softer emotional and narrative texture.
Choose character-specific or global memory, then use personas and branches to keep unrelated stories isolated.
Split lore into focused entries with names, aliases, and concise descriptions that can be retrieved independently.
Review memories and World Book usage, repair inaccurate entries, and remove information that no longer belongs in the story.
Guide FAQ
Put the guide to work