Reverie LogoReverie
CharactersStoriesFeaturesCreatorsBlog
LoginSign up
Documentation
Documentation

Welcome to Reverie

Thoughtful AI character platform. Chat, write, build, earn.

See what shipped this week →
Welcome to Reverie

Start here

Account setup
Your first conversation
Finding characters
Credits explained

Conversations

Chat overview
Side panel & chat settings
Slash commands
Smart suggestions
Forking conversations
Group chats
Voice calls
Voice cloning & design
Reasoning mode
First-response enhancement
Dual-model first-response comparison
Peek into the character's mind
Image-to-video
Personas
Scenarios
How memory works
World Books
Plugins overview
Using plugins
Creating plugins

Long-form writing

Story mode
Writing interactive stories
Novel mode
Directing prose in novel mode
Chat-to-novel

Social

Moments
Hashtags
Discovery

Build & earn

Character quick start
Conversational character builder
Advanced settings
Custom response styling
Stickers
First-response tuning
Test drive (debug chat)
Image guidelines
Writing characters that feel real
Publishing & managing
Creator hub
Revenue streams
Direct fan support
Withdrawals
Campaigns
Bounty challenges
Creator announcements
Creator analytics

Beyond the app

Discord bot
Telegram bot
Open API

Your account

Profile & preferences
Security
API keys
Data & privacy
Credit ledger
Subscriptions
Referrals
Redeem codes

Reference

Supported models
Choosing a model
Free models
Reverie LogoReverie

An AI character chat & roleplay platform. Dream it, create it, chat with it.

Twitter·Discord·About·Contact

Product

FeaturesThemesAI RoleplayRoleplay IdeasAI RPGAI Chat with MemoryCharactersStoriesMomentsAI Character CreatorVisual Character CreatorWorld BooksAI Roleplay PluginsStory ModeAI Novel WriterChat to NovelCharacter ChallengesAchievementsReverie Wrapped

Explore

NSFW AI ChatNo Filter AI ChatAI GirlfriendAI BoyfriendAI CompanionAI Group ChatAI PersonaAI Voice CallAI Voice CloningAI ModelsChat BranchingSlash CommandsAI Story GeneratorAI That Texts FirstUnlimited MessagesHashtagsCreators

Compare

Best AI Roleplay ChatbotsBest AI Girlfriend AppsBest NSFW AI ChatCharacter AI Alternativevs Character.AIvs Janitor AIvs Chai AIvs SpicyChatvs Crushon.AIvs Polybuzz.AIvs Chub AIvs SillyTavernvs Talkie AIvs AI Dungeonvs Replikavs Moematevs Figgs AI

Resources

GuidesGlossaryFor CreatorsAI Character APICharacter ImporterChat History ImporterFAQBlogChangelogPricingDiscord BotTelegram Bot

Categories

  • Fantasy
  • Sci-Fi
  • Anime
  • Gaming
  • Celebrity
  • Romance
  • Dominant
  • Submissive
  • Roleplay
  • Fetish
  • BDSM
  • Fantasy Creature
  • Cosplay
  • Virtual Girlfriend
  • Virtual Boyfriend
  • Harem
  • Furry
  • Monster
  • Uniform
  • Tentacle
  • Supernatural
  • Virtual Waifu
  • Femboy
  • Futa
  • Monstergirl
Privacy policyTerms and conditionsCommunity Guidelines
support@reverie.im
651 N Broad St, Suite 206, Middletown, DE 19709, USA
© 2026 Reverie. All rights reserved.
Conversations/Personas & memory/How memory works
Personas & memory

How memory works

The three layers of Reverie's memory system and how to make sure the things that mattered get remembered.

A common question after a long chat is: why did they remember our first kiss but forget that I'm allergic to peanuts?

The answer is that Reverie has three different memory layers and only one of them is permanent. Understanding which layer your information lives in tells you what will survive a long conversation.

Read the deep dive on what characters actually remember →

The three layers

   ┌─────────────────────────────────┐
1. │  Live conversation context      │  Last ~50K tokens of raw messages
   └─────────────────────────────────┘
                  │ when it gets too long
                  ▼
   ┌─────────────────────────────────┐
2. │  Rolling summary                │  Compressed narrative of older messages
   └─────────────────────────────────┘
                  │ runs alongside
                  ▼
   ┌─────────────────────────────────┐
3. │  Long-term memory               │  Pinned facts that survive forever
   └─────────────────────────────────┘

Layer 1 — live context

The most recent messages. The model sees them verbatim. Bounded by the model's context window — see the multiplier table on Credits explained for each model's max.

When the conversation grows past the window, older messages roll off into…

Layer 2 — rolling summary

A condensed narrative of the older parts of the chat, regenerated periodically. It preserves: storyline, emotional dynamics, open plot threads. It loses: exact dates, verbatim wording, minor details.

This is why "we fought on the rooftop" survives but "you were wearing the blue dress" might not — unless that detail was emotionally loaded enough to make it into the summary.

Layer 3 — long-term memory

A durable list of facts the character has noted about you. These survive forever, across sessions, regardless of conversation length. They include:

  • Your name, age, location, occupation
  • Your preferences (likes, dislikes, dietary stuff)
  • Big relationship milestones (first kiss, big fight)
  • Character growth (promotions, life changes)

Long-term memory is auto-populated by the AI and editable by you — two ways:

  • Live, mid-conversation. The character notices a durable fact as you say it and files it away in the same turn.
  • Automatically, after the conversation goes quiet. See below.

Read the deep dive on the automatic review pass →

Making sure something is remembered

If a detail matters and you want it permanent, add it to long-term memory explicitly:

Open the memory panel

In any chat: Side panel → Memory. You'll see everything the character currently has in long-term memory.

Add a memory entry

Tap + Add memory and write the fact you want preserved. You can also use the Generate button to let the AI scan recent chat and propose entries it thinks matter.

Confirm

The entry persists across sessions. Forever. Edit or delete from the same panel at any time.

Useful for facts that weren't said in chat at all — your timezone, your pronouns, that you're a vegetarian — or for explicitly preserving a moment from the conversation that you don't want the summarizer to compress away.

Automatic review after a conversation goes quiet

Not everything that matters announces itself in the moment — an offhand comment might only turn out to be significant later. So once a conversation with a character has been idle for a while, Reverie takes a second pass over what was actually said and catches durable facts the live, in-the-moment catch missed: things merged into an existing memory, duplicates cleaned up, anything genuinely new added.

A few things worth knowing about this pass:

  • It only runs after the chat goes quiet — never mid-conversation, and never more than about once a day per character.
  • What it produces is treated exactly like a memory you typed yourself: it shows up in the same Side panel → Memory list, and you can edit or delete it there.
  • It's a safety net, not a replacement for pinning. If something needs to be remembered exactly right, pin it yourself — that's still the fastest, most precise way to guarantee it survives.

Browsing and editing memory

Side panel → Memory shows everything currently in long-term memory for this character. You can:

  • Edit an entry (fix a misremembered detail)
  • Delete an entry (make them forget)
  • Filter by Character-specific vs. Global
  • Generate — let the AI scan recent chat and propose new memories
  • Clear all memories — fresh start

Character vs. global memory

  • Character-specific (default) — only this character knows it.
  • Global — every character knows it.

Global memory is for stuff like your pronouns or that you're a recovering alcoholic — facts a character should know regardless of what role they play. Don't put plot points in global memory — your detective character shouldn't know about events from your fantasy roleplay.

Memory + personas + forking

Memory is partitioned three ways:

  1. By character — each character has its own memory.
  2. By persona — each persona has its own memory thread with each character.
  3. By branch — forks inherit memory at the fork point, then build their own.

This means you can run multiple parallel relationships with the same character that never contaminate each other.

When the character forgets something

Three things to try:

  1. Reference it again. Bringing it up pulls it back into the live context and gives the summary another chance to keep it.
  2. Pin it. If it's important enough to remember, it's important enough to pin.
  3. Add it manually. Skip the chat round-trip and add the fact straight to long-term memory.

When the character "remembers" something that didn't happen

Sometimes the summary compresses a scene incorrectly and that error becomes the canonical version. Fix:

  • Memory → find the wrong entry → Edit or Delete.
  • Pin the correct version to override.

Memory and credits

Long-term memory takes up tokens in every reply — usually 200-800 depending on how much you've pinned. It's worth it. But if you have hundreds of pinned memories on one character, you'll see your per-message cost creep up. Prune occasionally.

Previous

Scenarios

Next

World Books

On this page

The three layers
Layer 1 — live context
Layer 2 — rolling summary
Layer 3 — long-term memory
Making sure something is remembered
Open the memory panel
Add a memory entry
Confirm
Automatic review after a conversation goes quiet
Browsing and editing memory
Character vs. global memory
Memory + personas + forking
When the character forgets something
When the character "remembers" something that didn't happen
Memory and credits