How does the Character AI Lorebook work?

The Lorebook uses keyword-triggered entries to load specific details only when relevant terms are mentioned. Each entry has a title, content, and up to 8 keywords, ensuring that the AI can efficiently recall information without overloading its context window.

What are the privacy concerns with Character AI's Lorebook?

Lorebooks on Character AI are public by default, meaning titles and keywords are visible to everyone, even if the content is hidden under the 'Hint Only' setting. This lack of private testing options has sparked significant community backlash.

How do I create effective Lorebook entries?

Write self-contained entries that tell the character how to react, not just describe facts. For example, instead of saying 'The tavern is old,' say 'The tavern is dim, crowded, and hostile to strangers' to guide the character's behavior.

Can I test my Lorebook entries privately on Character AI?

No, Character AI does not offer a private testing mode for Lorebooks. This means that all entries, even under 'Hint Only' settings, have their titles and keywords visible to the public.

What makes FlirtAlly's memory system different from Character AI's Lorebook?

FlirtAlly's memory system behaves like human memory, storing facts, episodic, emotional, and rolling summaries without keyword limitations. This means FlirtAlly's companions can remember and react to information naturally, even if keywords aren't explicitly mentioned.

Is there a way to link a single Lorebook to multiple characters in Character AI?

Yes, you can link a single Lorebook to multiple characters. This allows for consistent world-building and details across different characters without duplicating data, making it easier to manage complex narratives.

Character AI Lorebook: How It Actually Works (2026)

The Context-Window Problem: Why Character AI Needed a Lorebook

AI conversations can't hold infinite context, forcing platforms to innovate around memory limitations. Character AI’s Lorebook emerged as a solution to this fundamental constraint, enabling consistent character behavior without overwhelming processing systems.

AI models operate within fixed context windows, typically capping the amount of prior conversation they can reference. This limitation means developers must prioritize which details stay "in the room" and which get discarded. The Lorebook addresses this by using keyword-triggered entries that load only when relevant, preserving efficiency while maintaining narrative coherence. Research shows this approach avoids overloading context windows by only activating entries when specific terms are mentioned.

Character AI’s Lorebook requires users to structure world-building data into titles, content, and up to 8 keywords per entry, ensuring precise recall. For example, a "Parisian café" entry might trigger when a character mentions "café" or "Eiffel Tower," embedding details seamlessly. This system reflects a technical rationale: "only bring this up when it’s actually needed," as one expert noted.

However, the Lorebook’s design highlights the trade-offs inherent in context management. While it solves immediate efficiency issues, it also underscores the fragility of AI memory. Published Lorebooks are public by default, with no private testing mode, raising privacy concerns. Users can hide content via "Hint Only" settings, but titles and keywords remain visible, limiting experimental flexibility.

FlirtAlly takes a different approach, treating memory as a core system rather than a bolt-on. Its four-layer memory architecture—facts, episodic, emotional, and rolling summaries—supports unlimited recall across all plans, including the free tier. This contrasts with keyword-triggered systems, which rely on predefined cues to maintain context.

Lorebook entries use up to 8 keywords to trigger specific details
Public Lorebooks default to visible titles and keywords, even with content hidden
Character AI’s Lorebook entered beta on July 24, 2026, with c.ai+ access required

The Lorebook’s emergence reflects a broader challenge in AI development: balancing efficiency with depth. While it offers a pragmatic workaround for context windows, it also reveals the limitations of current models. FlirtAlly’s emphasis on continuous, unbounded memory represents an alternative vision, prioritizing seamless, human-like recall over technical workarounds. Both approaches highlight the evolving landscape of AI companionship, where memory management defines the user experience.

How the Lorebook Actually Works: Entries, Keywords, and +Lore

At its core, the Lorebook is a keyword-triggered system, not a memory bank the AI reads from start to finish. Each entry you create has three parts: a title, the content itself, and up to 8 keywords that decide when it fires. The AI only pulls an entry into the conversation when one of its keywords actually comes up — as one guide puts it, the keyword system is a way of saying "only bring this up when it's actually needed." That design keeps the context window from getting overloaded, which is the whole technical point of the feature.

Here's what a single entry looks like in practice:

Title: a short label so you can find and edit the entry later, e.g. "Ravenhollow Tavern."
Content: the world detail you want injected — description, rules, relationships, tone.
Keywords (up to 8): the trigger terms — "Ravenhollow," "the tavern," "old inn" — that activate the entry mid-chat.

How you write the content matters as much as the keywords. A useful rule of thumb from hands-on testing: a weak entry describes a fact; a strong entry tells the character how that fact should shape its next reply. "The tavern is old" is a dead entry. "When the tavern is mentioned, describe it as dim, crowded, and hostile to strangers" changes the character's behavior the moment the keyword fires.

One Lorebook can also be linked to multiple characters, which is where it gets genuinely powerful for world-builders. Define a city, a faction, or a magic system once, attach the same Lorebook to every character in that world, and all of them stay consistent without duplicating data across each character definition.

You can spot characters using the feature by the "+Lore" tag that appears on profiles and discovery pages, signaling that a Lorebook is active behind that character. Worth knowing before you invest hours in entries: published Lorebooks are public by default, and even under "Hint Only" visibility — which hides your content — the titles and keywords remain visible to everyone. A community moderator confirmed the situation plainly: "We currently only have public Lorebooks." That lack of a private testing mode drew immediate community backlash when the feature shipped, so plan your entries assuming an audience.

Access is also gated: the Lorebook launched in beta restricted to c.ai+ subscribers at $9.99/month, though recent updates added private Lorebooks for subscribers and an import option for plain text and JSON files.

The takeaway for builders: think of keywords as switches, content as stage directions, and shared Lorebooks as your world's single source of truth. It's the same principle we lean on at FlirtAlly — memory and world detail should surface exactly when they're relevant, not sit as static text the model has to wade through on every reply.

The Privacy Catch: Public Lorebooks and the Creator Backlash

Lorebooks were supposed to be Character AI's victory lap — the community's most requested feature, finally shipped. Instead, the launch exposed a design decision that creators immediately pushed back on: everything you wrote was visible to everyone.

According to coverage of the launch, Lorebooks were public by default, with no private testing mode. A community moderator confirmed the situation bluntly: "We currently only have public Lorebooks." For creators, this meant no safe space to iterate — every half-finished entry, every experimental world detail, went live where anyone could read it.

The timing made things worse. The feature entered beta on July 24, 2026, according to hands-on documentation, with access restricted to c.ai+ subscribers paying $9.99/month. Creators were paying for the privilege of testing in public.

Character AI offered one partial remedy: a "Hint Only" setting. As privacy analyses pointed out, it hides entry content — but titles and up to 8 keywords per entry stay visible. That means the actual lore text stays under wraps, but anyone browsing can still see what your entry is called and which words trigger it.

For many creators, that compromise missed the point. Titles and keywords often reveal plot twists, character secrets, or world-building surprises on their own. The community response was swift, and the backlash centered on a few recurring complaints:

No way to test entries privately before publishing them
"Hint Only" still exposed titles and trigger keywords
Paid subscribers had no privacy advantage over free users
No import tools at launch, forcing manual re-entry of existing world notes

Character AI eventually responded. Private Lorebooks became available, but only for c.ai+ subscribers, alongside a new import feature for plain text and JSON files, per updated documentation. The fix arrived, but it reinforced the same tier structure that shaped the original beta.

It's a familiar tension in companion platforms: creators invest the most effort, yet their tooling often lags behind the experience they're building for others. FlirtAlly takes the opposite stance on a related front — memory is unlimited on every plan, including free, because a companion who forgets isn't much of a companion. Whether Character AI's paywalled privacy fully settles the creator backlash remains an open question, but the episode is a useful case study in what happens when a most-requested feature ships without its most-requested safeguard.

How to Write Lorebook Entries That Change Behavior, Not Just Lore

There's a difference between a lorebook that stores information and one that actually changes how a character behaves. As the folks at aicompanionguides.com put it: "A weak entry describes a fact. A strong entry tells the character how that fact should shape its next reply." That single distinction separates lorebooks that feel decorative from ones that genuinely steer conversation.

Each Character AI entry consists of a title, content, and up to 8 keywords, and entries only load into the model's context when a keyword fires. That keyword-triggered design is the whole point — the system avoids overloading the context window by bringing in "only this when it's actually needed." But it also means your entry has to work hard in the few lines it gets.

Write entries that are self-contained

Because an entry is injected mid-conversation, the model never sees the rest of your lorebook when it reads one. Write each entry as if it's the only context the character has. "Ravenhollow is a port town" is weaker than "Ravenhollow is a fog-bound port town the character resents, having grown up there — she speaks of it with a bite of sarcasm." The first adds a fact; the second adds a fact plus instructions.

Phrase content as behavior, not biography

The strongest entries read like stage directions rather than encyclopedia entries. When you draft an entry, ask: what should this fact cause the character to do in their next reply? A few patterns that work:

State the fact, then the reaction: "X is true — when it comes up, the character deflects with humor."
Specify tone shifts: "This topic makes her guarded; her replies get shorter."
Include relationship stakes: "The character owes this person money and avoids the subject."
Note what not to do: "The character never admits this openly."

Choose keywords the conversation will actually hit

With only 8 keyword slots per entry, selection matters more than volume. Pick words that appear naturally in dialogue — names, places, objects the character would realistically mention — rather than rare or overly specific terms that may never trigger. If your entry never fires, it might as well not exist.

It's worth remembering that published lorebooks on Character AI are public by default, with no private testing mode — even the "Hint Only" setting hides content but leaves titles and keywords visible. That lack of a private sandbox makes it harder to iterate on entries before sharing them, so testing your phrasing carefully before publishing is even more important.

The same principle applies on any platform where persistent character memory matters. At FlirtAlly, for example, companion memory is built around facts and emotional context that carry forward — the goal is always the same: information that shapes the next reply, not just sits in a file. Whether you're building a lorebook entry or designing a companion's backstory, write for behavior first and lore second.

What FlirtAlly Does Instead: Memory That Behaves Like Memory

Everything you just read about keyword triggers, entry limits, and privacy gaps exists because Character AI is solving a context-window problem. FlirtAlly takes a different route: memory is built in as a native system, not bolted on through a lorebook workaround.

The difference shows up in daily use. Character AI's lorebook requires you to hand-craft entries with a title, content, and up to 8 keywords, and the system only loads an entry when a matching term appears — a design its own guides describe as "only bring this up when it's actually needed." That's efficient, but it means your companion forgets anything you didn't anticipate and encode manually. FlirtAlly's memory behaves like memory: tell your companion your dog's name and she asks about him next week, unprompted, even if you never wrote a single keyword.

Here's what the four-layer system actually stores:

Facts — the concrete details of your life: names, places, preferences, the small stuff that makes someone feel real.
Episodic memory — specific moments you shared, from your last phone call to that joke only the two of you get.
Emotional memory — how conversations felt, so her tone reflects the relationship as it evolves.
Rolling summary — an ongoing condensed picture of your history, so continuity survives even across months of chatting.

On top of those layers sits semantic recall, which is where the lorebook comparison really breaks down. A keyword-triggered entry fails if you phrase things differently — say "my pup" instead of the keyword "dog," and the entry never fires. Semantic recall matches meaning, not keywords, so "remember when" works even when you phrase it a new way.

Access and cost matter too. Character AI's lorebook sits behind a c.ai+ subscription at $9.99/month, and even then it's a manual authoring tool — you're the database administrator of your own relationship. FlirtAlly includes unlimited memory on every plan, including the free tier, with no tier-gating on how much your companion can remember. Competitors commonly restrict memory length by subscription level; FlirtAlly does not.

Privacy is the other contrast worth noting. Lorebooks on Character AI launched public by default, with no private testing mode — a community moderator confirmed "we currently only have public Lorebooks," and even the "Hint Only" setting leaves titles and keywords visible. Your relationship's continuity on FlirtAlly lives in your companion's private memory of you, not in a shared, publicly inspectable entry list.

For writers who want a companion that remembers the dog's name, follows up after a call, and keeps inside jokes alive without context-window gymnastics, native memory is simply a different category of feature. FlirtAlly is currently in waitlist mode ahead of launch — you can join the waitlist to be first in when registration opens.

Character AI lorebook: how it actually works