Janitor AI Lorebooks: Where They Are and Why Yours Is Not Firing (2026)

On Janitor AI, a lorebook is not part of your character. It is a script — a separate object, built in a different part of the site, and attached afterwards.

That single structural fact accounts for most of what people cannot find. They open the character editor looking for a lorebook section, work through every field twice, find nothing, and conclude the feature is missing, paywalled, or broken. It is none of those things. It is simply not there, because it was never a character field.

This guide covers where lorebooks actually live, how an entry decides to fire, what changed when secondary keys were fixed, what you can and cannot see of someone else’s, and the token accounting that the character editor will never show you.

If you are still building the character itself, that is a separate job and lives in how to make a Janitor AI bot. If chats are failing outright rather than forgetting things, start with Janitor AI not working.

Where Lorebooks Actually Live

Janitor introduced lorebooks as part of its Scripts system, and the system has two modes. From the platform’s own beta announcement:

Lorebooks — a visual editor mode for world-building and lore management. This is the one most people want.
Advanced — a code editor for more complex logic and full control.

You reach both the same way: click your profile icon, then the scripts tab. Not the character editor. The scripts directory sits at janitorai.com/scripts, and you pick a mode when you create a new script.

The reason for the separation is worth understanding, because it is also the feature’s main advantage. A lorebook is not owned by one character. It is its own object, which means one lorebook can serve a whole cast in a shared setting — build the world once, attach it to every character who lives in it. That is exactly why it could not be a field inside a single character form.

It also explains the shape of the advanced mode. Janitor described scripts as a way to add actual logic to a character rather than praying the AI follows your prompt — a character that only reveals a backstory after fifty messages, or that reacts to particular keywords. Lore storage and conditional logic are two uses of one system, which is why they share a home.

What Janitor Has Documented

Update, 23 September 2026: this section first said Janitor had not written lorebook documentation. That was true of the scripts launch, but Janitor now runs a help centre at help.janitorai.com with beginner guides to scripts and lorebooks. This section has been rewritten to say what those guides cover.

When the scripts beta launched, Janitor said plainly that it would not document the feature yet: a lot was going to change, the interface needed major work, and it did not want to document something that would be different the following week. It linked a community-made guide instead.

That has changed. Janitor’s help centre now has a beginner overview of scripts and a step-by-step guide to building one, and they settle several things this page could only infer before:

Both sides can trigger an entry. By default, a keyword fires an entry whether you type it or the bot does.
There is a Depth setting. It controls how many recent messages are scanned. Set it to 1 and only your most recent message is scanned, so the bot’s own replies can no longer trigger entries — the help centre’s suggestion if you want to stay in control, and a guard against entries setting each other off in a loop.
Each entry has one text box. If you want an entry to add different text in different places, the guide says to switch that entry to Advanced mode.
SillyTavern entries paste straight in. Copy the contents of the entries array into the script’s source or JSON view.

Much of the rest — the more complex community scripts, the tuning people do in Advanced mode — is still community knowledge, some of it now hosted on the help centre with the author’s name on it. Where this guide states something as fact, it comes from Janitor’s changelog, its help centre, or what its public endpoints return. Where it does not, the piece says so.

How An Entry Decides To Fire

A lorebook entry has, at minimum, one or more keys and some content. When a key turns up in what is being scanned, the content goes into the prompt. When it does not, the entry stays out and costs nothing.

This is the part people find counter-intuitive, and it is the cause behind almost every “my lorebook is not working” report:

A lorebook is not background reading. It is a set of conditional insertions, and it only inserts when triggered.

The model does not read your lorebook and absorb it. Your lorebook sits inert until a key matches, at which point that one entry’s text is added. An entry about the capital city exists, as far as the model is concerned, only in the turns where the capital city gets named. Write a beautiful thousand-word entry on your world’s religion, never mention the religion by name, and the model will never once have seen it.

The scan covers recent messages, not the entire conversation. How many messages are scanned is set by the script’s Depth setting, and Janitor’s help centre notes that a Depth of 1 scans only your latest message. Janitor does not publish the default, so we are not going to invent a number for it, but the behaviour it produces is consistent and worth planning around: a key that was mentioned a long time ago is not currently in scope. The city named twenty turns back is not triggering anything now.

Secondary Keys, And Why Your Entry Got Pickier

Janitor’s weekly changelog for 19 September 2025 records fixing a batch of lorebook bugs and getting secondary keys working properly, describing them as letting you be much more selective about which entries actually make it into your prompts instead of dumping everything.

Two things follow from that.

First, secondary keys are a narrowing device. A primary key alone fires an entry. Add secondary keys and the entry needs more than the primary key before it will go in. That is how you stop a common word from dragging a long entry into every single turn — key an entry on a character’s first name, then require a secondary key before the full biography appears.

Second, and this is the part that surprises people: secondary keys were previously buggy, and getting them working made entries fire less often, not more. If you built a lorebook before that fix and your entries seemed to appear reliably, some of them may have been firing when they should not have been. An entry that stopped appearing after the fix is not necessarily broken. It may be the first time the filter you wrote has actually been applied.

The Triage For An Entry That Will Not Fire

Work down this list in order:

Is the key actually in the recent conversation? Say the key word out loud in your next message and see whether the entry appears. This one test resolves the majority of cases.
Are secondary keys narrowing it out? Strip them temporarily. If the entry fires without them, the filter was the cause and you can rebuild it more loosely.
Are your keys the words you actually type? People key entries on formal names — “the Kingdom of Aventhal” — and then write “the kingdom” in every message. Add the casual forms you really use, and the plurals and possessives.
Is the lorebook attached to this character? A lorebook is a separate object and attachment is a separate step. An unattached lorebook is a well-organised document that nothing reads.
Is it the lorebook at all? If the bot is also losing track of recent events, that is a memory problem rather than a trigger problem, and no lorebook change will fix it. What happened in your own story belongs in chat memory, which is sent on every turn instead of waiting for a keyword.

The Token Cost Nobody Shows You

Janitor’s character records break a character’s token total into four parts: personality, scenario, example dialogue and first message.

There is no lorebook line among them.

That is a genuine gap in the interface rather than a criticism of the design — the lorebook is a separate object, so it is not part of the character’s own count. But the practical consequence catches people out. A character can show a small, healthy token total in its own counter and still push a large amount of additional text into the prompt every time entries fire. The budget is real. It is just not displayed anywhere you would think to look.

For scale, the default generation context length in Janitor’s own client configuration is 50,000 tokens. That is a comfortable budget, and it is also a shared one: character definition, chat history, persona and every fired lorebook entry are all drawing on it. A dozen long entries all keyed to words you use constantly will take a real bite out of the room left for the conversation itself.

The practical rule is the same one that applies on every platform with this feature: keep entries short and let several fire, rather than writing few long ones. A hundred words of the facts that matter beats six hundred words of atmosphere, because the atmosphere is what the model was going to improvise well anyway.

Importing From SillyTavern

If you have already built lorebooks elsewhere, you probably do not need to start over. Janitor’s scripts beta announcement states directly that the lorebook mode lets you import your existing SillyTavern JSON lorebooks.

That makes SillyTavern’s lorebook JSON the working interchange format between the two platforms, and it is the answer to a question a lot of people ask in the other direction as well — whether a migration means retyping everything. It does not. Export from SillyTavern, import on the Janitor side.

One caution. An import moving cleanly does not guarantee that every option inside those entries means the same thing on both platforms. SillyTavern’s lorebook entries carry a large number of fields — insertion order, probability, recursion controls, timed activation and more — and two platforms implementing the same file format do not necessarily implement every switch in it. Treat an import as bringing your keys and your content across reliably, and check anything that depended on more elaborate triggering rather than assuming it survived. The content is the part that took you weeks to write; that is the part that moves.

If you are going the other way, or running both, the same underlying system on the SillyTavern side is covered in SillyTavern memory, which sets out how keyword scanning and budget work there.

What You Can See Of Someone Else’s

This is one of the most-searched things about the feature, so here is the clean answer.

If the script is public, you can read it. Janitor rebuilt public script pages in May 2026, and the changelog describes the result: bigger cover images, creator details, created and updated dates, script stats, and actions for cloning or reporting. Crucially, it describes an about section for lore notes and creator context with the source section kept separate so you can inspect the lorebook or code without the page feeling like one big editor. There is also a used-by section with character cards showing which characters rely on that script. So a public lorebook is genuinely browsable, and clonable.

If it is private, you cannot. And there is no supported route to extracting a lorebook from a character you are merely chatting with. Janitor’s public character data does not carry lorebook contents — the character record has fields for description, personality, scenario, example dialogue and first messages, and no lorebook among them, which is consistent with the lorebook being a separate object rather than part of the card.

There is a third point that explains why people search for this and come up empty. Janitor’s public sitemap lists characters and profiles — not scripts. So search engines are not indexing lorebook pages, and no amount of clever searching will surface one. If you want to browse lorebooks, you have to do it from inside the site.

Building One That Works

A short method, in the order that wastes least time:

List the proper nouns first. People, places, factions, objects, events. Proper nouns make excellent keys because they are distinctive and because you naturally type them when they are relevant.
One entry per thing. Not one entry per topic. Separate entries for the city and its ruler fire independently and cost only what is needed; a single combined entry costs everything whenever either is mentioned.
Write the entry as facts, not prose. Bullets and short declarative lines. The model is not reading it for style, it is reading it for content, and the prose costs tokens for nothing.
Key generously, including the casual forms. Every way you actually refer to the thing, including nicknames, short forms and plurals.
Add secondary keys only where an entry is firing too often. They are a fix for a real problem. Applied pre-emptively they produce entries that never fire and are very hard to debug.
Test by naming things. Write a message that names each key and check that the bot’s reply reflects the entry. Do this before you build the next twenty entries, not after.
Attach it. Separate object, separate step, and the most common single reason a finished lorebook appears to do nothing at all.

Checklist

A Janitor lorebook is a script, not a character field. Profile icon, scripts tab.
Two modes: lorebooks is the visual world-building editor, advanced is a code editor for logic.
One lorebook can serve many characters, which is why it lives outside the character form.
Entries are conditional insertions. No key in the recent conversation means no entry in the prompt.
Secondary keys narrow an entry. They were fixed in September 2025, and the fix made entries fire less, not more.
Lorebook text is not in the character’s token counter, but it is in the prompt.
SillyTavern JSON lorebooks import. Keys and content travel reliably; check anything more elaborate.
Public scripts expose their source and can be cloned. Private ones cannot be read, and scripts are not in the public sitemap.
Janitor has said openly that it has not documented this properly. Treat community guidance, including this page, accordingly.

Frequently Asked Questions

What is a lorebook on Janitor AI? It is a store of world information that gets added to the prompt when something in the conversation matches one of its keys, so you can give a bot a setting, a cast and a history without putting all of it in the character definition. The thing that confuses most people is where it sits. On Janitor a lorebook is not a field on your character. It is one of the two modes of the Scripts system, which the platform introduced in its scripts beta as a visual editor mode for world-building and lore management, alongside an advanced mode that is a code editor. So a lorebook is a separate object you build and then attach, not a box inside the character form.

Where is the lorebook option on Janitor AI? Under your profile, not in the character editor. Janitor’s own announcement puts it plainly: you find scripts by clicking your profile icon and going to the scripts tab, and from there you choose between the lorebook mode and the advanced mode when you create one. People hunt for a lorebook button while editing a character, do not find one, and conclude the feature is missing or restricted to some tier. It is neither. It is in a different part of the site because it is a different kind of object, one that can be attached to more than one character.

Why is my Janitor AI lorebook not triggering? Usually because nothing in the recent conversation contains the key. A lorebook entry is keyword-activated, so an entry only enters the prompt when its key appears in what is being scanned, and the scan covers recent messages rather than the whole chat. An entry about a city is invisible until the city is named. The second common cause is secondary keys, which narrow an entry so that it needs more than the primary key before it fires. Janitor’s changelog records fixing a batch of lorebook bugs and getting secondary keys working properly, describing them as a way to be more selective about which entries make it into prompts instead of dumping everything, so an entry that fired before that work may be pickier now than you remember.

Can you import a SillyTavern lorebook into Janitor AI? Yes, and it is an advertised feature rather than a workaround. When Janitor launched the scripts beta it described the lorebook mode as a visual editor for world-building and lore management and said you can also import your existing SillyTavern JSON lorebooks. That makes SillyTavern’s lorebook format the practical interchange format between the two platforms, and it is the reason people with large existing lorebooks can move to Janitor without rebuilding them by hand. Export from SillyTavern as JSON and import on the Janitor side.

How do I see someone else’s lorebook on Janitor AI? Only if they made the script public. Janitor rebuilt public script pages in 2026 to carry creator details, dates, stats and a separate source section, described in the changelog as letting you inspect the lorebook or code without the page feeling like one big editor, along with a used-by section showing the characters that rely on it. So a public script’s contents are visible on its page. A private one is not, and there is no supported way to pull a lorebook out of a character you are chatting with. Janitor’s public character data does not carry lorebook contents, and scripts are not included in the site’s public sitemap, so search engines will not surface one for you either.

Do lorebook entries count toward the character token total on Janitor AI? Not in the token count shown on a character. Janitor’s character records break the token total into personality, scenario, example dialogue and first message, and there is no lorebook line among them. That means a character can look small in its own token counter while dropping a large amount of extra text into the prompt whenever lorebook entries fire. The budget is still real, it is just not displayed in that counter, so it is worth keeping entries short for reasons the character editor will never show you.

Is there official documentation for Janitor AI lorebooks? Yes, now. At launch Janitor chose not to document scripts, saying too much would change, and pointed people to a community guide instead. Janitor now runs a help centre at help.janitorai.com with a beginner overview of scripts and a step-by-step guide to building a lorebook. It covers who can trigger entries, the Depth setting that decides how many recent messages are scanned, and how to paste in SillyTavern entries. The more advanced material is still mostly written by the community.

What is the difference between lorebook mode and advanced mode on Janitor AI? They are two ways into the same scripts system. Lorebook mode is the visual editor, built for world-building and lore management, and it is what most people want: entries with keys and content, managed in a form. Advanced mode is a code editor for full control, intended for conditional logic rather than stored lore, such as revealing information only after a certain number of messages or reacting to particular keywords. Janitor described the advanced mode as straight code editing and the lorebook mode as the more visual of the two, with both still being made more intuitive over time.

The location and two modes of the scripts system, the secondary-keys fix, the SillyTavern JSON import and the public script page contents are quoted from Janitor AI’s own announcements, changelog and help centre (checked 23 September 2026), dated 5 September 2025, 19 September 2025 and 4 May 2026; the character field structure, token-count breakdown and sitemap contents were checked against the platform’s public endpoints on 22 September 2026. Janitor’s help centre documents who can trigger an entry and the Depth setting; exact entry limits and insertion order are not published, and no figures for them are claimed here. Janitor AI is a third-party platform and Arcanum is an independent publication with no affiliation to it or to any model provider named here; all names are used nominatively.