How to use Janitor.AI to RP with ChatBots? - Cara's Guide for Beginners
I've noticed a lot of people have never ventured far from c.ai before (which I get, but you guys don't know what you're missing out!) so I decided to make this little tutorial for anyone who wants to give Janitor.AI a try.
DISCLAIMERS! READ THIS!
Janitor.AI is NOT a website for minors. Unlike with c.ai, you can write smut/NSFW content with the bots.
"Well, Cara, I'm an adult (+18!), so what does that mean for me?"
It just means you should be mindful of bot's tags and stuff. Most people have a brain in their heads and don't use inappropriate pictures for their bots, not even the ones JUST FOR SMUT. But some of them... do. And Janitor's AI can only flag so many of them.
Why should I try out Janitor.AI? I'm happy with C.ai!
Well, honestly, that's okay. If you like c.ai and have a good experience with it, more power to you. Me, personally? I was unfulfilled. The smut/adult content constraint was a bother, sure, but I dealt with it just fine. What made me look for other options was that the quality of the answers from the bots just wasn't what I wanted. They weren't long enough, they weren't deep enough, it just seemed all very superficial.
That said, J.ai IS different from C.ai. So let's move on to the tutorial part of this tutorial! It's under the cut, for the sake of the few people that follow me.
Index (the topics that are talked about under the cut):
Access JanitorAI and create an account.
Create a Persona.
Search for the Characters you want (using filters).
Pick a bot!
Let's write? Nope, not yet.
Advanced/Custom Prompt.
Tweak your settings.
Start writing.
Chat Memory.
What you need to immediately start writing with bots:
1. Access JanitorAi and create an account! You need a valid e-mail (they'll send you a confirmation code) and you need to be eighteen. That's about it.
2. Create a Persona! Much like with c.ai, you can have different personas in your account. Your "default persona" is attached to your PROFILE. So I personally don't mess with it. I just added a picture to my profile and a name and that was that. So, let's make a new persona, shall we?
Click on your profile on the upper right, select "My Personas" and then click "Create Persona". You'll get the pop up window in the picture above.
Name: I'd personally enter ONLY your character's first name or what your character usually goes by. Like, if your character's legal name is "Alexandra" but people rarely call her that, she goes by "Lexi", just add "Lexi" to this part.
Select Pronouns. Give your Persona a dandy picture.
Groups: When you start having a lot of Personas, dividing them by groups can make your life easier (especially if like me you have one Persona that you like to use for different 'universes' so her information will vary slightly). For now, don't worry about this.
Description: This is the important part! The LLM (Language Model - the thing that will read your messages and answer them) will look at this and it'll know/remember things about your character like eye/hair color and even scent. There's no WRONG way to do this, but this (a template) is what I found out works out for me!
When you're done, your Persona should look something like this:
Yay! Persona created. Let's move on. Click that "Save" button!
3. Search for the Characters you want! So, like a proud Dean girl, let's go ahead and search for our man, right?
Well damn, would you look at that?
I know, I know, we're all eager, but HOLD YOUR HORSES. Let's refine this search a little, okay?
First, understanding what all of this means:
Limited Only: These are bots that don't allow NSFW/Smut. So if you're looking for bots with a familial/platonic vibe, they're usually here.
The tags: You can click them on the bar right below the search bar and search exclusively for Dean Winchester bots that have the tag "Smut", for example.
STILL ABOUT THE TAGS: Notice how each bot has Tags underneath their description. They're important. Some bots are MLM (male characters made to be shipped with male personas). Some are WLW, some are AnyPov (Personas with any gender assigned to them can use it. They will usually start the RP referring to your Character as "they/them", but it'll adjust with time). Most importantly: SOME BOTS HAVE A TAG CALLED "Dead Dove". What does this mean? Dead Dove is an umbrella term for anything that the reader might consider disturbing. This could be anything from inc*st to non-c*n to... idk, cann*bal*sm. People will write anything. I'm not saying to avoid or block these tags, just be mindful of them and read the Bot's description to find out exactly why they have this tag. ;)
On the upper left corner, there's this little button:
Click that. This is where the good stuff is!
Availability: All / Limited Only. Like I explained, this one will filter out bots that allow smut.
Sort by: - Popular (if it's your first time searching for that character, start with this. However, this might include bots that were made a long time ago and had different quality standards than they do nowadays!) - Creation Date (will show you bots that were JUST made! FRESH OUT OF THE OVEN. This is my go to if I'm searching a character that I've searched a lot of times before. Like, I went through all of it already, so lemme see what's new.) - Trending / Trending 24h (... I honestly don't know. I think it's self-explanatory, but I rarely use these, if I'm being honest.) - Relevance (Self-explanatory as well, but again, I rarely use it. I tend to swap between the first two options).
Messages: - Greater than or equal / Less than or equal: Choose the parameter. - Enter message count. (This is also a parameter I rarely mess with. Having only a few messages isn't indicative to a low quality bot, it might just mean it's new or niche. And it's the same for the opposite: bots with tons of messages sometimes just have been around for too long or are, um, *cof* Smut Slop *cof*.)
Tokens: - Greater than or equal / Less than or equal: Choose the parameter. - Enter token count. THIS IS THE IMPORTANT STUFF!!!! Now pay attention!
WHAT ARE TOKENS AND WHY ARE THEY IMPORTANT?
I'll make this quick and easy:
Every word you type out for a bot, be that creating it or chatting with it, is turned into tokens.
JLLM (The Language Model JanitorAI uses by default and that you can use for free with their website) holds between 8,000 to 9,000 tokens per chat.
As a rule of thumb, for the English language, 1 token ≈ 4 characters.
So... why is this important for bots?
Short answer: Memory and Quality. Longer answer: If you want to know more about Tokens and what they are, how they work, etc, read Janitor's Guide about it here! They have a really cool, easy to understand explanation over there. It explains what tokens are, what they do and the difference between temporary/total and permanent tokens!
Here's what you need to know if you just want to get to writing:
When you click a bot and open their page to see their Description, Comments, etc, you can see the amount of "Total and Permanent Tokens", like this:
In my experience as a botmaker and someone that uses them:
If you're looking for SERIOUS/LONG TERM RPs with LONG MESSAGES, don't go for anything under 800 Permanent tokens.
If you want a bot that doesn't have the memory of a goldfish, don't pick anything above 3.5k total tokens.
So, the screenshot above is an example of a fairly decent (at least upon first glance) token count for people who, like me, enjoy a more in-depth RP experience without excessive, needless information.
What are examples of bots that are most likely bad?
Not enough tokens:
^ This basically means that Dean's personality was probably described in five words, the bot's first message is about two sentences long and that's about it.
Why is that bad? 'Cause while Dean is a canon character that Janitor's LLM SHOULD know... it might not. This 130 token Dean will probably be Dean in name only. Forget about referencing anything in Canon, it'll probably fill the gaps with made up stuff.
Too many tokens:
My brother in Christ, did you just copy/paste the Wiki in here?
Anyways. This just... won't work. Not only the LLM won't absorb all this information, it'll start saying gibberish and it will forget anything you say to it in about 3 messages. Kinda like Dory introducing herself over and over.
So, what's ideal?
This is a personal preference thing. I usually go for 1k minimum for total tokens, 3k max. Find out what works for you.
4. Pick a bot and let's chat! For this example, I'll link one of my Dean bots. Here he is.
What's in here? The bot description, which is where botmakers will give you a quick rundown of what to expect from the bot. Mine has the core information, the First Message (so people know exactly what Persona to use) sent by the bot, a few other notes and tips.
Scroll alllll the way down and find this:
By default, Janitor will pick your "Default Persona" (the one that is your profile) to chat. If you want to change it to the Persona you made, click the "Playing as" button and pick your preferred Persona.
Then click "Chat with Dean".
5. LET'S WRITE! Right? ... Well, you can. But if you want the optimal Janitor experience, let me give you a few more tips.
You see this way up there on your screen?
There's a button that says "Using Janitor" and a little hamburger menu next to it. Click "Using Janitor" first.
AH, WHAT ARE ALL THESE WORDS? HELP!
I know, it's gross, we just want to nerd out with our favorite characters. But this is worth it, I promise.
So this is for people who will be using JANITOR'S LANGUAGE MODEL, usually referred to as JLLM.
6. Click "Advanced Prompt".
It'll open a text box. This is where you can give the LLM some instructions to refine the type of answer it'll give you. Like length, tone, narration style, etc.
If you have no idea what to put here, you can copy my prompt HERE. Give it a read before using it, figure out what works FOR YOU and change it however you want. Just, as always, keep it short and sweet. Bullet points and direct information are your friends with AI.
7. Save your Settings and click the Hamburger Menu. (☰)
First thing: UNTOGGLE "TEXT STREAMING". It improves performance and it's honestly just useless in my opinion.
Click Generation Settings:
These are my settings.
Temperature: 0.9
Max tokens: 0 (Default is 500. If you like longer answers, this might cause issues. 0 is safer).
Context is JLLM's default context. What does this mean? It's about tokens, so you can read more about it here. Basically? It's Memory.
Show thinking: toggle it off. This basically means that when the bot sends you a reply, it won't show you it's thought process. I choose to leave it off because it ruins immersion for me.
8. You can start writing!
I'm not kidding, you're done, this is it. Start writing. The following step is for people who, like me, like to get caught up in LONG TERM RPs. Like... Over 100 messages with the same bot in the same chat.
9. About Chat Memory:
So, this is how I USE IT!
If you click the Hamburger Menu (☰), you'll see that it has a "Chat Memory" option.
Upon clicking it, you'll see the following window.
This is where you add a summary of your RP with the bot when it starts getting too long.
What I recommend is to update this every time you finish a "scene" (move from one location to another, end a conversation, etc. Think of it like a movie or a show). If you're writing a super long "scene", check this window and see how many tokens need to be summarized. When you hit about 2-3k tokens, SUMMARIZE YOUR RP.
This is the template I use for my Chat Memory. I used Mrshmellow's Chat Memory template and tweaked it to my liking. You're welcome to use mine.
Update this frequently and you should be fine!
Does this mean the AI will be PERFECT now?
No, far from it. AI is really dumb actually. If it wasn't we wouldn't have to do all this. It might still say some gibberish, get stuff wrong, write your character's dialogue/actions. If it does that:
One star the reply and generate another one by clicking the little arrow below the reply.
Or, just edit the reply and take out any parts in which the bot speaks for you.
What you don't do:
Complain to botmakers that the AI is talking for you. Most likely it's not their fault, it's LLM silliness.
THIS IS IT! The end.
If you guys need a Proxy tutorial, if you have ANY QUESTIONS, feel free to send me an ask! <3