How to Use Janitor AI Chat: Beginner's Guide to AI Character Roleplay on Mobile and Web

Janitor AI looks simple on the surface. Pick a character, start typing, watch it respond. But if you want characters that actually stay in character, conversations that don't loop, and roleplay that feels like collaborative storytelling rather than a chatbot guessing what you want, there's a learning curve.

This guide walks through everything from creating your account to dialing in settings that make the AI feel less robotic. New user or frustrated by flat responses, here's what makes the difference.

Creating Your Janitor AI Account

Go to janitorai.com and click Sign Up. You can register with an email address or link a Google account. Email registration requires a verification step, check your spam folder if the confirmation email doesn't show up in a few minutes.

One thing the platform doesn't advertise upfront: you need to be 18 or older. Janitor AI's content policy allows NSFW material, so age verification matters. If you're under 18, the platform won't work for you.

Once you're in, you'll land on the main character browsing page. The default view shows trending and popular characters. Don't jump into a chat yet. Take two minutes to set up your preferences.

Setting Up Your Account for Better Responses

Navigate to Settings (gear icon, usually top right). The most important settings live here:

API Settings. This is where you choose which AI model powers your chats. The default is JanitorLLM, which is free but inconsistent. If you want better responses:

Select "OpenAI Compatible" or "Reverse Proxy" as the API type
Enter your API key and proxy URL if using an external model like DeepSeek or Claude
If you don't have an API key, stick with JanitorLLM and adjust the generation settings instead

Generation Settings. Temperature controls creativity. Higher values (0.8-1.0) make responses more varied and surprising but sometimes nonsensical. Lower values (0.5-0.7) make them predictable but consistent. For roleplay, start at 0.8 and adjust based on how the character responds. Max output tokens determines how long responses can be, set this to at least 400 for proper roleplay paragraphs.

Persona Settings. This is your character in the roleplay, what the AI sees when it looks at you. Fill in your name, description, and appearance. The more detail you provide, the more the AI references your persona naturally in conversation. A one-line persona gets one-line acknowledgement. A paragraph of personality traits, appearance details, and background gets woven into the story.

How to Have Janitor AI Speak to You

Here's the lesson most new users learn the hard way: the AI's response quality depends almost entirely on your opening message.

The first message sets the tone. If you type "hi," the character replies "hi." If you write three paragraphs of atmospheric description that establishes a scene, introduces your character, and gives the AI something to react to, you get three paragraphs back. The AI mirrors your effort level.

An effective opening message does three things:

1. Establishes where the scene takes place

2. Describes what your character is doing

3. Gives the AI character a clear reason to respond

Instead of: "Hey, how are you?"

Try: The tavern had gone quiet when she walked in. Rain dripped from her cloak onto the wooden floor, and every pair of eyes in the room tracked her movement toward the bar. She didn't sit, she leaned against the counter, close enough that you could smell the storm on her clothes. "You're the one they said could help."

The AI now has context, atmosphere, and a clear prompt to work with. The response will be dramatically better.

Guiding the conversation without breaking immersion. If the AI character starts getting repetitive or veering off-track, you can course-correct without breaking character. Add an OOC (out of character) note at the end of your message in parentheses: (OOC: remember that we're in the library, not outside). The AI picks up on these cues and adjusts.

Using narration alongside dialogue. The most engaging Janitor AI chats blend dialogue and narration. Describe what your character sees, feels, and does, not just what they say. The AI responds in kind, building a scene rather than just exchanging lines of dialogue.

Browsing and Finding Characters

The character browser has search, tags, and trending sections. Tags range from "fantasy" and "romance" to hyper-specific genres like "slow burn" and "enemies to lovers." Use them to narrow the library.

When evaluating a character before chatting:

Read the character definition. A well-made character has a clear personality description, backstory, and conversation style notes. Vague definitions produce vague responses.
Check the token count. Characters with very short definitions won't have much personality. Aim for characters with at least 500 tokens in their definition.
Look at the opening message. It shows you what the creator expects from the roleplay. If the opening message matches the vibe you want, the character probably does too.

Creating Your Own Janitor AI Character

Creating a character goes beyond picking a name and an avatar. The character definition field is where the AI gets its personality from. Fill it with specific details:

Personality traits with examples of how they manifest ("stubborn, refuses to admit when she's wrong, even when the evidence is right in front of her")
Speech patterns ("uses formal language with strangers, drops into casual slang when comfortable, calls everyone 'darling' when annoyed")
Backstory elements that shape behavior ("grew up in a fishing village, still flinches at thunderstorms because of the shipwreck that killed her father")
Behavioral rules ("never initiates physical contact, deflects personal questions with humor, protective of children to the point of recklessness")

The more specific and behavioral your descriptions are, the more consistently the AI roleplays the character. Vague definitions like "nice, friendly, smart" give you a generic chatbot. Specific definitions give you a character.

Common Chat Problems and Solutions

The AI forgets what happened three messages ago. JanitorLLM's context window is limited. External models like DeepSeek and Claude have longer memories. If you're on the free model, periodically summarize key plot points in your narration to refresh the AI's memory.

Responses feel repetitive or generic. Raise the temperature slightly (0.85 to 0.95). If it's still repetitive, edit your last message to include more specific detail and descriptive language. The AI uses your writing style as a prompt, better input, better output.

The character breaks character or says something nonsensical. Regenerate the response (the refresh button next to messages) or edit your last message to add an OOC correction. Some models are more prone to this than others. Claude and DeepSeek tend to stay in character better than JanitorLLM.

Conclusion

Getting good results from Janitor AI chat comes down to two things: writing strong openings and setting up your API properly. Most of the complaints about "bad AI" trace back to the default JanitorLLM model and low-effort first messages. Spend ten minutes configuring a better model through the proxy settings and another five minutes writing a detailed opening scene, and the platform transforms from a chatbot into an interactive storytelling engine.

If you're new to Janitor AI, you can download the app from APKPure.