Janitor AI Script: The Honest 2026 Review

A Janitor AI script is the underlying configuration — often called a system prompt or personality text — that instructs the language model how to behave when responding to users. Unlike all-in-one platforms like Character.AI or Replika, Janitor AI is a front-end that routes your messages through an external LLM (typically OpenAI's GPT-3.5/GPT-4 or Kobold AI). The script is not a fixed file but a set of instructions you paste into the character's definition box. It includes the character's name, appearance, personality traits, speech patterns, and backstory. The script also sets behavioral rules — for example, whether the character speaks in first person, uses slang, or avoids certain topics. Without a well-crafted script, the AI defaults to generic, boring responses. Writing a good script is the difference between a character that feels alive and one that sounds like a customer support bot. Because Janitor AI offloads all generation to your own API key, the script is your only tool for shaping output quality — there's no built-in moderation or style layer on top.

Writing a Janitor AI script requires precision. Start by opening the character creation panel and locating the 'Personality' or 'Definition' field. This is where your script goes. Begin with a one-line summary of the character's core identity: 'You are a sarcastic 25-year-old barista named Chloe who works at a busy café.' Then expand with bullet points or short paragraphs covering appearance, personality traits, speech style, likes, dislikes, and backstory. Example: 'Appearance: shoulder-length blonde hair, always messy bun. Speech: uses slang like 'dude' and 'literally,' interrupts frequently. Backstory: dropped out of art school, hates corporate customers.' For advanced control, include explicit behavioral rules: 'Never speak in third person. Never ask what the user wants to do. Stay in character at all times.' Some users add a section called 'Constraints' that lists forbidden behaviors. The script length matters — for GPT-4, 500-1000 tokens is ideal; for GPT-3.5, keep it under 400 tokens to avoid truncation. Test your script with sample messages before saving. If responses feel off, adjust the script — add more examples of dialogue or clarify ambiguous instructions.

Even experienced users hit script errors. The most common issue is the model ignoring the script entirely, often because the script exceeds the model's context window. GPT-3.5 has a 4096-token limit including conversation history; if your script plus past messages exceed this, the model drops earlier parts. Solution: keep scripts under 400 tokens for GPT-3.5. Another frequent error: the character speaks generically despite a detailed script. This happens when the script lacks explicit speech examples. Add 3-5 sample dialogue lines: 'Example line: "No way you just said that. I can't even."' The third common failure: the model contradicts the script mid-conversation. This stems from the model's built-in tendencies overriding weak instructions. Reinforce key traits in every script section — don't assume one mention is enough. If using OpenAI's API, set the 'system' role message to your script and the 'user' role to the chat history. For Kobold AI users, the script goes into the 'Character Definition' field. Finally, check your API key permissions — a key with usage limits or expired credits returns empty responses, which users mistake for script errors.

For roleplay-focused scripts, specificity beats generalities. Instead of 'friendly waitress,' write: 'You work the night shift at a diner called Mel's. You call everyone 'hon' and refill coffee without asking. You're tired but keep smiling.' Include emotional triggers: 'You get defensive if anyone criticizes your cooking.' This gives the model emotional levers to pull. Use 'show, don't tell' — demonstrate traits through example dialogue rather than labels. 'You say 'uh-huh' a lot when customers complain' is better than 'you have low patience.' For romantic or NSFW roleplay, explicitly state the character's consent boundaries and tone: 'Flirty but not aggressive. Let the user initiate physical contact.' Janitor AI does not filter content, so your script must set all guardrails. Test scripts with multiple conversation starts — a script that works for a coffee shop scene may fail in an emergency scenario. Many users keep a library of scripts for different scenarios and swap them via the character editor. Version your scripts (e.g., 'Chloe_v3') to track what works. Community forums on Reddit and Discord share successful scripts; adapt them to your character's voice rather than copy-pasting.

Janitor AI does not host its own language model — it's an interface that sends your script and chat history to an external LLM via API. This means every response consumes tokens from your own API credit balance. OpenAI charges roughly $0.0015 per 1K tokens for GPT-3.5 and $0.03 per 1K tokens for GPT-4. A typical 20-message conversation with GPT-3.5 costs about $0.03. Kobold AI offers free tiers but with slower speeds. The script itself counts as tokens — a 500-token script adds $0.00075 to each API call. The advantage: you control the model and its parameters (temperature, top P, frequency penalty). The disadvantage: setup friction. New users often struggle with obtaining an API key, setting up billing, and ensuring the key has sufficient funds. Without a valid key, Janitor AI returns an error. Alternatives like AIAngels include built-in language models with no API key required — you pay a flat subscription and start chatting immediately. If you value control and low cost per conversation, Janitor AI's API model is powerful. If you value simplicity, all-in-one platforms eliminate the script-and-key overhead.

On Character.AI, configuration is limited to a few text fields — description, greeting, and short 'definition' — and the platform's proprietary model handles generation. You cannot inject detailed system prompts. Replika uses a curated model with built-in memory and personality sliders; scripts are impossible. Candy.AI and DreamGF offer no script access at all — you pick from presets. Janitor AI's script system offers unparalleled flexibility: you can design a character with hundreds of tokens of explicit instruction, control model temperature, and swap models on the fly. The trade-off is complexity. Platforms like AIAngels offer a middle ground: you customize appearance, personality, and interests through structured forms (not raw scripts), and the built-in model interprets those settings without demanding an API key. For users who want deep control and don't mind technical setup, Janitor AI's script approach is unmatched. For users who want a working companion in 30 seconds, platforms with pre-configured models and no API friction are the better choice. The right pick depends on whether you enjoy tinkering or just want to chat.