Janitor AI exploded in popularity as a home for character-based roleplay, but it is also one of the more confusing AI chat platforms to actually use. New users immediately run into questions the interface never quite answers: what are these "tokens" everyone talks about, do I need an API key, can the people behind the site read my private chats, and why is it suddenly so slow or refusing to respond? This guide answers all of that in plain language. You will learn what tokens really mean for your conversations, how the privacy picture actually works, the most common reasons Janitor AI breaks and how to fix each one, and where to go if you want something more reliable. 18+ only. Last reviewed: June 2026.
What is Janitor AI?
Janitor AI is a platform for chatting with user-created AI characters, widely used for roleplay including NSFW scenarios. Instead of one assistant, it offers thousands of community-made characters, each with its own personality and backstory, that you talk to in an ongoing conversation.
The wrinkle that trips up newcomers is that Janitor AI does not always run its own language model for you. To get good, uncensored responses, many users connect it to an external model through an API key or a proxy, which is where most of the platform's complexity — and most of its problems — come from. Understanding tokens and setup is the key to making it behave.
What are tokens in Janitor AI?
Tokens are the units of text an AI model reads and writes — roughly a few characters each — and in Janitor AI they govern how much your character can "remember" and how much you can say. A token is not a coin or a currency you buy; it is a measure of text. The word "hello" is one token; a long message is many.
Why this matters for roleplay comes down to the context window: the model can only hold a limited number of tokens at once, covering the character definition, your conversation history, and the new reply. Two practical consequences follow.
| Token concept | What it means for your chat |
|---|---|
| Context window | The total tokens the model can "see" at once — character + history + reply. |
| Memory limit | When the chat exceeds the window, the oldest messages drop off and the character "forgets" them. |
| Bloated character cards | A huge character definition eats tokens, leaving less room for actual conversation memory. |
| Cost (with paid models) | If you connect a paid API, you are billed per token in and out. |
The takeaway: if your character keeps forgetting things, you are hitting the token limit. Trimming a bloated character card or summarizing the story so far frees up context and improves memory.
Setting it up: API keys and proxies
Janitor AI lets you generate responses in a couple of ways, and which you choose affects quality, cost, and reliability.
- The built-in option is the simplest — no setup — but it can be slower, more heavily loaded, and more filtered.
- Connecting an external model via API (your own key with a provider) generally gives faster, higher-quality, less-restricted responses, at the cost of setup and possible per-token billing.
- Proxies route requests through a third party. They can unlock models but carry a real privacy cost — see the next section, because a proxy operator can potentially see what passes through it.
If the words "API key" and "proxy" already sound like more work than you want, that is a sign a dedicated companion app may suit you better — they handle all of this invisibly.
Privacy: can Janitor AI see your chats?
Treat Janitor AI chats as not private. However immersive a roleplay feels, your messages pass through servers, and depending on your setup, more than one party can potentially access them.
Two layers matter. First, the platform itself: like most web services, it can log activity, and you should never assume your conversations are invisible to operators. Second, and more important, your setup: if you route requests through a proxy, that proxy's operator can potentially see everything you send, including explicit content and anything personally identifying. Using a reputable, official model connection rather than a random community proxy meaningfully reduces that exposure.
The safe behavior is the same as anywhere else online: do not share your real name, location, financial details, or anything you would not want tied back to you. For the broader playbook on keeping adult activity separate from your real identity, see our guide on staying anonymous on adult sites.
Why Janitor AI is slow or not working — and how to fix it
Most "Janitor AI is broken" moments come down to a handful of causes, each with a straightforward fix. Work through them in order before assuming the site is down.
| Problem | Likely cause | Fix |
|---|---|---|
| Very slow replies | Server overload at peak times, or a slow model/proxy | Try off-peak, switch to an API connection, or change model |
| Not working / no response | Site outage, an expired API key, or a dead proxy | Check status, regenerate your key, swap the proxy |
| Character forgets things | Token/context limit reached | Trim the character card; summarize the story so far |
| Talks/acts for you | The model is continuing your turn | Add an instruction to never speak as your character; edit replies |
| Generic errors | Browser cache, extensions, or network issues | Clear cache, disable extensions, try another browser or device |
If you are constantly fighting one of these, that is the platform telling you its rough edges may not be worth it for your use case.
Janitor AI alternatives
Janitor AI rewards tinkering, but plenty of people just want immersive AI chat that works out of the box, with no API keys, proxies, or token wrangling. Dedicated companion apps fill that gap, handling the model, memory, and safety setup for you.
- AI companion apps offer polished roleplay, voice, and image generation without any technical setup. Our best AI girlfriend apps roundup and the wider AI companion category compare the top options.
- AI sexting apps focus specifically on explicit chat — see our tested best AI sexting apps roundup.
- Character AI alternatives are a good starting point if you came to Janitor AI to escape filters — our Character AI alternatives list covers them, and our explainer on the Character AI NSFW filter gives the background. If you switched from Replika, our Replika NSFW guide covers what changed and where to go next.
Janitor AI FAQ
Quick answers to the questions people ask most.
What are tokens in Janitor AI? Tokens are units of text (a few characters each) that determine how much your character can remember at once. Hitting the token limit is why characters "forget" earlier parts of a chat.
Can Janitor AI see my chats? Assume yes. Messages pass through servers, and if you use a proxy, that operator can potentially see everything. Never share identifying or sensitive details.
Why is Janitor AI so slow? Usually server load at peak times or a slow model/proxy. Try off-peak hours, connect an API, or switch models.
Why is Janitor AI not working? Common causes are site outages, an expired API key, or a dead proxy. Check the platform's status, regenerate your key, and try a different connection or browser.
Is Janitor AI free? The platform is free to use, but connecting certain external models via API can incur per-token costs from that provider.
Wrapping up
Janitor AI is powerful but unpolished: once you understand that tokens are really about memory, that your chats are not as private as a roleplay feels, and that most "it's broken" moments trace back to servers, your model setup, or your browser, the platform becomes far less frustrating. Keep your context lean, treat anything you type as potentially logged, and work through the fixes methodically before assuming the site is down. And if the constant tinkering wears thin, remember that purpose-built AI companion apps deliver the same kind of roleplay with none of the proxy setup — sometimes the best fix for a finicky tool is a tool that just works.