The Short Answer
Janitor AI is a free, web-based platform for text roleplay with community-made AI characters. Replies come from a language model you connect with an API key, or from Janitor AI's own hosted model, JanitorLLM. Tokens measure how much text that model can hold at once, which is why long chats start forgetting earlier details.
Janitor AI is one of the most popular places online for character roleplay, and one of the most confusing to set up.
This explainer gives adults plain-English answers to the questions the interface skips: what tokens are, whether you need an API key, what it really costs, who can see your chats, and why the site slows down or stops replying.
The facts come from our full Janitor AI review and the platform's own setup options. The guide finishes with a troubleshooting table and a shortlist of alternatives that need no setup. 18+ only. Last reviewed: June 2026.
What Is Janitor AI?
Janitor AI is a free, web-based platform for text roleplay with AI characters created by its community.
It hosts hundreds of thousands of user-made characters, each with a personality, backstory and scenario, and people use it for everything from adventure stories to romantic and 18+ roleplay. Sign-up needs an email and an 18+ confirmation, with no ID check.
The wrinkle that trips up newcomers is that Janitor AI mostly supplies the interface, not the "brain".
The platform manages characters, chat history, personas and lorebooks, while the replies come from a language model you choose: its own hosted JanitorLLM, or an outside model such as OpenAI's GPT, Anthropic's Claude or Google's Gemini, connected with an API key.
That split explains most of the platform's flexibility and most of its problems. It is also text-only, with no built-in voice or image generation.
What Are Tokens in Janitor AI?
Tokens are the small units of text, a few characters each, that an AI model reads and writes, and in Janitor AI they decide how much your character can remember at once. A token isn't a coin you buy; it's a measure of text.
The word "hello" is one token, and a long message is many.
What matters for roleplay is the context window. The model can only hold a limited number of tokens at a time, and that budget has to cover the character definition, your chat history and the new reply.
| Token concept | What it means for your chat |
|---|---|
| Context window | The total tokens the model can "see" at once: character, history and reply. |
| Memory limit | When the chat outgrows the window, the oldest messages drop off and the character "forgets" them. |
| Bloated character cards | A very long character definition uses up tokens, leaving less room for conversation memory. |
| Cost (with paid models) | If you connect a paid API, you are billed for the tokens going in and out. |
If a 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 recall.
Do You Need an API Key for Janitor AI?
You don't strictly need an API key for Janitor AI, because its hosted JanitorLLM model works without one, but connecting your own model usually gives faster, higher-quality and less restricted replies.
- JanitorLLM (built in): no setup, but it can be slower at peak times and more filtered than a model you choose.
- Your own API key: connect a provider such as OpenRouter, OpenAI or Anthropic for better roleplay, in exchange for setup time and per-token billing.
- Proxies: these route requests through a third party to unlock models, and the proxy operator can potentially see everything that passes through.
Connecting a key takes about ten minutes once you've done it before, and can take an hour the first time:
- Create an account with an API provider. Our review recommends OpenRouter for first-timers, since it offers many models, including free ones, and simpler billing.
- Add credit with a card to cover token usage.
- Generate an API key in the provider's dashboard.
- Paste the key into Janitor AI's settings, choose a model and start chatting.
Is Janitor AI Free?
Janitor AI itself is free; what you pay depends on the model behind your chats. Light users can spend nothing or a few dollars a month, while heavy users on premium models can pay more than a flat-rate companion subscription.
| Option | Typical cost (per our review) |
|---|---|
| Janitor AI platform | $0, with no usage limits on the platform itself |
| OpenRouter free or cheap models | $0, or about $1–$3/month for light use |
| Your own OpenAI or Claude key | About $0.10–$2.00 per day of moderate use, billed per token |
| JanitorLLM (built in) | Free, with unlimited chats; the optional Janitor+ plan costs $12.99/month or $99/year |
For comparison, Candy AI costs $13.99/month (about $3.99/month billed yearly), with unlimited text chat but images and voice paid for with tokens.
Can Janitor AI See Your Chats?
Assume Janitor AI chats are not private: the platform stores your conversations on its servers, and the model provider you connect receives every message too. Your messages pass through at least two companies, and a proxy adds a third.
- The platform keeps conversations server-side to maintain context. You can delete your data from account settings.
- Your model provider, whether OpenAI, Anthropic, Google or another, handles your messages under its own data and retention policy.
- A proxy operator can potentially see everything you send. An official model connection, rather than a random community proxy, meaningfully reduces that exposure.
On the plus side, accounts are pseudonymous, no ID is required, and your card details go to your API provider rather than to Janitor AI. Never share your real name, location or financial details in a chat.
Our guide to staying anonymous on adult sites covers the wider playbook.
The character library is user-uploaded, and its moderation has been inconsistent at times, though it has improved. Stick to fictional adult characters, and skip anything involving minors, non-consent or real people.
Why Is Janitor AI Slow or Not Working?
Most Janitor AI problems come down to server load, an expired API key or dead proxy, the token limit, or your browser, and each one has a straightforward fix. Work through the table in order before assuming the whole site is down.
| Problem | Likely cause | Fix |
|---|---|---|
| Very slow replies | Peak-time server overload, or a slow model or proxy | Try off-peak hours, switch to an API connection, or change model |
| No response at all | A site outage, an expired API key, or a dead proxy | Check the site's status, regenerate your key, swap the proxy |
| Character forgets things | The context limit has been reached | Trim the character card; summarize the story so far |
| Talks or acts for you | The model is continuing your turn | Add an instruction never to speak as your character; edit replies |
| Generic errors | Browser cache, extensions or network issues | Clear the cache, disable extensions, try another browser or device |
What Are the Best Janitor AI Alternatives?
The best Janitor AI alternatives are SpicyChat AI for a similar character library with easier setup, CrushOn AI for a curated library, and Candy AI for a polished companion with voice and images.
All three work out of the box, with no API keys, proxies or token management.
| Alternative | Why switch | Starting price |
|---|---|---|
| SpicyChat AI | Familiar Character AI-style interface, permissive content policy, usable free tier | Free (rate-limited); paid from $5.95/month |
| CrushOn AI | A more curated character library, plus metered voice and images | Free (100 credits a month); paid from $5.99/month |
| Candy AI | Voice, image generation and image-to-video in one polished app (memory is per-chat) | 5-message demo; $13.99/month (about $3.99/month billed yearly) |
If you came to Janitor AI to get away from Character AI's content rules, our explainer on whether Character AI has an NSFW filter gives the background.
Frequently Asked Questions
Janitor AI is reasonably safe with sensible habits. You sign up under a pseudonym without ID, and any card payment for models is handled by your API provider. The bigger risks are that two companies hold your chats and that community proxies can read them.
Janitor AI doesn't filter what the model generates, so adult roleplay depends on the model you connect and the character you pick; filtered models refuse at certain boundaries. The platform is for over-18s only, and roleplay should stay with fictional adult characters.
No, Janitor AI is text-only. Users who want spoken messages or character pictures either pair it with a separate image tool or move to a companion app that includes both, such as Candy AI or DreamGF .
Yes. SillyTavern character cards, saved as PNG files, can be imported into Janitor AI. Character AI doesn't export characters, but the community has recreated many popular ones in the Janitor AI library.
Experienced users rate Anthropic's Claude models highly for creative writing, and OpenAI's GPT-4-class models work well for general roleplay. On a tight budget, OpenRouter's cheap and free models are adequate for most chats.
Wrapping Up
Janitor AI is powerful but unpolished. Once you know that tokens are really about memory, that your chats pass through more than one company, and that most "it's broken" moments trace back to servers, your model setup or your browser, it becomes far less frustrating.
Keep your context lean, treat anything you type as potentially logged, and work through the fixes in order before assuming the site is down. If the tinkering wears thin, a purpose-built companion app offers the same kind of roleplay with no setup.
Our explainer on how AI companion apps work is the next place to look.
