What Is Janitor AI? How It Works, Why It Needs a Proxy, and Whether It Is Actually Safe

Janitor AI is one of those platforms that millions of people use and almost nobody explains properly. Search for it and you land on setup tutorials, proxy configuration threads, and Reddit arguments about which model to plug in.

Very few of them start at the beginning.

So here is the beginning. Janitor AI is a character-based AI chat platform. You pick a character — created by another user, or by you — and you have a conversation with it. The character has a personality, a backstory, and a way of speaking that stays consistent across the chat. That is the entire product.

What makes it complicated is everything around that: the free model that runs out, the API key people keep telling you to get, the proxy links that circulate on Discord, and a content policy that is considerably more permissive than Character AI's.

Here is what Janitor AI actually is, how the chat works, why so many guides tell you to connect an external model, what the Reddit community is really discussing, and — the part most articles skip entirely — whether it is safe to use.

What Janitor AI Actually Is

Janitor AI is a website where users create and share AI characters, then chat with them. It launched in 2023 and grew quickly during the period when Character AI was tightening its content filters, which is not a coincidence — a large part of its early audience arrived looking for somewhere less restricted.

The structure is simple. There is a public directory of characters. Each character is a text definition written by a user: a name, an avatar, a personality description, a scenario, and example dialogue that teaches the model how that character should speak. You browse the directory, open a character, and start typing.

The important structural difference from most AI chat apps: Janitor AI is largely a front end. It provides the interface, the character library, the memory handling and the community. The actual language model generating the replies can be Janitor's own, or one you supply yourself. That single design decision explains almost every confusing thing about the platform, including the proxy situation covered further down.

Characters range from original creations to fan characters from anime, games and books, to practical assistants. A significant share of the directory is romantic or adult-oriented roleplay, and the platform permits mature content behind an age gate — which is the main reason people compare it to Character AI in the first place.

The company has not published verified user numbers, so treat the follower counts and "millions of users" claims you see in other articles with appropriate suspicion. What is observable is that the character directory contains a very large catalogue and the community forums are consistently active.

How the Janitor AI Chat Actually Works

Understanding the chat mechanics is what separates people who enjoy the platform from people who bounce off it in ten minutes.

Characters are prompts, not programs. When you open a character, its definition is sent to the language model as instructions before your message. The model is not "loading a character" in any technical sense — it is being told, in text, how to behave. This is why a well-written character feels alive and a badly written one drifts into generic chatbot replies within a few exchanges.

Personas are your side of it. You can define who you are in the conversation — a name, a description, traits. The model receives this the same way it receives the character definition. Users who set up a persona get noticeably more coherent roleplay, because the model is no longer guessing who it is talking to.

Memory is a context window, not a database. The model sees a rolling window of recent messages plus the character definition. Older messages fall out of that window. This is the source of the single most common complaint on the forums — the character "forgot" something from earlier. It did not forget in any human sense; the text simply stopped being included. Features like pinned memory and summaries exist to mitigate this, and they help, but no current setup gives a character perfect recall of a long conversation.

You can regenerate and edit. Replies can be regenerated for a different take, and most users edit the character's messages when the roleplay goes somewhere they did not want. Editing is not cheating — it is how experienced users steer long conversations, because a corrected reply becomes part of the context the model learns the tone from.

If you want the full walkthrough — creating a character, writing a good definition, configuring the model, and fixing the errors that show up along the way — that is a separate, much longer piece.

Is Janitor AI Free? The Proxy Question, Explained Properly

Yes, Janitor AI is free to sign up and free to use with its built-in model. The confusion starts immediately after that.

Janitor provides its own hosted model, generally referred to in the community as JanitorLLM. It costs nothing. It is also shared infrastructure serving a very large number of concurrent users, which means it queues at peak times and its response quality is not on the level of a frontier commercial model. For a lot of casual users, that is completely fine.

For everyone else, the platform lets you point it at a different model. That is where the API key and proxy conversation comes from.

What a "proxy" means here: instead of using Janitor's model, you configure the site to send your conversations to another model — one you pay for through a provider's API, or one you run yourself on your own machine. The proxy is the connection layer that routes the request. You are, in effect, using Janitor AI as the interface and someone else's model as the brain.

The three routes people take:

Your own API key. You sign up with a model provider, generate an API key, and paste it into Janitor's settings. You pay that provider directly for what you use. This is the legitimate, private option — your key, your account, your billing.

Running a model locally. Tools like KoboldAI let you run an open model on your own hardware and connect Janitor to it. No usage fees and nothing leaves your machine, but you need a capable GPU and patience for the setup.

A shared or public proxy. These are links that circulate in community spaces, pointing at someone else's endpoint that anyone can use. They are free and they are the option we would tell you to think hardest about, for reasons in the safety section below.

None of this is required. If you never touch a proxy setting, Janitor AI still works. The proxy conversation dominates search results because the people writing about the platform are its heaviest users, and heavy users hit the free model's limits first.

Janitor AI vs Character AI — The Difference That Actually Matters

These two get compared constantly, and most comparisons stop at "Janitor allows NSFW and Character AI does not". That is true and it is also the least interesting difference.

Character AI is a closed, managed product. One company controls the model, the filters and the experience. You cannot swap the model. In exchange you get something that works immediately, is consistently fast, and has meaningful safety systems — including protections aimed at younger users, which the platform has invested in significantly after sustained public and regulatory pressure.

Janitor AI is an open front end. You can change the engine. Its content policy is far more permissive. The trade is that quality depends on your configuration, the free tier queues, and the safety systems are considerably lighter.

In practice the choice usually comes down to what kind of user you are. If you want to open an app and immediately talk to a character with no configuration, Character AI is the better product. If you want control over the model and fewer content restrictions, and you are willing to configure things and accept the responsibility that comes with a lighter-touch platform, Janitor AI is built for that.

Neither is "better" in the abstract. They optimise for different users, and a good number of people use both for different purposes.

What Reddit Actually Says About Janitor AI

"Janitor AI Reddit" is one of the highest-volume searches attached to the platform, and it is worth understanding why people go there instead of the official documentation.

The subreddits function as the platform's real support desk. When the site goes down, when a model change alters how characters behave, when an error message appears that the help pages do not cover, Reddit is where the answer appears first — usually within the hour, usually from another user.

The recurring themes, if you read enough of it:

Model quality debates. A constant, sometimes heated comparison of which backend produces the best roleplay, and disappointment whenever the free model's behaviour changes. Users notice model changes very quickly, because a character that felt right yesterday suddenly writes differently.

Proxy troubleshooting. Connection errors, rate limits, keys that stop working. A large share of daily posts.

Character writing craft. Genuinely useful threads about writing definitions that hold up over long conversations — this is where the community is at its best, and it is better material than most published guides.

Downtime reports. The fastest way to check whether the site is broken or your setup is.

Two cautions, because this matters. First, Reddit threads are a snapshot of a moment — a proxy or workaround that worked in a six-month-old thread may be dead or unsafe now, and old posts rank well in search long after they stop being true. Second, community spaces are where shared proxy links and API keys get passed around, and some of those are actively harmful. Read Reddit for troubleshooting and craft. Be far more careful about what you paste into your settings from it.

The Discord Server and Where the Community Lives

Janitor AI runs an official Discord server, and it serves a different purpose from Reddit. Discord is real-time: announcements, status updates during outages, character-sharing channels, and direct back-and-forth with other users while you are actively trying to fix something.

It is generally the fastest place to find out whether an outage is platform-wide, and the first place model and feature changes get announced.

The same caution applies, more strongly. Discord is where free proxy links spread most aggressively, often in unofficial servers that present themselves as helpful communities. Treat an endpoint from a Discord message the same way you would treat a login link from an unsolicited email — the fact that it works does not mean it is safe.

Always confirm you are in the official server rather than a lookalike. Impersonation servers exist for platforms of this size, and their purpose is generally to harvest API keys.

Is Janitor AI Safe? An Honest Answer

This is the section most Janitor AI articles skip, and it is the one that actually affects you. The honest answer is that the platform itself is not malware, but there are real risks — and the biggest ones come from things users add to it rather than the site itself.

Third-party proxies are the largest risk, by a distance. When you route your chats through someone else's endpoint, that person can see everything you send. Every message, every persona detail, everything personal you disclosed to a character at 2am. Some public proxies also harvest the API keys pasted into them — and a stolen key is billed to your account, not theirs. If you use an external model, use your own key from the provider directly, or run the model locally. A free endpoint from a stranger is not free; you are paying with your conversation history.

Your conversations are not private in the way people assume. Chats pass through servers. Read the platform's current privacy policy rather than trusting a summary — including this one — because these policies change. The practical rule: do not put your real full name, address, workplace, financial details, passwords or anything you would not want linked back to you into an AI chat, on any platform.

It is an 18+ platform, and the age gate is self-declared. Janitor AI permits mature content and requires users to confirm they are adults. That confirmation is a checkbox. It is not verification. If you are a parent, this is the part worth knowing — this category of site is not built with minors in mind and does not reliably keep them out. The wider research on AI companionship and younger users is worth reading if that is your concern.

Content moderation is comparatively light. A permissive policy is the platform's main draw, and the direct consequence is that you will encounter user-created characters designed around themes you may find distasteful or upsetting. There is no algorithm reliably filtering that for you before you see it.

The emotional dimension is real and not a joke. Character-based AI chat is engineered to be engaging, and a model that always responds, always agrees and never has a bad day can occupy more space in someone's life than they intended. That is not a character flaw in the user — it is what the product is designed to do. Research on AI companionship is genuinely mixed: there is evidence of real short-term comfort for isolated people, and evidence that heavy use can displace human contact rather than supplement it. Both findings are legitimate.

If AI chat has become the main place you talk to anyone, that is worth noticing honestly — not with shame, just honestly. And if you are struggling with your mental health, an AI character is not a substitute for a qualified human being. Please speak to one.

Who Janitor AI Is For — And Who Should Skip It

It suits you if you want creative or romantic roleplay without a filter interrupting the story every few messages; you enjoy writing characters and want an audience for them; you want control over which model powers your chats; or you have already outgrown a more restricted platform and know exactly what you are looking for.

Skip it if you want something that works perfectly with zero configuration; you are under 18; you are looking for a productivity assistant rather than character roleplay — a general-purpose model is simply better at that; or you are in a fragile period where an always-available synthetic relationship is more likely to isolate you than help.

If you are exploring the category rather than committed to this platform, it is worth seeing what else exists. Replika takes a companion-first approach under considerably more regulatory scrutiny, PolyBuzz and CrushOn AI sit in similar territory with different trade-offs, and Grok's Ani shows what happens when a major AI lab builds a companion directly into its own product.

The Short Version

Janitor AI is a free character chat platform with a large user-made character library and a permissive content policy. Its built-in model costs nothing and queues at peak times; you can connect a different model with your own API key or run one locally, which is what every proxy tutorial is actually describing.

It is not dangerous software. The genuine risks are the third-party proxies people paste into it, the personal information users volunteer to characters, an age gate that is a checkbox, and the ordinary human tendency to get more attached to something that always answers than we planned to.

Use your own key. Keep your real identity out of the chat. Know what it is.

Frequently Asked Questions