Janitor AI
Last reviewed
Jun 4, 2026
Sources
13 citations
Review status
Source-backed
Revision
v1 · 1,923 words
Improve this article
Add missing citations, update stale details, or suggest a clearer explanation.
Suggest edit
CtrlK
Last reviewed
Jun 4, 2026
Sources
13 citations
Review status
Source-backed
Revision
v1 · 1,923 words
Add missing citations, update stale details, or suggest a clearer explanation.
Janitor AI is a character-based AI chatbot and roleplay platform hosted at janitorai.com, where users hold open-ended text conversations with user-created fictional characters. It launched in mid-2023 and grew extremely fast, becoming one of the most-visited generative AI websites in the world by web traffic. The site is best known for permitting mature and uncensored roleplay that mainstream assistants block, and for a hybrid technical design in which conversations can be powered either by the platform's own in-house model (JanitorLLM) or by external large language model backends that a user connects through an API key or a community-run reverse proxy. It is frequently compared to Character.AI as the two leading character-chat services, with the two distinguished mainly by their opposite stances on content moderation.
Janitor AI was created by Jan Zoltkowski, an Australian-born software developer who had previously worked in the cryptocurrency space and built the project at a hackathon. In a September 2023 profile, Semafor described the site as launching on June 1, 2023, and reaching one million users within 17 days. By the time of that profile, roughly three months after launch, Semafor reported that the platform had about three million users who had exchanged on the order of 2.5 billion chat messages.
The site filled a gap left by mainstream chatbots: at the time, ChatGPT and Character.AI applied strict filters that blocked sexual and other mature content, and Janitor AI attracted users who wanted fewer restrictions on fictional roleplay. Early coverage emphasized its use for romantic and erotic roleplay, including AI "boyfriend" and "girlfriend" characters, and Semafor headlined its piece "The NSFW chatbot app hooking Gen Z on AI boyfriends."
Some later promotional write-ups state that women make up more than 70 percent of the user base, but that figure traces to PR-style sources rather than the platform or an independent audit, and it conflicts with web-analytics estimates (see Audience below). It should be treated with caution.
At launch, Janitor AI's conversations were powered by OpenAI's models through the ChatGPT API. In July 2023, OpenAI sent a cease-and-desist letter and cut off the platform's API access over content-policy violations, according to Semafor. Rather than shut down, Zoltkowski began building a custom language model for the platform, based on an open-source architecture and trained in part on the large volume of conversations Janitor AI users had already generated. This effort became JanitorLLM, the platform's own model.
In its early months the company published little information about itself. Semafor reported that Zoltkowski initially pitched venture investors with help from Martin Shkreli, the convicted former pharmaceutical executive, but that Shkreli's involvement turned some investors off and that he was no longer involved by September 2023, when Zoltkowski said he expected to close a funding round soon.
Startup-data aggregators later listed a single funding round for the company. According to profiles on PitchBook and Crunchbase, JanitorAI raised approximately $3 million and counts Khosla Ventures, North Equity, Premise Ventures, and Sky9 Capital among its investors, with the company headquartered in San Francisco. These details come from third-party databases rather than a company announcement and the round's exact date is not clearly documented.
Janitor AI presents a gallery of community-created character "cards." Each card is a character profile that bundles a persona, a description, an avatar image, an opening scene, example dialogue, and hidden instructions that steer how the character behaves. Anyone can browse the gallery and start chatting, and registered users can publish their own characters. By 2026 the catalog had grown to tens of thousands of public characters spanning original creations, fan-made versions of fictional figures, scenario-driven adventures, and companion-style personas.
The conversation experience is open-ended text roleplay: the user types messages, the character replies in persona, and the exchange continues with the model maintaining the scene and character voice across many turns. The interface exposes detailed controls, including generation settings, custom personas for the user's own side of the chat, and the ability to edit or regenerate responses.
Janitor AI separates the front end (the website, character gallery, and chat interface) from the language model that actually generates replies. Users choose which backend powers a chat:
| Backend option | Description |
|---|---|
| JanitorLLM (JLLM) | The platform's own free in-house model, served from a shared queue. It is offered in beta and can be slow or throttled at peak times. |
| OpenAI-compatible API | The user pastes in their own API key (for example from OpenAI) so that an external commercial model generates the responses, billed to that user's own account. |
| Reverse proxy | A community-run intermediary server, sometimes called a "proxy," that relays requests to a model. Users enter a proxy URL plus a key. Proxies have historically been shared within the community and can be unstable. |
| Third-party model routers | Services such as OpenRouter can be connected as a proxy, giving access to a range of hosted models, including free options. Official help docs walk users through this setup. |
This design means the platform itself does not have to run an expensive model for every conversation: heavy or power users can bring their own compute by supplying an API key or proxy, while casual users rely on the free, rate-limited JanitorLLM. The reverse-proxy ecosystem in particular grew up as a workaround that let the community route conversations through commercial models, and it has been a recurring feature of how the service is used.
Janitor AI is built around permissive, user-directed content rather than a strict global filter. Adults are allowed to create and engage in mature and explicit roleplay, with responsibility shifted to the user, who is expected to be 18 or older. The platform's guidelines nonetheless prohibit categories such as sexual content involving minors, hate speech, and other material the site deems over the line, and characters or content that cross those lines are subject to removal.
This stance is the platform's defining contrast with Character.AI, which enforces strict global moderation that keeps interactions within safe-for-work boundaries. Where Character.AI is positioned as a polished, broadly accessible app with heavy emphasis on safety, Janitor AI is positioned around freedom and configurability, at the cost of a steeper setup process for users who connect their own models.
Janitor AI became one of the highest-traffic AI websites within months of launch. Independent figures should be read against a specific source and date, because monthly visit counts vary considerably between providers and over time.
| Source and period | Reported web traffic |
|---|---|
| WriterBuddy AI industry analysis, Sept 2022 to Aug 2023 | About 192.4 million total visits, averaging roughly 48.1 million per month, ranking it the 9th most-visited AI tool, with an unusually long average visit of about 27 minutes. |
| Similarweb, March 2026 (janitorai.com) | Roughly 149 million monthly visits. |
| Similarweb, April 2026 (janitorai.com) | Roughly 99 million to 141 million monthly visits, depending on the snapshot, with an average visit of about 17 to 18 minutes and a global rank in the low hundreds. |
The platform reported on the order of two million daily active users by 2025 to 2026. Web-analytics estimates from Similarweb in April 2026 put the largest single national share of traffic in the United States (roughly a third), followed by smaller shares from India, the Philippines, Russia, and the United Kingdom, and estimated a predominantly male audience skewed toward the 18 to 24 age group. That male-majority estimate is one reason the widely repeated "70 percent women" claim should be treated skeptically.
The very long average session times reported across multiple periods, often two to three times the typical web average, reflect the platform's roleplay use case, in which a single conversation can span a long, continuous session.
The high-traffic site is the janitorai.com domain. A separate domain, janitor.ai, exists but draws only a tiny fraction of the traffic and is not the main service.
Janitor AI and Character.AI are routinely described as the two leading character-chat platforms, and both let users create custom personalities and roleplay with them. Character.AI is the larger and more mainstream of the two, with its own reported monthly visits in a comparable range and roughly twenty million monthly active users, and it is built as a safety-first, broadly accessible product. Janitor AI competes by allowing the uncensored adult roleplay that Character.AI forbids, trading polish and ease of setup for openness.
In February 2026, Janitor AI released native mobile apps on the Apple App Store and Google Play, which required tighter content controls and the hiding of certain extreme tags from the mobile interface to satisfy app-store rules.
In 2026 the platform also began rolling out age verification in specific jurisdictions to comply with new online-safety laws, namely Brazil's Digital ECA (Lei No. 15.211/2025) and Australia's age-restricted material codes under the Online Safety Act 2021. According to the company's help documentation, verification was scheduled to roll out on April 24, 2026, using the third-party provider k-ID (also used by Discord, Twitch, and Snapchat). Users in those regions can verify through on-device facial age estimation, an ID scan, or a reusable "AgeKey," and the company states that it receives only a yes-or-no result on whether a user is over 18, with no documents or facial data retained. The company indicated it did not plan to extend mandatory verification to other regions at that time. These measures restrict full uncensored access for unverified or underage users in the affected countries while leaving existing characters, chats, and settings intact.
The platform's permissive stance has drawn the kinds of scrutiny common to uncensored AI companion services. Reporting and commentary have raised questions about the effects of emotionally intensive AI companionship on younger users, about the handling and privacy of intimate chat data, and about the reliability and safety of community-run reverse proxies, which route private conversations through third-party servers. The 2026 introduction of legally mandated age verification in Brazil and Australia, and the content restrictions tied to the mobile app launch, reflect the broader regulatory pressure on adult-oriented AI chat platforms. As of mid-2026 the service remained active and heavily trafficked.