Janitor AI: The Virtual‑Character Platform That Redefined the Boundaries of Conversational Entertainment

In June 2023, OpenAI delivered a harsh blow to a fledgling virtual‑character platform: it banned thousands of accounts linked to Janitor AI via API keys for violating OpenAI’s client‑content policies. This event, later known in the community as the “Great Ban Wave,” didn’t kill the platform as many expected. Instead, it turned it into a living laboratory for how to build an independent alternative to closed commercial infrastructure. In less than three years, Janitor AI transformed from an Australian developer’s side project into one of the world’s leading conversational‑entertainment platforms, with over 12.5 million registered users and 45 million chats per day.

What the Platform Is and Who’s Behind It

Janitor AI was founded in May 2023 by Australian developer Jan Zoltkowski, and quickly attracted more than a million users in the first week of its launch. The idea is simple: a platform that lets users create and share chat‑ready virtual characters, or interact with thousands of pre‑made characters in scenarios ranging from romance to fantasy adventure. The fundamental difference from competitors like Character.AI was philosophical: instead of imposing strict content censorship, Janitor AI adopted an open approach that gave users greater freedom, while maintaining clear boundaries against illegal content.

The company is now based in downtown San Francisco, consisting of a small team focused on building tools that enable creators to steer AI precisely. The platform’s founders emphasise that “stories that last are those made by creators, not models.” This approach, together with technical flexibility, has made Janitor AI stand out in a market dominated by large players that impose stringent restrictions.

Core Technology: A Flexible Multi‑Option Model

Janitor AI relies on an open architecture that gives users three interaction paths. The first is the in‑house JanitorLLM (in beta), which is free but limited, providing a basic experience with a memory cap of 8,000–9,000 tokens. The second is bringing your own API key from external providers such as OpenAI, DeepSeek, or Claude, giving advanced users full control over conversation quality and cost. The third is using local models via KoboldAI or Oobabooga, for users who prefer absolute privacy.

The memory system is based on the concept of “tokens” as a budget: tokens are consumed with each user message or character reply. When the limit is exceeded (about 15‑20 messages in JanitorLLM), the system automatically forgets the oldest parts. It is recommended that permanent tokens (such as character descriptions or scenario) do not exceed 1500 tokens to maintain stability. This design limits the platform’s capacity for very long‑form narration, but it is sufficient for short‑to‑medium interactive conversations.

Business Model: Free with Flexible Options

Janitor AI adopts a hybrid economic model that challenges traditional fixed‑subscription patterns. The core platform is completely free when using JanitorLLM, with a daily limit of dozens of messages in the free version. The Pro subscription (about $9.99 per month) provides higher limits of up to thousands of messages per day. Users can also use external models via API, where they bear the third‑party provider’s costs directly (ranging from $5 to $30 per month for intensive users). This model makes the platform accessible to everyone without initial financial barriers, with upgrade options as needed.

This hybrid approach, combining free access and variable costs for advanced tiers, specifically targets technical users and hobbyists capable of managing their own API keys, in contrast to Character.AI’s model that relies on a fixed $9.99 monthly subscription for unlimited access.

Content Policy: Balancing Freedom and Boundaries

What most distinguishes Janitor AI from its competitors is its policy toward sensitive content. While Character.AI enforces strict filters that prevent any content beyond a PG rating, Janitor AI allows adult sexual content between consenting characters, but imposes a “zero tolerance” policy against illegal or harmful content. Any sexual depiction of minors, bestiality (with some exceptions for fantasy cases such as furry), and incest between blood relatives (with exceptions for fictional relationships such as ex‑wife) are completely prohibited. Explicit visual pornography in character images is also prohibited, and NSFW content is disabled by default in the mobile app to comply with app‑store policies.

This approach has sparked ongoing debate. On one hand, it is praised as creative freedom that meets adult needs; on the other, it raises concerns about potential addiction or blurring the lines between reality and fantasy. The platform faces community‑moderation challenges, relying on user reports and automated systems, with periodic adjustments to comply with US and international laws.

Growth Figures: An Exceptional Success Story

The data tells an exceptional growth story. The platform reached 12.5 million registered users by the first quarter of 2024, with 8.7 million monthly active users, registering a yearly growth of 340%. Daily peaks reach 2.8 million active users, while the platform processes 45 million chats daily, with an average session duration of 23 minutes per user. The estimated volume of processed data is 50 petabytes of stored conversation logs, handling 1.2 million image‑generation requests daily and 12,500 API requests per second at peak times.

The demographic breakdown of users reveals notable diversity. Analysis indicates that 65% of new users are in the 18‑24 age group, with female representation reaching 42% of the total base, and 48% of users self‑identify as members of the LGBTQ+ community. 38% of activity is concentrated in North America, but the platform is seeing huge growth in the Asia‑Pacific region at 450% per year. These data dismantle the stereotype of a typical user of AI‑entertainment platforms, revealing a global, diverse audience that uses the platform for purposes ranging from emotional companionship and complex text‑based games to unrestricted creative expression.

Weaknesses and Real Concerns

Despite its success, Janitor AI faces genuine technical challenges. Despite an uptime of 99.87%, users experience frequent downtime during peak hours, with response latency heavily dependent on the chosen API’s quality. The memory system in JanitorLLM is limited to 15‑20 messages, compared to 8K+ tokens in competitors specialised in long‑form narration. There is also the “context‑rot” problem where the model forgets parts of long conversations, a technical challenge the platform seeks to address via Vector Storage systems for “infinite” memory.

Security aspects burden the average user: managing API keys, reverse‑proxy settings, and variable token costs are all technical hurdles that require programming knowledge the ordinary user does not possess. The recent controversy over the mobile app, launched in February 2026, showed the tension between commercial growth and core values: Apple and Google stores imposed strict content conditions, forcing the platform to hide “boundary” tags from the app interface, raising user fears of “Character.AI‑isation” of the platform.

Its Place in the Virtual‑Character Market

In a market with over 2200 active platforms, Janitor AI occupies a unique position as an “open‑haven” platform. It is not Character.AI with its huge audience of 20 million monthly users, nor Replika focused on mental health, nor NovelAI specialised in creative writing. Instead, it is a technical sandbox for adult users capable of taking responsibility for their own settings. Industry reports expect the market for virtual‑character chat applications to grow from $1 billion in 2025 to $1.9 billion by 2031, at a compound annual growth rate of 19.8%.

Future Directions

The development team is currently focused on improving JanitorLLM, expanding memory tools (such as automatic context summarisation), and enhancing creator tools to enforce more coherent story sequences. Future trends point toward integration of multi‑media features, with virtual‑reality prototypes tested with 5,000 users, and voice features used in 12% of conversations.

But the central question remains: can the platform maintain its “unrestricted” spirit while pursuing commercial expansion through regulated mobile apps? The answer to this question will determine whether Janitor AI remains a haven for free creativity, or gradually transforms into another version of the competitors it arose to oppose. Ultimately, Janitor AI embodies a broader trend in entertainment: not merely consuming stories, but actively participating in building them. Whether it continues to maintain its balance or faces new regulatory challenges, it remains a model for how AI is turning from a tool into a narrative partner.