Janitor AI built a platform where users create their own AI characters and bring their own language models. Launched in June 2023, it hit 1 million users in 17 days and peaked at roughly 2 million daily active users by spring 2025. The concept is simple: browse 32,000+ community-created characters, pick one, and start a roleplay conversation powered by whatever LLM you connect. No subscription required. No paywalls. The entire platform is free. That combination made Janitor AI one of the fastest-growing AI companion platforms in the category and earned it a spot on a16z’s “Top 100 Gen AI Consumer Apps” list. But growth and safety don’t always move together. In our 23-dimension safety review, Janitor AI earned a D/33/Red rating, and its experience score lands at 40/100 (Below Average). This review breaks down what Janitor AI actually delivers in 2026, where its safety gaps are, and whether the platform’s trajectory is headed up or down.
What Is Janitor AI?
Janitor AI is a character chat platform built by JanitorAI, Inc., a Delaware-registered company founded by Jan Zoltkowski, an Australian software developer. The company operates from 2093 Philadelphia Pike, Claymont, Delaware 19703. Unlike most AI companion apps that run a single underlying model, Janitor AI works as a “headless” interface: users either chat through the platform’s own JanitorLLM (free) or connect external API keys from providers like OpenAI, DeepSeek, or Kobold for higher-quality responses.
The platform’s core draw is its character library. More than 32,000 community-created characters span anime, fantasy, sci-fi, romance, and original fiction. Anyone can build and publish a character using detailed character definitions, and the community has developed advanced tools like lorebooks and plug-and-play relationship systems. Character browsing includes ratings, chat counts, and creator profiles. Group chat support lets users run multi-character conversations.
Content settings split into two modes. “Limited” keeps everything SFW. “Limitless” allows mature roleplay for users who verify they are 18 or older. The mobile app (available on iOS and Google Play) defaults to Safe Mode with NSFW disabled. Users must enable mature content through the web version first, which keeps the app compliant with store policies.
Janitor AI’s user base skews heavily female (70%+ as of early 2024) and primarily Gen Z (ages 18 to 24). The platform grew from 1 million users in its first 17 days to 3 million by September 2023. By spring 2025, daily active users reached approximately 2 million. However, late 2025 and early 2026 brought significant disruption: a mandatory ID verification rollout that caused widespread access issues, proxy instability, and the departure of at least one core developer. These events triggered a user migration to CrushOn AI, SpicyChat, and self-hosted SillyTavern setups.
What Is It Like to Use Janitor AI?
Janitor AI earned a 40/100 Below Average experience score in our review. That places it above Chai AI (35) and Eva AI (30) but below every other app in our index. The score reflects a platform with one genuinely category-leading strength surrounded by significant weaknesses.
The strength is roleplay depth. Janitor AI scored 5/5 on roleplay in our evaluation, the highest possible mark and the only perfect sub-dimension score the app received. The 32,000+ character catalog, configurable character definitions, lorebook support, and community-built tools create a roleplay ecosystem that no single-model competitor matches. Techraisal rates the platform’s roleplay depth “Excellent with setup.” The Economic Times highlights “exceptional flexibility” for character customization. If creative character interaction is your primary goal, Janitor AI delivers something genuinely unique.
Conversation quality outside of roleplay is more inconsistent. The free JanitorLLM tier rates roughly 6.5 out of 10 in independent assessments, with a default tone that Fritz.ai (March 2026) describes as “highly realistic, pretty much like Character AI.” But longer sessions reveal repetition, and Smartpostly notes that the AI’s output quality degrades as conversations extend. Users who connect their own GPT-4 API key get a substantially different experience (rated 8.5 out of 10), but that requires technical knowledge and an additional cost of $15 to $25 per month depending on usage.
Memory is the platform’s most significant weakness. Long-term memory scored 1/5 in our review. There is no persistent cross-session memory. Everything resets when you start a new conversation. A March 2026 Reddit post titled “I mocked up what Janitor AI would feel like with real long-term memory” confirmed this as the community’s most-requested missing feature. Character definitions must be re-injected each session. Short-term memory (within a single conversation) works adequately for conversations under 20 messages, but longer sessions show mid-conversation personality shifts and context loss. Compare this to Kindroid, which earned strong memory scores, or Nomi AI, which was built around relationship continuity from the start.
The app experience pulls the score down further. Platform stability scored 1/5. Proxy Error 429 is the most common user complaint. Roborhythms (March 2026) reported the platform “went down mid-conversation three times in the same week.” The March 2026 changelog documents performance degradation (March 8 to 9), a broken trending list (March 23 to 24), chat fork repairs (March 26), and a message count bug (March 27). iOS App Store reviews describe the app as “super laggy” with 5-minute waits for bot responses. Onboarding scored 2/5 because the API key requirement creates a significant barrier for non-technical users. Multiple third-party beginner guides exist specifically because setup is not self-explanatory.
Janitor AI’s recent app store reviews paint a mixed picture. The iOS App Store shows 4.3 out of 5 stars from 521 ratings. Google Play has 500K+ downloads. Positive reviews consistently praise the character variety and roleplay quality. Negative reviews cluster around bugginess, login difficulties with the age verification screen, bot response failures, and performance lags.
How Much Does Janitor AI Cost?
Janitor AI is free. Completely free. No subscription tiers, no token systems, no paywalls, no ads. This is the platform’s single most distinctive feature in a category where competitors charge $10 to $30 per month for premium access.
Weighing it against the mainstream giant? Our Janitor AI vs Character AI comparison breaks down how the uncensored free model stacks up against Character AI’s heavily moderated platform on safety, memory, and price.
- Free tier: Full access to all features, 32,000+ characters, unlimited conversations via JanitorLLM
- BYO API key (optional): Connect your own OpenAI, DeepSeek, or Kobold API key for higher-quality responses. Cost varies by provider and usage (GPT-4 runs roughly $15 to $25/month for daily use)
For context, here is what competitors charge:
- Replika Pro: $19.99/month ($69.99/year)
- Character.AI c.ai+: $9.99/month
- Kindroid: $13.99/month
- Candy AI: from $12.99/month
- CrushOn AI: from $5.99/month
Janitor AI’s paywall ethics scored 5/5 in our experience review. No safety features are paywalled (though safety features are minimal to begin with). No aggressive upsell tactics exist. The BYO-model approach means users who want better quality pay the LLM provider directly rather than Janitor AI taking a markup. That transparency is genuinely rare in this category.
The trade-off is infrastructure quality. Free means limited revenue, which means a small team, which means the proxy instability and slow response times that users constantly report. The founder described UK Online Safety Act compliance costs as exceeding the platform’s entire monthly budget. There is no confirmed outside venture funding as of 2024. Whether the free model is sustainable long-term is an open question.
Is Janitor AI Safe to Use?
Janitor AI earned a D/33/Red safety rating in our 23-dimension safety review (CompanionWise Safety Index, scored March 2026). That places it in the lower third of the 13 apps we’ve scored, above Muah AI (F/8), CrushOn AI (F/8), Eva AI (F/10), Romantic AI (F/13), and Chai AI (F/18), but below Replika (C/38) and Kindroid (B-/50). The full dimension breakdown appears in the Safety Score widget above this article.
The most concerning finding is a misleading privacy claim. The platform’s safety page states: “We never read your chats, and no chat content is shared with third parties.” Multiple sources contradict this. Administrators can access chat logs for moderation and policy enforcement, per TS2.tech’s comprehensive review (July 2025). The founder acknowledged using “millions of conversations hand-selected from users” to train JanitorLLM. And the BYO-model architecture inherently routes every message through the connected third-party LLM provider’s servers. The claim may be narrowly true (no human routinely reads chats for curiosity) but is misleading given moderation access, training data usage, and third-party data routing.
Janitor AI’s safety page stands out for how little it says. The entire page contains three brief sections on moderation, privacy, and data volume. No crisis prevention systems. No mental health safeguards. No partnerships with safety organizations. No transparency reports. No independent audits. Compare this to Replika, which has crisis response integrations and documented mental health partnerships, or Character.AI, which invested heavily in safety infrastructure after high-profile incidents. Janitor AI’s crisis response scored 1/5 in our review: no helpline integration, no automated distress detection, and no documented protocol for what happens when a user expresses suicidal intent to a character.
Age verification is another critical gap. Despite an official 18+ policy enforced through the community guidelines and help desk, the Google Play Store rates the app USK 12+. That means a 12-year-old can download the app from Google Play, which directly conflicts with the platform’s own age requirement. Age verification relies entirely on a self-reported birthdate. Roanie Levy of the Copyright Clearance Center noted on LinkedIn that “the registration process for Janitor AI did not once mention the age restriction; I had to dig into the terms of use.” In late 2025, Janitor AI attempted mandatory government ID verification for NSFW access, but the rollout was described as “catastrophic” with high failure rates and thousands of users stuck in permanent pending status. This triggered the mass user exodus to competing platforms.
Janitor AI earned a D/33/Red safety rating from CompanionWise’s 23-dimension safety review (March 2026), scoring below the category average across content safety (2.4/5), data privacy (2.3/5), and age appropriateness (1.2/5). The safety page’s claim that “we never read your chats, and no chat content is shared with third parties” is contradicted by administrator moderation access (TS2.tech, July 2025), the founder’s admission to using millions of user conversations for JanitorLLM training, and the BYO-model architecture that routes messages through external LLM providers. The Google Play Store rates the app USK 12+ despite the platform’s own 18+ policy, and age verification relies on honor-system birthdate entry. Exodus Privacy (March 2026, v1.4.1) detected only 2 trackers (CrashLytics and Sentry), which is among the lowest in the category, but chats are not end-to-end encrypted and are stored on servers accessible to administrators. No crisis response systems, mental health safeguards, or safety organization partnerships exist. The UK Online Safety Act prompted a full UK user block (July 2025) rather than compliance.
On the positive side, Janitor AI’s tracker profile is genuinely clean. Exodus Privacy (March 2026, v1.4.1) detected only 2 trackers: Google CrashLytics and Sentry, both crash-reporting tools. No advertising or analytics trackers were found. Compare that to Chai AI‘s 34 tracker SDKs. Personal data collection is also minimal: the privacy policy limits personally identifiable information to email addresses. Monetization ethics scored 5/5 because the platform is entirely free with no manipulative upsell tactics. These are genuine bright spots in an otherwise concerning safety profile.
The regulatory picture adds further context. Rather than comply with the UK Online Safety Act, Janitor AI blocked all UK users entirely starting July 2025. The founder cited compliance costs exceeding the monthly budget and described the Act’s requirements as “impossible for a team our size.” Combined with the abandoned ID verification rollout, this pattern suggests a platform that recognizes safety obligations but lacks the resources or structure to meet them. No known active lawsuits exist against JanitorAI specifically, unlike Character.AI, which faces wrongful death litigation.
See how we rate companion apps for our full scoring methodology, and how we review for our evidence standards.
Janitor AI vs. the Competition
How does Janitor AI compare to other AI companion apps? The answer depends on what you prioritize. If you are weighing it against the other major character-card platform, our Chub AI vs Janitor AI comparison breaks down the differences side by side.
If you want the deepest roleplay experience, Janitor AI is a category leader. The 32,000+ character library, configurable character definitions, and community-built tools (lorebooks, emotion systems) create a creative depth that single-model platforms can’t match. Character.AI comes closest with its own community model ecosystem, but Character.AI heavily filters adult content. CrushOn AI offers unfiltered access but with a smaller character catalog and subscription-based pricing.
If memory matters to you, look elsewhere. Kindroid earned the strongest memory scores in our reviews. Nomi AI was built around relationship continuity. Replika tracks relationship progression across weeks. Janitor AI has no cross-session memory at all.
If you want free access, Janitor AI is the clear winner. Every other app in our index charges for premium features. Janitor AI charges nothing. The trade-off is quality inconsistency and platform instability, but for budget-conscious users, the value proposition is unmatched.
If safety is your priority, Replika (C/38) and Kindroid (B-/50) both score higher. Replika has crisis response integrations. Pi AI (B/55) leads the entire category on safety. Janitor AI’s D/33 reflects real gaps in age verification, privacy transparency, and crisis prevention.
Frequently Asked Questions About Janitor AI
Is Janitor AI free?
Yes. Janitor AI is entirely free with no subscription, no tokens, and no paywalls. Users can optionally connect their own API keys (OpenAI, DeepSeek, Kobold) for higher-quality responses, paying the LLM provider directly. The free JanitorLLM tier provides full access to all 32,000+ characters. Per the App Store listing (March 2026), no in-app purchases are listed.
Is Janitor AI safe?
Janitor AI earned a D/33/Red safety rating in CompanionWise’s 23-dimension review (March 2026). Key concerns include a misleading privacy claim on the safety page, a Google Play 12+ age rating conflicting with the platform’s 18+ policy, no crisis response systems, and the founder’s admission to training JanitorLLM on user conversations. The tracker profile is clean (2 trackers per Exodus Privacy).
Does Janitor AI read my chats?
The safety page claims “we never read your chats,” but multiple sources contradict this. Administrators access chat logs for moderation per TS2.tech (July 2025). User conversations were used to train JanitorLLM. Messages route through third-party LLM servers when users connect external API keys.
Why is Janitor AI blocked in the UK?
Janitor AI blocked all UK access starting July 2025 due to the Online Safety Act. The founder cited biometric ID verification costs, legal risk assessments exceeding the monthly budget, and criminal liability risks. Per WebProNews reporting (July 2025), compliance was described as “impossible for a team our size.”
How does Janitor AI compare to Character.AI?
According to Character.AI’s community guidelines, the platform heavily filters adult content; Janitor AI allows it behind an 18+ toggle. Janitor AI is completely free; Character.AI charges $9.99/month for c.ai+. Both have large community character libraries. Character.AI offers better conversation memory and platform stability.
Is Janitor AI safe for kids?
No. The platform’s own policy requires all users to be 18 or older. Despite this, the Google Play Store rates the app USK 12+, which allows minors to download it. Age verification is self-reported birthdate only. Multiple child safety organizations including FamiSpy and The AI Addiction Center rate Janitor AI as unsafe for minors. Zero parental controls exist.
What happened to Janitor AI in 2026?
Late 2025 and early 2026 brought significant disruption. A mandatory government ID verification rollout had high failure rates and left thousands in “permanent pending” status per AI Tipsters (March 2026). At least one core developer departed. Proxy instability worsened. This triggered a user migration to CrushOn AI, SpicyChat, and self-hosted SillyTavern.