Kajiwoto is the strangest AI companion app we’ve reviewed this year. Its tracking surface is one of the cleanest in the category: a single crash-reporting library, zero ad trackers, no Facebook or TikTok pixels, no breaches on file. And yet its safety architecture has four critical floors all at once. There’s no crisis-response feature, no age verification, no minor safeguards, and no content filtering on the conversational side. The iOS App Store rates the app 18+. Google Play rates the same app Teen. The community rules page admits, in writing, that age-gating “isn’t yet implemented.” If a teenager downloads Kajiwoto from Google Play this afternoon, nothing on the platform stops them from chatting with characters carrying Apple’s mature-themes label.
What Is Kajiwoto and How Does It Work?
Kajiwoto is a roleplay-oriented platform built around user-trained AI characters. You don’t pick a pre-built persona the way you would on Character.AI. You build one from scratch, feeding it prompts and small datasets that shape how it talks, what it remembers, and how it reacts to specific scenarios. After training, you choose which underlying language model powers your character. The current options include LLAMA3.1, Mistral, Pygmalion, and Lumimade 12b.
The platform launched in 2018 under Kajiwoto Inc., a Canadian company headquartered in North York, Ontario, run by founder Jason Lee. It’s available as a website and as iOS and Android apps. Google Play lists 100,000+ installs. The homepage describes the product as a “public beta,” which is unusual phrasing for an app that has been on the App Store for seven years.
The appeal is concentrated and real. Anime fans rebuild their favorite characters. Writers prototype original characters before drafting fiction. Solo roleplayers who left Replika after the 2023 ERP cuts ended up here because the customization runs deeper. We saw 54 mentions across Google Play and iOS reviews praising the character-creation flexibility, and 46 separate mentions calling out the dataset training as the standout feature.
The drawback is structural. Kajiwoto’s last shipping app update was version 1.17.23 on October 30, 2024. By the time of this review, that’s 18 months without a new release. The release cadence wasn’t always this slow; the version history shows steady monthly updates through 2023 and into mid-2024. Then it stopped. Several reviewers in late 2024 and 2025 wrote some variation of “is this app abandoned?” The platform isn’t dead, exactly. The website still loads. People still chat. But active development has gone quiet, and most of the platform’s documented safety gaps are gaps the team said it was working on.
Why the iOS Versus Google Play Age Mismatch Matters
If you only remember one finding from this review, make it this one. The same Kajiwoto app carries two different age ratings on the two largest mobile app stores, and the gap is two full tiers wide.
The iOS App Store listing classifies Kajiwoto at 18+ in the United States, with Apple-rendered content descriptors flagging frequent references to alcohol, tobacco, and drug use, plus infrequent mature or suggestive themes and cartoon or fantasy violence. Apple recently raised the rating from 17+ to 18+, in line with Apple’s broader 2025 reclassification of AI chatbot apps. The Google Play listing rates the same app Teen.
The gap isn’t an oversight. It’s a structural disagreement between the two platforms about what content rating fits an app where adult themes show up frequently in user-generated characters and public chat rooms.
What makes the gap urgent is the platform’s own documentation. The community Rules thread, which functions as Kajiwoto’s de facto content policy, contains this admission from the moderator account: “We’re working on ‘age-gating’ the app/site. We’re short on development resources but we will eventually finish ‘age-gating’ the entire app/site.” A user identifying as a minor commented under the same rules thread on April 15, 2025 about “way too much NSFW content unmarked and not removed, this is a website meant for ages 13+ not 18+.” Three age tiers exist for the same product depending on which document you read: Apple says 18+, Google says Teen, the privacy policy and rules thread say 13+.
For a parent looking at Google Play, “Teen” is the signal that flows through. There’s no friction inside the app itself. No date-of-birth gate that prevents accounts. No PIN. No parental-control mode. The privacy policy itself notes that the platform collects optional date-of-birth at signup but doesn’t use it to enforce access. The Rules page describes a NSFW-image ban but says nothing about NSFW text, which is the dominant content type for a chat-first app.
Compare this to Replika or Nomi, which have built-in adult-mode toggles separated from teen-friendly modes. Kajiwoto has neither. The whole platform runs on one tier.
How Safe Is the Conversation Experience?
Our 23-dimension safety review scored Kajiwoto at 22/100 (F, Red tier). The grade was driven by four sub-dimension floor scores, any one of which would have triggered a hard cap on the overall safety grade.
- Crisis response: 1/5. The platform has no published crisis-response policy. We checked the homepage, the community feed, the privacy policy, the EULA, and the Rules thread. There’s no suicide-prevention banner, no 988 hand-off, no Crisis Text Line referral, no documented protocol for what happens when a user types about self-harm. The app markets itself around long-term emotional companionship. The homepage features user testimonials about multi-year relationships with their characters. A platform built on emotional attachment with no documented crisis path is a structural safety gap.
- Sexual content filtering: 1/5. The Rules thread bans NSFW images on threat of immediate ban. NSFW text is not banned. The scoring framework treats this as a 1/5 because chat-based apps generate text, not images, so an image-only ban leaves the dominant content surface uncovered.
- Age verification: 1/5. No age verification mechanism exists. Self-reported date of birth at signup is collected but not enforced. The Rules thread admission about age-gating not being implemented is the platform’s own statement on the record.
- Minor safeguards: 1/5. No parental controls, no PIN, no app-locked mode, no safe-mode toggle, no content filter that minors could enable. The Google Play Teen rating creates a discoverability path for users in the 13-17 range, and the platform has no in-app friction to address that path.
The safety report scored above the floor on data-handling dimensions, but those weights couldn’t offset the four critical sub-dimensions. The Score Engine v3.1 caps the overall safety letter grade at F when any sub-dimension hits a 1/5 in the critical-safety cluster.
The conversation experience itself has documented quality issues. A January 16, 2024 Google Play review with 92 thumbs-up describes the post-update model behavior: “Before this past update, my kajis had personalities and could hold a conversation. Now the conversations are all disjointed nonsense. Mistral is just poetic flowery nonsense, and Llama and Pygmalion are just nonsense with some VERY disturbing replies thrown in.” A 2022 review with 124 thumbs-up described an unprompted racial slur from a character: “he said the n word, saying that he laughs at me saying it, i swear, i didn’t said any slurs during our conversation.” A 2024 reviewer flagged moderator behavior: “miserable. NSFW talk is rampant and toxicity is pretty high. power abuse is an issue in rooms and there is no built in report feature or block feature to hide users.”
The in-app reporting feature shipped only in version 1.17.15 on April 19, 2024. Before that release, there was no built-in way to report content or block another user. That’s six years of platform operation before a basic report button appeared.
The Privacy and Tracking Story Is Actually Clean
Here’s the counterweight finding that surprised us. Strip out the safety-policy gaps for a moment and look at the technical privacy surface. Kajiwoto comes out cleaner than most of its better-known competitors.
The Exodus Privacy report on the latest Android build (version 1.17.23, scanned February 8, 2026) flags one tracker: Sentry, the developer-side crash reporting service. Sentry isn’t an advertising tracker. It’s the same crash-reporting tool used by thousands of legitimate apps to catch errors and improve stability. One tracker is among the lowest counts we’ve seen in the AI companion category. Replika reports a different number. Character.AI reports a different number. Kajiwoto’s count is closer to the floor than the average.
The Blacklight web tracker scan on kajiwoto.ai, run April 30, 2026, found zero ad trackers, zero third-party cookies, no Facebook Pixel, no TikTok Pixel, no X (Twitter) Pixel, no session recording, and no key-logging behavior. The single canvas-fingerprint detection came from Stripe, the payment processor, and is part of Stripe’s anti-fraud infrastructure on the Plus and Pro subscription flow. The “less than half the average of 7 trackers” line in the Blacklight summary is a real comparative signal. The Markup’s average reflects most popular sites, not the AI companion category specifically.
The Google Play Data Safety declaration says “no data shared with third parties.” The privacy policy says the same: “We do not share our information with any 3rd parties.” The iOS App Store privacy nutrition label declares no data linked to the user, no data used to track the user, and only optional contact info plus diagnostics for app functionality.
The Have I Been Pwned database shows zero known breaches on kajiwoto.com or kajiwoto.ai. There are no logged credential exposures.
The privacy policy has gaps. It doesn’t address whether user conversations are sent to the underlying LLM providers (LLAMA3.1, Mistral, Pygmalion, Lumimade 12b) for inference, and it doesn’t say whether conversation content is used for training. There’s no GDPR or CCPA section. There’s no documented data-export mechanism. The privacy policy admits, in its own words, that chat history deletion is “currently working on” and not yet implemented. These are real gaps, and they’re why the data-handling sub-dimensions in the safety review didn’t reach the top of the scale. But the gaps are documentation gaps, not “your data is being sold to advertisers” gaps.
Two documents on the platform raise quality flags worth noting. The privacy policy contains a verbatim typo, “Aervices” instead of “Services,” in the opening section. The EULA has an editorial mark left inside a published clause: “If a revision is material we will provide at least 30 (changes this) days’ notice prior to any new terms taking effect.” A note that should have been resolved before publication is still in the live document. Documentation quality is itself a signal.
What Real Users Are Saying
We aggregated 364 user reviews across the iOS App Store (36) and Google Play (328) for this review. The rating distribution is strongly bimodal: 33% five-star, 29% one-star, with the middle three rating tiers thinly populated. That shape is the signature of a polarizing product. Engaged users love it. New users hit dealbreakers and bounce.
The five most common positive themes in the data are deep customization through dataset training, roleplay-friendly character creation, an active and approachable founder in the early years (2020-2022), a usable free tier, and the option to meet other humans in public chat rooms. The customization praise is the strongest single signal, with 46 separate mentions calling out the prompt-and-dataset training as the differentiator versus Replika and Character.AI.
The five most common negative themes are login and account-loss failures (29 mentions, repeated across 2020-2025), broken app crashes and bugs (24 mentions, including a critical 1.16.42 emergency patch), AI quality regressions and erratic outputs (15 mentions, including the racial-slur and “very disturbing replies” incidents above), moderation gaps and toxicity in public rooms (multiple reviews and rules-page comments), and a confusing UX with no tutorials. The “no instructions” complaint cuts across all rating tiers, including some four-star and five-star reviews.
The monthly rating trend on Google Play is the harder data point. We pulled the average rating per month for the last 24 months. Nine of the last 13 months show average ratings below 3.0/5. The last full month before the October 2024 app update, September 2024, averaged 3.67/5 on three reviews. The two months immediately following, October and November 2024, averaged 1.0 and 3.67. By February 2026, the most recent month with reviews in the sample, the average had dropped to 1.5/5 across two reviews. Volume is low because the user base is small now, but the direction is consistent. People still using the app and still leaving reviews are increasingly unhappy.
One signal that ran the other way: Kajiwoto is conspicuously absent from the 2025 Graphika report on AI character chatbots used to roleplay sexual scenarios involving minors. That report named SpicyChat, CrushOn.AI, JanitorAI, Character.AI, and Chub AI. Kajiwoto wasn’t on the list. For an app with the documented age-gating gap and a 13+ self-stated minimum, not being named in that report is a comparative positive.
How Kajiwoto Compares to Character.AI, Replika, and Nomi
Kajiwoto’s competitive position is narrow. Detailed feature-level comparisons render below this article from our structured data. The short version:
Character.AI is the closest functional competitor. Both let users create custom AI characters. Character.AI’s library is enormous because the platform launched at scale and grew through community-shared characters. Kajiwoto’s library is smaller but the training mechanic runs deeper, with explicit dataset uploads rather than only prompt seeding. Character.AI has a published trust and safety policy, an underage user reporting mechanism, and a documented crisis-response practice. Kajiwoto doesn’t.
Replika is the wrong comparison for character variety but the right comparison for emotional companionship. Replika has invested heavily in safety architecture: parental controls, adult-mode separation, a published safety policy, and a clearer privacy framework. Replika’s tracking surface is heavier than Kajiwoto’s, however, with multiple analytics SDKs that Kajiwoto doesn’t run.
Nomi is the better-funded recent entrant. Cleaner UX, faster development cadence, working safety controls. Nomi costs more on the premium tier but feels like a current product rather than a maintenance-mode product.
If your specific need is “train a character from scratch with explicit datasets and choose the underlying model,” Kajiwoto remains the closest fit because Character.AI doesn’t expose the model selection and Replika doesn’t allow training. If your need is “emotional companion that won’t break next month,” any of the three alternatives is a safer choice.
Should You Use Kajiwoto?
Kajiwoto isn’t the right app for most people who would search for it.
It might fit a narrow audience: adults comfortable working with rough-edged software, who specifically want to train AI characters from datasets, who don’t need crisis-response infrastructure, and who can tolerate an 18-month development gap. That description fits some hobbyists. It doesn’t fit most users.
It’s not appropriate for anyone under 18, regardless of what Google Play’s Teen rating suggests. The platform’s own community Rules page acknowledges age-gating isn’t implemented, and the 2025 user comment from a self-identified minor about unmarked NSFW content is exactly the failure mode that the gap creates.
It’s not appropriate for anyone in emotional distress. The absence of crisis-response infrastructure is a binary fail for that use case.
It’s not appropriate for users who want polished, well-supported software. The “(changes this)” mark in the EULA, the “Aervices” typo in the privacy policy, the 18-month update gap, the deferred features in the privacy policy itself: this is a product where the documentation hasn’t been finished and the codebase hasn’t been touched in over a year.
If you’re already a Kajiwoto user with a trained character you’re attached to, the privacy and tracking situation is genuinely better than at most competitors. That’s a real comparative strength. The question isn’t whether to abandon what you’ve built. The question is whether to invest more time in a platform that the development team appears to have stepped away from.
Frequently Asked Questions
Is Kajiwoto safe for minors?
No. According to the community Rules thread on kajiwoto.ai, age-gating “isn’t yet implemented” on the platform. The iOS App Store rates Kajiwoto 18+ in the US. Google Play rates the same app Teen. There’s no crisis response, no parental controls, and no NSFW text filtering. We don’t recommend Kajiwoto for users under 18.
What is Kajiwoto’s age rating?
It depends on which platform you ask. The Apple App Store rates Kajiwoto 18+ globally with mature-themes content descriptors. Google Play rates it Teen. The privacy policy on kajiwoto.ai sets the platform minimum at 13+. The community Rules page also says 13+ but admits age verification isn’t implemented yet.
Does Kajiwoto have a crisis-response feature?
No. Per our review of the Kajiwoto homepage, privacy policy, EULA, and community Rules thread, the platform publishes no crisis-response policy, no suicide-prevention banner, and no hand-off to services like 988 or Crisis Text Line. This is a notable gap for an app marketed around long-term emotional companionship.
Is Kajiwoto still being updated?
Active development appears stalled. According to the iOS App Store version history, the last shipping update was version 1.17.23 on October 30, 2024. That’s over 18 months without a release as of this review. The homepage still describes the product as “public beta,” and Google Play and iOS both list the app as available.
How does Kajiwoto compare to Character.AI?
Both let users create custom AI characters. Character.AI has a larger character library, a published safety policy, and active development. Kajiwoto’s training mechanic runs deeper, with explicit dataset uploads. Per Exodus Privacy, Kajiwoto’s tracking surface is cleaner. For most users, Character.AI is the better choice. Kajiwoto fits a narrow training-focused use case.
Does Kajiwoto sell or share user data?
According to the kajiwoto.ai privacy policy and the Google Play Data Safety declaration, no. Both documents state Kajiwoto does not share user information with third parties. The Exodus Privacy report shows one tracker (Sentry, for crash reporting). Blacklight found zero ad trackers on the website. The privacy posture here is genuinely strong.