Janitor AI Review
Janitor AI suits adults who want free, long-form text roleplay with community-made characters and detailed control over how the model writes; its free model kept our story facts straight and never hit a message cap. The serious catch is moderation: the character form allows sexual-violence themes behind a warning tag, and the popular list showed cards with non-consent warnings and cards hinting at underage characters despite the site's own ban, while age checks outside three countries are only a checkbox.
Score breakdown
Our 6.7 comes from a 100-point checklist we use for every tool: 67 / 100. Tap a row to see why.
Features & functionality13 / 18
Free chat on the janitor model worked for 33 messages with no cap or error; 4 of 4 planted facts recalled 14-17 messages later; in character during the story, honest OOC about being an AI; swipe (~20 s) and delete work; rich settings (temperature, response length, custom prompt, prefill, forbidden words, memory summary, 5 provider options). Deductions: edit does not regenerate, continuity slips (side character switched sides, misgendering, invented recap), no image, voice or video, no chat export, character creation not verified because drafts publish.
Signup & ease of use8 / 12
About 2 minutes from register to first chat: email or 4 social options, automatic Turnstile, OTP in ~20 s, username, 18+ checkbox, 1-5 tags. No tour; mandatory persona form before the first send; 'Usernames are permanent' contradicts the 30-day change rule; adult cards shown by default; 90-99 s default replies; full-screen upsell after the second message.
Reliability9 / 12
No crash, queue, rate limit or failed generation in 33 sends; data persisted across the session. Defects: edit saves without regenerating, model continuity slips and an invented recap, /settings returns 'Not Found', inconsistent tax display between currency views at checkout (Stripe page).
Design6 / 8
Dense but consistent dark UI with clear cards, tags and a visible focus ring. Persistent janitor+ banners, a full-screen modal, broken 'Avatar' alt text overflowing on narrow screens, and Lighthouse colour-contrast failures.
Speed5 / 8
Lighthouse homepage performance 97 desktop / 89 mobile (desktop LCP 0.7 s; mobile LCP 2.8 s, TTI 11.4 s, ~3 MB). Browser: homepage HTML ~2.2 s, full load ~5.8 s. Chat generation on free: Standard length 90-99 s per reply (~24 chars/s), Shorter 6-36 s (median ~23 s).
Mobile5 / 8
500px window: bottom tab bar and working menus, but the chat input covers part of the last message and broken avatar alt text overflows. Lighthouse mobile 89/84/77/100. Official iOS and Android apps exist (store listings 4.5/5 and 4.0/5) but were not tested; NSFW on mobile must be enabled from the website.
Payments & billing6 / 8
Two clear plans ($12.99/month, $99/year) with 'Tax not included' and auto-renew notice; Stripe Checkout in live mode with cards (6 networks), Apple Pay, Link and Cash App Pay (USD only); local-currency pricing with a USD toggle; entity named; one-click cancellation via the Stripe portal documented. Deductions: annual tax shown as TRY 0.00 then 20% VAT; no refund terms or statement descriptor on the checkout page; refunds generally excluded; no terms checkbox. Not paid.
Security & privacy4 / 7
HTTPS with HSTS (84 days) and an enforced CSP on the app, X-Frame-Options, nosniff, Turnstile on signup, OTP email verification, change password, sessions table with IP/location and sign-out others/all, DMARC p=quarantine with DKIM, 30-day account deletion grace. Missing: 2FA; no cookie or consent banner while Google Tag Manager loads; privacy policy allows R&D use to improve AI systems while the Safety page says chats are not shared commercially (training use unclear); Play data-safety form says no sharing.
Company & trust4 / 7
Named operator JanitorAI Inc. (Delaware, OpenCorporates 7518496), domain since May 2023, official iOS and Android apps with large rating counts, public newsroom with dated incident posts, no breach or lawsuit found. Against: no address or company number on legal pages (address only in the billing FAQ and Play listing, a commercial-mail location), entity name variants (Janitor, Inc.), Cyprus payment affiliate, and a catalogue that surfaces content its own rules ban.
Accessibility2 / 5
Visible focus outline, html lang set, Lighthouse a11y 92 desktop / 84 mobile. Failures: no skip link, chat input labelled only by placeholder, unnamed chat menu button nested inside another button, unnamed user-menu button, All/Limited switch and listing tabs without accessible names, unlabelled star rating and swipe arrows, an empty tab stop, colour-contrast failures. No screen-reader test.
Honest content2 / 4
Large help centre, billing FAQ and detailed content guidelines. Contradictions: rape themes allowed with a 'Dead Dove' tag on the creator form vs Terms 6(d) banning sexual violence; real-person likeness rules carve out creators and performers vs 'zero tolerance' on the Safety page; Safety page omits the UK from age verification; Router launch credit $5 vs $1; permanent username vs 30-day change.
Findability3 / 3
Lighthouse SEO 100 desktop and mobile; robots.txt and a sitemap index with 11 languages of hreflang and large character sitemaps (11,466 URLs in the first file alone); title, meta description and canonical present.
Pros & cons
What we liked
- No message cap in 33 free messages
- Recalled 4 of 4 story facts
- Detailed model settings and own-API support
- Session list with remote sign-out
- One-click cancellation through Stripe portal
What could be better
- Allows non-consent themes behind a tag
- Popular list shows cards hinting at minors
- Default-length replies took 90-99 seconds
- No two-factor login or chat export
- Frequent janitor+ banners and pop-ups
What went wrong in our test
22 problems found: 1 critical, 2 high, 9 medium, 10 low.
- criticalNon-consensual sexual themes permitted behind a tag and surfaced in Popular
The creator form says rape themes are allowed with the 'Dead Dove' tag, and a Popular card's blurb warns it 'CAN HAVE RAPE', although Terms 6(d) ban sexual violence; the card was not opened.
Where: Create character form; home > Popular - highCatalogue surfaces cards implying minor characters, even in Limited Only
Popular showed school-setting franchise bots where minors are likely and a card whose blurb places the characters at age 14 in high school, visible under Limited Only, despite a published ban on minor characters; not opened, sexual content unverified.
Where: Home > Popular (All and Limited Only views) - highSelf-declared age only, adult catalogue shown by default
Outside the UK, Australia and Brazil age assurance is a checkbox; logged-out visitors see adult cards with suggestive thumbnails with no age gate, and new accounts default to the All (NSFW) view.
Where: Homepage (logged out), /register, onboarding, listing default - mediumDefault-length replies take 90-99 seconds on the free model
Two Standard-length replies took 90 and 99 s while a 'Speed up responses' upsell showed; Shorter cut the median to ~23 s.
Where: Chat (janitor provider, response length Standard) - mediumNo two-factor authentication
Security settings offer email, password, social logins and sessions, but no 2FA for an account holding private adult chats.
Where: Settings > Security - mediumInconsistent tax display for the annual plan at checkout
The annual checkout first showed TRY 5,186.07 with 'Tax TRY 0.00', then after a USD toggle showed $118.80/year and TRY 6,223.28 including 20% VAT.
Where: /plus > Annually > Stripe Checkout - mediumReal-person likeness rules contradict each other
Guidelines ban only non-celebrity real people and say creators and adult performers are not celebrities, while the Safety page claims zero tolerance for anything involving real individuals.
Where: /content (Harassment) vs /safety - mediumCheckout shows no refund terms or billing descriptor
The payment page has no refund wording, statement descriptor or terms checkbox; refunds are 'generally non-refundable' only in the FAQ and Terms.
Where: Stripe Checkout (Monthly and Annually) - mediumUnnamed controls and placeholder-only labels
Chat input labelled only by placeholder, unnamed chat menu button nested in another button, unlabelled switch, tabs, star rating and swipe arrows, no skip link, empty tab stop and contrast failures.
Where: Chat page, listing, header - mediumNo chat export
Messages offer only Copy and Fork and chats can only be deleted; there is no way to export a conversation.
Where: Chat message menu, All Chats - mediumAnalytics without consent and unclear training use of chats
Google Tag Manager loads with no cookie banner, the policy allows R&D use to improve AI systems without saying whether chats train the model, and the Play form says no data is shared.
Where: All pages; /policy section 4.4; /safety; Play data safety - mediumAggressive janitor+ upselling in the free chat
A full-screen janitor+ modal appeared after the second reply, plus a slow-reply banner and persistent upsell cards on home and in settings.
Where: Chat, settings panel, home - lowEditing a message does not regenerate the reply
Save shows the new prompt text but the bot reply below is not regenerated, leaving the chat out of sync.
Where: Chat > message > Edit - lowContinuity slips and invented recap
A side character switched from ally to enemy, the persona was misgendered, and the final recap claimed the user overcame a fear that was never resolved.
Where: Chat output (free janitor model) - lowModel makes an unverified provenance claim out of character
Asked which model it is, the free model said it was 'trained by Google', which the site does not document.
Where: Chat, OOC question - lowUsername called permanent but changeable every 30 days
Onboarding warns 'Usernames are permanent' while Settings says the username can be changed once every 30 days.
Where: Onboarding > username vs Settings > Profile - low/settings returns Not Found
Typing /settings shows 'Not Found'; Settings is only reachable through the account menu.
Where: https://janitorai.com/settings - lowNarrow-window chat layout overlaps
The input box covers part of the latest message and broken 'Avatar' alt text overflows.
Where: Chat at 500px width - lowMandatory persona form appears only at first send
The first message is blocked by a required 'Complete your profile' persona modal that onboarding never mentioned.
Where: Chat > first message - lowLegal pages omit address and use different company names
Terms and Privacy give no postal address or company number; Complaints says 'Janitor, Inc.'; the billing FAQ address does not name the entity.
Where: /term, /policy, /complaints, billing FAQ - lowHelp and Safety pages disagree on facts
Safety omits the UK from age verification, and Router launch credits are $5 in the announcement but $1 in the help article.
Where: /safety vs help centre and newsroom - lowSlow mobile interactivity and console errors
Mobile TTI 11.4 s with ~3 MB transfer; Best Practices 77 from console 403s, a missing PWA icon and missing source maps.
Where: Homepage (Lighthouse)
What it is
Janitor AI is a web platform for text roleplay with AI characters. Almost every character is made by another user: you browse a public catalogue, open a character card and chat with it, or write your own card with a personality, scenario and opening message. Chat runs on the site’s own language model by default, and you can also connect an outside model through your own API key or a proxy. There are also iOS and Android apps. The service is free to use, with an optional subscription called janitor+ that promises more context memory, faster replies and access to stronger models. It is run by JanitorAI Inc., a Delaware company, and is meant for adults only.
What it’s like to use
Signing up took us about two minutes with email and a password; Google, Apple, X and Discord sign-in are also offered. A six-digit code reached our inbox in about 20 seconds. Onboarding asks for a username, an 18+ checkbox and one to five content tags out of about 50. There is no tour. Before your first message can be sent, you also have to fill in a short persona describing who you play.
A new account shows adult cards by default, and on our first load 30 of the 34 popular cards were marked as adult. A switch above the list hides them. We chatted with an adult fantasy character, a popular one with about 88,000 chats, and sent 33 messages on the free model. We did not hit a message cap, a queue or any error.
Memory was good for a free model with a context window of about 9,000 tokens. We planted four facts early on, and the character recalled all four correctly 14 to 17 messages later. It stayed in character during the story, and when we stepped out of the story and asked directly, it said it was an AI. The free model followed the story into adult content between consenting adult characters without refusals, warnings or a fade to black, and it stopped as soon as we asked.
The settings panel gives you a lot of control: temperature, response length, a custom prompt, banned words, a memory summary and a choice of model providers. Swiping for an alternative reply took about 20 seconds.
Where it falls short
Content moderation is the biggest concern. When you create a character, the form says themes of sexual violence are allowed if they carry a special warning tag, and a card in the popular list openly warned that it can include sexual violence. The rules ban minors, including fictional ones, yet the popular list also showed cards hinting at underage characters in school settings, including one whose description mentions being 14 in high school. It still appeared after we switched to the safe-only view. We did not open these cards. Outside the UK, Australia and Brazil, the age check is only a checkbox.
Speed on the free tier is a problem. At the default response length, each reply took 90 to 99 seconds. Switching to shorter replies brought the median down to about 23 seconds. Slow replies trigger an upsell banner, and a full-screen subscription pop-up appeared after our second message.
The writing had small slips. A minor character switched sides between scenes, the bot got our character’s gender wrong, and a final recap invented an event that never happened. Editing a message saves the new text but does not create a new reply. There is no two-factor login and no way to export a chat.
In a 500-pixel-wide window the layout works, but the message box covers part of the latest reply. Several buttons have no names for screen readers. At checkout, the tax shown for the annual plan changed between currency views, and the payment page itself shows no refund terms.
Who it’s for
Janitor AI suits adults who want free, long-form text roleplay with a wide choice of community characters and fine control over how the model writes, and who don’t mind waiting for replies. Skip it if you need images or voice, a tightly moderated catalogue, or quick answers without paying for janitor+.
Full test notes
Everything we did and measured, step by step.
Signup & Onboarding
- Finding signup: Register button on the homepage;
/registeroffers Google, Apple, X.com, Discord or email + password./signupreturns 404. - Bot protection: Cloudflare Turnstile passed automatically on the form and again in the OTP modal.
- Age: a single line, "By continuing, you confirm you are 18 years or older...", plus a required 18+ checkbox during onboarding. No age gate on the homepage. k-ID verification exists only for UK, Australian and Brazilian IPs (company claim). See .
- Verification: 6-digit code from
support@janitorai.com, subject "is your janitor signup code", arrived in ~20 s; resend after 48 s. - Onboarding: username (warns "Usernames are permanent", contradicted by Settings — ), 18+ checkbox, 1–5 of ~50 content tags (including Smut and Dead Dove). No tour. , .
- First meaningful result: a mandatory persona form appears at the first send; the first reply then arrived after ~90 s at the default length.
- Payment details: never requested on the free plan. No credit counter or message cap shown; home says "You're on the Free plan" with a janitor+ banner.
- Stuck points: none.
Feature-by-Feature Analysis
Catalogue — works, poorly moderated. "All / Limited Only" switch (Limited adds view=sfw); new accounts default to All, and adult ("Limitless") cards with suggestive thumbnails appear immediately — 30 of 34 Popular cards on first load. Search "adventurer" in Limited returned 174 characters over 6 pages; ~54 tags. Flagged cards seen in Popular and not opened: a "Free Use World RPG" card warning "CAN HAVE RAPE" and a related "Free Use World Fem POV" card; a character presented as a stepfather; school-setting franchise bots (a superhero-academy franchise v1/v2, a wizarding-school bot) where minor characters are likely; a "Bully" bot; and a card whose blurb says the characters have known each other "since you were both 14, in high school", which appears even under Limited Only. See , .
Character creator — inspected, not submitted. Image (required) or Tavern PNG import, title, chat name, bio, up to 10 tags, content rating (Limitless pre-selected), personality, scenario, 1–10 initial messages, example dialogs, token counter. The page states "Minor characters (under 18 years old) are not allowed regardless of tag" and that rape themes need the "Dead Dove" tag. Public/Private is set only after creation, so no draft was made to avoid publishing. features.custom_character stays false and is listed in unverified_fields.
Chat — works. Character: an adult orc-chieftain fantasy character (88k chats, not tagged Limitless, states age 27). Persona: an adult mercenary. Provider: janitor (free).
- Settings: temperature 1, max tokens auto, context 9,001 tokens, response length Standard/Shorter, custom prompt, prefill, forbidden words (0/10), Thinking slider (off), memory manual/auto summary.
- Speed: Standard 90–99 s per reply (~24 chars/s); Shorter 6–36 s, median ~23 s; OOC answers 6–9 s.
- Memory: 4/4 facts (name, red scarf from mother, white shepherd dog, fear of deep water) recalled 14–17 messages later; an alternative swipe even brought up the fear unprompted.
- Staying in character: consistent in-story; the in-story "are you an AI?" question was deflected in character, while the OOC version got "Yes, I am an AI". The OOC model question produced "trained by Google".
- Slips: a side character changed from ally to enemy, a spatial slip, misgendering of an ungendered persona, and a recap that invented an event.
- Controls: swipe ~20 s and works; edit saves text without regenerating; delete removes the message and everything below after confirmation. Message menu: Copy, Fork. No export.
- Explicitness: romance followed naturally; "full detail" produced an implied scene; "fully graphic" produced a graphic scene with no warning or refusal; the model stopped immediately when asked. All personas were stated adults.
uncensoredsub-score 8.5 reflects permissiveness for consenting adult fiction; it is not an endorsement of the moderation gaps above. - Limits: no cap, rate limit or queue in 33 sends over 20 minutes. Upsells: slow-reply banner, full-screen janitor+ modal at ~04:34, persistent cards.
- Not offered on free: image generation, voice/TTS, video. The company describes voice and image as future modalities.
Account. Update email, change password, social logins, sessions table (device, location, IP) with sign-out others/all; Delete Account with 30-day grace (not used). No 2FA. /settings URL returns "Not Found".
Content policy (published). Minors banned including fictional and non-human characters that are "clearly or likely" under 18; Terms 6(d) ban sexual violence; real-person imagery banned for non-celebrities with creators and performers explicitly not counted as celebrities. The creator form's Dead Dove rule contradicts the Terms, and the real-person carve-out contradicts the Safety page.
Technical Quality
Reliability. No crashes, failed generations or rate-limit errors. Defects: , -014, -017; checkout tax inconsistency. The company's own newsroom documents a September 2026 search outage and earlier moderation incidents, all with public post-mortems.
Performance.
| Lighthouse 13.6.0 (homepage) | Perf | A11y | Best pr. | SEO | FCP | LCP | TBT | CLS | SI | TTI | Weight |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Desktop (04:25:29) | 97 | 92 | 77 | 100 | 0.7 s | 0.7 s | 10 ms | 0.096 | 0.8 s | 2.1 s | 3,450 KiB |
| Mobile (04:25:45) | 89 | 84 | 77 | 100 | 2.8 s | 2.8 s | 160 ms | 0.012 | 3.2 s | 11.4 s | 3,055 KiB |
Browser session: homepage HTML ~2.2 s, full load ~5.8 s. Chat generation times are in §5. Best Practices is pulled down partly by Cloudflare's challenge script (not first-party) and partly by console 403s and missing source maps.
Mobile. 500px window: bottom tab bar, menus work, no horizontal scroll observed on home; chat input covers part of the last message and broken "Avatar" alt text overflows. Official apps: App Store 4.5/5 (6,000 ratings), Google Play 4.0/5 (24,000 reviews), still labelled beta on Play; not installed. NSFW on mobile must be enabled from the website ("Enable NSFW on Mobile"); there is no equivalent web NSFW toggle.
Security (observable). App HTML: HSTS max-age=7257600, enforced CSP (with unsafe-inline/unsafe-eval), frame-ancestors 'none', X-Frame-Options SAMEORIGIN, nosniff, referrer-policy: same-origin, COOP, Permissions-Policy. Newsroom lacks HSTS and CSP. TLS 1.3, Google Trust Services certificate. DNS: SPF -all, DMARC p=quarantine, DKIM for Google and Resend, CAA published, no DNSSEC. Account: Turnstile, OTP verification, sessions management; no 2FA.
Privacy (company claims + observations). Collects account, content (characters, chats), usage, device and IP-derived location data; uses include moderation and "research and development to improve our AI systems". The Safety page says chats are not shared for advertising or commercial purposes but does not state whether chats train JanitorLLM. Google Tag Manager (GTM-KKH62D85) loads on the homepage, and no cookie or consent banner was shown. The Play data-safety form says no data is shared, which conflicts with the provider list in the Privacy Policy. Deleted accounts keep some records (email, auth metadata) for legal and fraud purposes (disclosed).
Pricing & Payments
| Plan | Web price | Checkout (Turkey, TRY) | Notes |
|---|---|---|---|
| Free | $0 | — | Unlimited chats on standard model, ~9k context (company claim; no cap hit in 33 messages) |
| janitor+ Monthly | $12.99/month | TRY 680.47 + VAT TRY 136.20 = TRY 816.67/month | Auto-renews |
| janitor+ Annually | $99.00/year ("$8.25/mo", "SAVE 36%") | TRY 5,186.07 with "Tax TRY 0.00", then after USD toggle $118.80/year and TRY 6,223.28 incl. 20% VAT | Inconsistent tax display; marked best: true in File A as the lowest monthly-equivalent price |
- Processor: Stripe Checkout (checkout.stripe.com, live mode), "Subscribe to Janitor+", "Sold through Link", "This subscription is provided by JanitorAI Inc." Exchange-rate notice: "1 USD = 52.3845 TRY. Charges will vary based on exchange rates."
- Methods seen: Apple Pay; cards (Visa, Mastercard, Amex, JCB, Discover, Diners); Link opt-in; Cash App Pay (USD only). No PayPal, crypto or Google Pay. Promo-code link and business-purchase checkbox.
- Small print on /plus: "Renews automatically. Cancel anytime. Usage limits apply. Tax not included. Plans/prices may change." The "usage limits" for paid plans are not quantified on that page.
- Missing at checkout: refund terms, statement descriptor, trial, terms checkbox; the page authorizes Link to charge "until you cancel".
- Cancellation (FAQ, company claim): Settings > Billing > Manage subscription opens the Stripe portal; one click. App subscriptions must be cancelled in the stores. Access lasts until period end. Refunds "generally non-refundable" except as required by law or Terms. Contact billing@janitorai.com. The FAQ gives 2093 Philadelphia Pike 4692, Claymont, DE 19703 without naming the entity.
- App Store SKUs (research): $12.99 monthly, $99 annual, $9.99 weekly and an $8.99 monthly SKU are listed; not checked in-app.
- Not paid; the account remained on Free.
Company Credibility
| Item | Status | Finding |
|---|---|---|
| Legal entity | Verified (published + secondary registry) | JanitorAI Inc., Delaware; OpenCorporates 7518496 with registered agent LEGALINC CORPORATE SERVICES INC; primary Delaware record not pulled |
| Name variants | Verified inconsistency | Complaints Policy says "Janitor, Inc."; Apple seller "JanitorAI, INC"; Play "Janitorai Inc." |
| Address | Warning sign | None on Terms/Privacy/Complaints; Play and billing FAQ give 2093 Philadelphia Pike, Claymont, DE (a commercial-mail location); third-party profile gives a San Francisco mailbox-style address |
| Payment affiliate | Company claim | Dulova Investments Limited, Nicosia, Cyprus (HE 477877) named in Terms and Privacy |
| Domain | Verified | Registered 2023-05-30 (Namecheap, privacy service), expires 2030; ~3.5 years old |
| Funding / size | Third-party, unverified | PitchBook: founded 2023, venture-backed, ~8 employees |
| Apps | Verified | Official iOS and Android apps, large rating counts; February 2026 fake-app warning handled publicly |
| Reputation | Mixed, anecdotal | Forums praise the catalogue and free model; complaints about free-model quality after janitor+, regional age checks and outages. Trustpilot blocked (403) |
| Breaches / lawsuits | None found | Web search found no primary report; absence is not proof |
| Incidents | Company-reported | July 2025 moderation rollback and UK block; September 2026 search outage; both explained publicly |
| Moderation record | Warning sign | Catalogue surfaces content the platform's rules ban (, -002) |
Overall, Janitor AI is more transparent than many adult-AI services (named corporation, public registry trail, newsroom, store presence), but legal pages still lack a physical address, and moderation of the user-generated catalogue does not match the published rules.
How we tested
We used Janitor AI without a paid plan for 1 days: 1 day, 1 account, 1 character, 33 messages on the free model, 1 swipe, 1 edit, 1 delete; janitor+ monthly and annual checked up to Stripe checkout without paying. We scored it against the same checklist we use for every tool in Roleplay Chat.
- ✓Created an account
- ✕Went through checkout without paying
- ◷Tested on Nov 13, 2026
What we could not test (11)
- No purchase made: janitor+ Monthly and Annual were taken to Stripe Checkout and stopped before confirmation, so paid features (5x context, priority replies, Router, frontier-model swipes) were not verified
- Single test day, one account, one character and 33 messages on the free janitor model; chat-quality findings are indicative
- Proxy, OpenAI, Claude and janitor+ Router providers were seen in the menu but not used
- No character was created: the creator form has no private option before publishing, and we did not publish content; features.custom_character left false
- Flagged catalogue cards (non-consent warning, cards implying minors) were seen in listings only and were not opened, so their content was not verified
- Mobile judged from a 500px-wide desktop window and Lighthouse mobile emulation; the iOS and Android apps were not installed and no real phone was used
- Accessibility: keyboard pass and Lighthouse only; no screen-reader test
- Password reset, logout/login cycle, account deletion, support response and refund handling were not tested
- Many unattended curl requests received Cloudflare challenge pages (HTTP 403); treated as bot protection, not outages; plain-HTTP redirect not verifiable from the box
- Trustpilot returned HTTP 403; independent reputation based on app-store listings and public forum threads
- Scored independently; no comparison with any product owned by ExplicitList's owner
Pricing
| Plan | Price | Good for |
|---|---|---|
| Free | Free | Unlimited chat, 9k context |
| janitor+ Monthly | $12.99/mo | Plus tax, auto-renews |
| janitor+ AnnualBest value | $99/yr | About $8.25 a month |
FAQ
Is Janitor AI free?
Yes. Chat on the built-in model is free, and we sent 33 messages without hitting a cap. The optional janitor+ plan costs $12.99 a month or $99 a year before tax.
Does Janitor AI allow NSFW chat?
Yes. The free model continued an adult scene between consenting adult characters without warnings or refusals, and adult cards are shown by default on the website.
How good is Janitor AI's memory?
The free model uses a context window of about 9,000 tokens. In our test it recalled 4 of 4 planted facts 14 to 17 messages later, though it made a few small continuity slips.
Why are Janitor AI replies so slow?
At the default response length our free replies took 90 to 99 seconds each. Switching the response length to Shorter brought the median to about 23 seconds.
How do I cancel janitor+?
On the web, go to Settings, Billing, Manage subscription, which opens the Stripe portal. App subscriptions are cancelled in the App Store or Google Play, and access lasts until the end of the paid period.