Uncensored AI is a subscription aggregator that resells frontier models from Anthropic, OpenAI, Google and others as "uncensored" variants, alongside two in-house Model T models. Operated by Found Tech LLC in Wyoming, it publishes nine compliance documents but names no model supplier in any of them.
Used this tool? Rate it
Used this tool? Rate it
Uncensored AI is a subscription platform that sells access to a catalogue of large language models, most of which it did not build. Its own homepage defines the product in dictionary form: uncensored(ai) is "the most capable AI models, made more truthful and unbiased." That definition is unusually candid about the business model. The value proposition is not a novel model — it is access to other people's frontier models with the guardrails loosened, gathered behind one subscription and one interface.
The service is operated by Found Tech LLC, registered at 1309 Coffeen Avenue STE 17591, Sheridan, Wyoming 82801, with a published phone number and a support address at the uncensored.com domain. That entity name appears in the footer of all nine compliance documents, in the copyright line, and in the 2257 statement where it is listed as Custodian of Records. The Complaints Policy states the relationship in plain language: "Uncensored AI is managed by Found Tech LLC., a US-based company." For a category where anonymous operators are common, a named legal entity with a consistent address across every legal page is a genuine point in the product's favour, and it should be weighed as such.
The catalogue page reports 28 models across 9 providers. Two of them — Model T 2.0 Uncensored and Model T 2.0 Extended Uncensored — are labelled as the platform's own flagship. The remaining twenty-six are labelled with their originating vendor: seven Claude variants under Anthropic, three GPT tiers under OpenAI, four Grok variants under xAI, four Gemini variants under Google, three Kimi models under Moonshot AI, two GLM models under Z.AI, plus DeepSeek R1 and MiniMax M2-Her. Individual model pages carry structured fields stating the provider outright, and describe several entries as variants built on a named base model — the Claude Opus 4.7 v2 page, for instance, calls itself "a variant of Claude Opus 4.7, tuned for roleplay and creative writing on top of the base model's frontier reasoning."
The catalogue's own summary line is more measured than the brand name: it invites you to browse every model available "— most of them uncensored." That qualifier is worth noticing, because it concedes that not every entry in the catalogue ships an unrestricted variant.
Nor does the platform position itself as rule-free. The Acceptable Use Policy prohibits child sexual abuse material and anything that sexualises minors, real or fictional; incest, bestiality and non-consensual sexual content, along with anything suggestive of those themes; any realistic human face or real likeness without logged consent; content that harms, harasses, defames, threatens or promotes hate, violence or terrorism; malware and phishing; and intellectual property infringement. Enforcement is described as four layers: automated real-time scanning of all inputs and outputs, instant blocking of prohibited input before it reaches the model, human review of flagged material, and a built-in reporting control on every generated output. For anyone uploading media, the policy adds three requirements: you must confirm that you own the rights to the content, you must agree not to upload media violating the policy above, and uploaded content is used only for your personal AI session rather than being published or shared. Repeated violations may lead to content removal, account suspension or permanent termination.
What that list does not include is adult sexual content as such. The Content Removal Policy sets the boundary explicitly by making "fails to meet standards for legal and consensual adult content" a ground for removal — a legality-and-consent line rather than a blanket prohibition. Readers should take that as the actual policy position: consensual adult material appears to be within scope, while the categories above are not.
The practical proposition is breadth. Rather than holding separate subscriptions with several model vendors, a subscriber gets one account that reaches Claude, GPT, Gemini, Grok, DeepSeek, Kimi, GLM and MiniMax entries alongside the in-house Model T line. The Discover surface is built around this, described on the homepage as "Compare across 50+ models. Pick a winner," with a one-click path to an API for developers.
For someone whose actual need is comparing how different frontier models handle the same prompt, that consolidation has real value independent of the uncensored framing.
The homepage lists six capabilities: frontier models, real-time web search, speech, a Code Mode described as handling live data fetching and updates, AI predictive inputs, and vision models. The three-way product split is Chat, Generate (image creation and editing) and Discover. A voice affordance is described at the interaction level — tapping a response plays it aloud — and the platform advertises conversation in multiple languages.
Several catalogue entries are marked "Community requested · roleplay & writing" and identify the base model they were built from: a Claude Fable 5 Roleplay variant built on Claude Fable 5, a Claude Opus 4.6 v2 built on Claude Opus 4.6, and a Claude Opus 4.7 v2 built on Claude Opus 4.7. A Gemini Emotional Uncensored entry is described as built on Google's Gemini 2.5 and "made empathetic." This is the clearest window the site offers into what it is actually doing to the models it resells — tuning them toward roleplay and creative writing on top of a named base.
Two specific mechanisms are claimed rather than left as slogans. The Chat surface advertises auto-redaction to keep sensitive details private, and preference learning over time. The feature list pairs "Anonymous & Private" with real-time web search and multi-language chat. How far the anonymity claim actually reaches is covered in the limitations section below, because the Privacy Policy describes a rather different data posture.
Beyond terms and privacy, the site publishes an Acceptable Use Policy, a 2257 record-keeping statement, an Anti-Trafficking Policy, a DMCA Policy, a Content Removal Policy, a Complaints Policy and a Fulfillment Policy. Nine documents is more than most products in this category maintain, and the substance is operational rather than boilerplate — the 2257 statement, for example, distinguishes AI-generated content (claimed exempt because no human performers are involved) from user-uploaded media, which must pass third-party age and consent checks through a service named Didit before each approved upload is assigned a record number and stored.
This is the step most worth slowing down on, because the pricing structure has a division that is easy to miss. Every tier includes the chat application features — text models, voice, vision models — plus a free or included allowance for supported non-chat features. Credits are a separate currency layered on top, and they buy only non-chat capacity.
Three plans are published. The monthly plan is $29.99. A premium monthly plan is $99 and includes 1000 credits. The yearly plan is $149.99, which works out well below twelve months of the base monthly rate. If your use is overwhelmingly text conversation, the credits in the $99 tier will not help you, and the yearly plan is the obvious value choice. If you generate a great deal of imagery, the credit allowance starts to matter.
Credits can be purchased in-app to top up a balance, at one credit per ten US cents, and are consumed at feature-specific rates that the policy does not enumerate. The included allowances are described as capacity-based and may vary with system demand, so they are not a guaranteed monthly quantity. Heavy use may trigger a prompt to enrol in consumption-based billing.
With 28 catalogue entries, the default is rarely the right choice for a given task. The model pages state the context window and a short capability note — several Gemini and Claude entries advertise a one-million-token context window, GPT-5.4 and GPT-5.6 Sol Pro likewise, and Grok entries carry architecture notes. Reading the provider field and the highlights before starting a long session is worth the minute it takes.
Cancellation runs through the "Manage Subscription" control in account settings, which redirects to the payment partner's portal, or by emailing support. Cancelling stops future billing and preserves access to the end of the paid period. There are no prorated refunds for the unused remainder, and the general rule is that subscription payments are final and non-refundable except at the company's discretion or where law requires otherwise. Set a calendar reminder before renewal rather than relying on being able to unwind a charge afterwards.
Charges appear on card statements as "UAI Premium" rather than under the product name. If you are reconciling a statement months later, that descriptor is what you will be looking for.
The strongest legitimate case is evaluation. Running the same brief through a Claude variant, a GPT tier and a Gemini entry inside one interface, with the ability to switch and re-run, is a workflow that would otherwise require several accounts and several billing relationships.
The community-requested variants exist specifically for this, and the large advertised context windows suit sustained narrative work. Several catalogue entries are described as tuned for roleplay and writing on top of a frontier base model.
There is a genuine and legitimate need here: writers, researchers and clinicians regularly encounter refusals on subject matter that is entirely lawful to discuss. A platform that declines to refuse on those grounds addresses a real friction point, and the Acceptable Use Policy's prohibited list remains in force regardless.
Image generation and editing, web search, voice output and vision models under a single subscription suits users whose work crosses formats and who would rather not assemble a stack of separate tools.
It is not the right vehicle for regulated or safety-critical output. The terms state plainly that outputs may be inaccurate, incomplete or outdated, and that AI-generated content is not reliable as a sole source of truth. Nor is it suited to anyone who needs a documented, auditable supply chain for the models they rely on, for reasons set out in the limitations section.
The clearest fit is an individual buyer who wants breadth of model access at a consumer price point, is doing personal or low-stakes work, and specifically values not being refused on lawful subject matter. For that person, one subscription reaching several vendors' models is a straightforward convenience, and the $149.99 yearly price is competitive against holding multiple direct subscriptions.
A second fit is the evaluator — someone comparing how frontier models from different labs handle the same material, who would otherwise be assembling several accounts to do it. The Discover surface and the breadth of the catalogue serve that use directly, and the comparison value holds regardless of whether the unrestricted variants interest you.
Writers working in genres that mainstream assistants handle poorly are a third fit, and the community-requested variants tuned for roleplay and long-form writing exist for exactly this audience.
Where the fit breaks down is professional and commercial work with accountability requirements. If you may need to state which model produced a given output, demonstrate a documented supply chain, or point to a contractual commitment about how your data reaches a model vendor, the disclosure gaps below make this the wrong instrument — the governing contract calls every model "our AI models" and names no supplier. Going direct to a vendor costs more and restricts more, and buys you a paper trail.
It is also the wrong choice for anyone who needs verified scale or track record before committing. The published statistics do not survive checking, the independent review sample is nineteen entries, and no independent press coverage exists. None of that proves the service is not what it says; it means a prospective buyer has very little outside the company's own account to weigh.
Model detail pages state access as web, iOS and Android. In practice the web application at uncensored.com is the surface that can be verified directly, and the site provides no store links: neither apps.apple.com nor play.google.com appears anywhere in the homepage or catalogue markup.
That absence matters when assessing the mobile claim. A search of Apple's official lookup interface across three query variants returned no application published by Found Tech LLC and no "Uncensored AI" application corresponding to uncensored.com. There is an "Uncensored AI" application on Google Play, but its developer details identify a different company at a Nebraska address with a support address on the uncensored.ai domain — a separate entity, covered under same-name disambiguation below.
The platform also maintains a developer-facing surface, with an API and models section, a referral programme and an external link to Dataset AI, alongside a Discord community.
Three published tiers: $29.99 monthly, $99 monthly with 1000 credits included, and $149.99 yearly. All tiers grant access to the full chat application feature set — text models, voice interaction, vision models — plus a free or included allowance for supported non-chat features.
The credit layer sits alongside the subscription. One credit costs ten US cents, credits are consumed at feature-specific rates, and included allowances are capacity-based rather than fixed. The constraint that most affects value judgement is stated explicitly: credits apply only to supported non-chat features and cannot be applied to, exchanged for, or redeemed toward chat-text model access. A buyer choosing the $99 tier for its 1000 credits should understand those credits do nothing for text conversation.
All plans auto-renew at the end of their billing cycle unless cancelled beforehand. Payments are handled by a payment partner, with the company stating it does not store complete card details on its own servers. Failed payments trigger an email notification, with premium access potentially suspended until payment information is updated. Refunds are, as a general rule, final and non-refundable, subject to company discretion or legal requirement, with no proration on cancellation.
The honest comparison set is other multi-model aggregators and other loosened-restriction chat services, and the useful framing is what you trade away.
Going direct to a model vendor gets you a documented commercial relationship, published model cards, a stated data-handling posture from the company that trained the model, and a supply chain you can audit — at the cost of separate subscriptions and stricter content policies. Using a mainstream aggregator gets you breadth with supplier relationships that are typically disclosed in the terms, again with standard content policies. Choosing this product gets you breadth plus loosened restrictions, and the trade is the disclosure gaps set out below.
Trustpilot's own "people also looked at" panel places the service beside Venice at 1.9 from 54 reviews, JOI at 4.1 from 262, and Secrets AI at 4.4 from 200. Those sample sizes are all small enough to be indicative at best, and the panel is a browsing artefact rather than a curated comparison, but it does map roughly where the product sits in users' consideration sets.
This is the most consequential gap and it was verified by counting rather than by impression. Across all nine compliance documents, Anthropic, OpenAI and xAI each appear zero times. Google appears three times, all within the Privacy Policy and all referring to Google Authentication and Google Analytics — login and analytics, not model supply. The model brand names Claude, Gemini, Grok and GPT likewise appear zero times across those nine documents.
The Terms of Service go further than silence. They describe the models in the first person possessive: "our AI models generate text, images, code," and "our Large Language Models (LLMs), while powerful, may sometimes produce Output that contains inaccuracies." A reader of the contract alone would conclude the company built what it is selling.
To be precise about the scope of this gap: disclosure does exist elsewhere on the site. The catalogue page names providers repeatedly — Anthropic twelve times, Google nine, xAI eight, OpenAI six, Moonshot five, DeepSeek three — and individual model pages state a Provider field outright. So the accurate statement is that supplier attribution is present in the product catalogue and absent from every contractual document. Which of those a buyer should weigh more heavily is a judgement call, but a marketing page is not a contractual commitment, and the document that governs the relationship is the one that says "our AI models."
Searched across all nine legal documents plus the catalogue page, "not affiliated" appears zero times. The word "respective" appears once, in the Fulfillment Policy, describing "respective billing cycles" — nothing to do with trademark ownership. "Trademark" appears once, in the Content Removal Policy, as a prohibited-conduct item ("infringes on copyrights or trademarks"), not as an attribution notice.
Set that against usage: the homepage uses Gemini 22 times, Claude 21, Grok 21, Opus 19, GPT 16, alongside Kimi, DeepSeek, MiniMax and GLM. The brands do heavy work in the marketing; the companies behind them are never named on the homepage at all, and nowhere on the site is there a statement that those marks belong to their respective owners or that this service is independent of them. Readers should note the absence and draw their own conclusion about what it signals.
Searched for the standard formulations — "18 years," "eighteen," "age of majority," "18+" — the Terms of Service return zero matches, and "minor" returns zero as well. The contract that governs use does not make age a condition of entering it.
Age appears in two other places with different force. The Privacy Policy states that the services are not intended for individuals under 18 and that the company does not knowingly collect information from anyone under that age, deleting it promptly if discovered — a data-collection posture, not an access control, and the policy describes no verification at registration. The 2257 statement is stronger but narrower: user-uploaded media must pass third-party 18+ age and consent checks via Didit before approval. That is a real check, but it gates the upload pathway rather than account creation or conversation.
The homepage statistics block displays "4.4 Avg. App Store rating." Queried directly against Apple's official lookup interface using three separate search terms, no application published by Found Tech LLC and no Uncensored AI application matching this service was returned in the US store. The site itself provides no App Store link. The figure may refer to a store, storefront or historical listing not surfaced by that interface, but as published it cannot be checked by a reader, and no reader should treat it as an independent rating.
In the same block, "Happy customers" and "Model iterations" both render as zero — literally "0 m+ Happy customers" and "0 Model iterations" — and the block repeats three times on the page with zeros each time. Elsewhere the homepage states "Enjoyed by 10m+ users worldwide." A statistics panel where two of three figures are broken placeholders and the third is an unverifiable rating is not a sound basis for judging scale, and this page does not repeat any of those numbers as fact.
Trustpilot shows 3.1 from 19 reviews for uncensored.com. Nineteen is below the twenty-review floor at which a rating starts to carry signal, so it is reported here for completeness rather than as a verdict. The profile is claimed as of July 2026, and Trustpilot itself flags two caveats: no recent history of inviting customers, meaning the reviews may not be representative, and no replies to negative reviews.
The customer reviews on the homepage span August 2023 to July 2026 and carry developer responses, in the format of app store reviews, but without a stated source platform. Every quoted rating is four or five stars; nothing at three or below appears. Against an independent score of 3.1, the absence of any middling or negative entry indicates selection rather than a representative sample. Treat that section as marketing.
"Always anonymous" appears in the product's own descriptions, and the feature list pairs "Anonymous & Private." The Privacy Policy describes something more ordinary: accounts are created through Google Auth with the email address stored, and log data collected through third-party analytics includes IP address, device name, operating system details, access times and usage statistics. The reasonable reading is that conversations are not tied to a public identity, not that the operator holds nothing identifying. The stated retention target after account deletion is 90 days, and data may be transferred to and stored on servers outside the user's country. The policy's own effective date is January 2023.
Worth restating because it is the single most likely purchase mistake: the 1000 credits bundled with the $99 tier cannot be applied to chat-text model access. Anyone upgrading from $29.99 to $99 for more conversation is buying capacity for a different part of the product.
Subscription payments are final and non-refundable as a general rule, with exceptions at company discretion or where law requires. There is no proration for the unused portion after cancellation. Combined with auto-renewal on every plan and a card descriptor that does not carry the product name, this is a billing structure where attention before renewal matters more than recourse afterwards.
The Complaints Policy states it does not cover complaints about content moderation decisions or copyright infringement. Copyright has a separate DMCA route, which is reasonable. Moderation decisions have no equivalent dedicated policy, which means a user whose content was removed has no documented channel for contesting that specific decision.
The catalogue summary reads "most of them uncensored," which is a narrower claim than the brand name makes. A prospective subscriber whose interest is a specific model in an unrestricted form should confirm that particular entry rather than assuming the catalogue is uniformly unrestricted.
Its stated effective date is 21 January 2023. The catalogue advertises models released well after that, and the compliance document set has clearly grown since. A privacy policy three years older than the product it governs should be read with that gap in mind.
Anyone researching this service will encounter a second, unrelated product with the same name, and confusing them will produce wrong conclusions.
The Google Play listing titled "Uncensored AI" — 3.7 stars from 544 reviews, over 100,000 downloads, rated Everyone, updated July 2026 — lists its developer at 1901 Howard St Ste 306, Omaha, Nebraska, with a support address on the uncensored.ai domain and a Nebraska telephone number. Every one of those details differs from Found Tech LLC's Wyoming address, uncensored.com support address and Texas-area phone number.
Checking the domains confirms the split. The site at uncensored.ai returns its own page without redirecting, carries the slogan "Look Into It," and hosts its own distinct assets and product description. Meanwhile chatuncensored.ai redirects to uncensored.com, and the media assets on uncensored.com's homepage are served from a media subdomain of chatuncensored.ai — so that domain belongs to the operator described here, and uncensored.ai does not.
The practical consequence: the Play Store ratings, download figures and reviews belong to the other product and say nothing about this one. Every finding on this page is anchored to uncensored.com.
A targeted search across the major technology publications returned no reporting, funding disclosure or independent review of this service. That absence carries some weight for a platform advertising "10m+ users worldwide" and reselling several vendors' frontier models, since such an arrangement would ordinarily leave traces in industry coverage. Combined with the broken statistics block, the unverifiable store rating and the nineteen-review Trustpilot sample, it means every scale figure on this page is vendor-reported and independently unaudited.
Entries with similar names behave differently. A "v2" variant is described as tuned for roleplay and writing on top of a base model, which is a different proposition from the base entry itself. The provider field, the context window and the short capability note are on every model page and take a moment to check.
The yearly plan is priced well below twelve monthly payments, which makes it attractive — but refunds are final and non-refundable by default with no proration. Establishing that the service actually fits your work on a monthly cycle first turns a good price into a good decision.
Nothing in the published policies commits to export, portability of conversation history, or retention of your chats for any particular period. If a session produces something you need later, copy it out as you go.
Because the catalogue itself says "most of them uncensored," the label on the catalogue page is not a guarantee for any specific entry. A short test in your actual subject area tells you more than the marketing does.
Auto-renewal applies to every plan, refunds are final by default, and the card descriptor reads "UAI Premium" rather than the product name. A calendar reminder a few days before renewal is the only reliable control a subscriber has.
For casual personal use, the absence of supplier naming in the contract may not matter to you. For work with professional or commercial exposure — anything where you might need to state which model produced an output, or demonstrate a documented supply chain — it matters a great deal, and going direct to a vendor is the safer path.
Found Tech LLC, a US company registered at 1309 Coffeen Avenue STE 17591, Sheridan, Wyoming 82801. That name appears in the footer of all nine compliance documents, in the copyright line, and as Custodian of Records in the 2257 statement, and the Complaints Policy states directly that the service is managed by that company. A published phone number and support email accompany it.
The catalogue page and individual model pages name the providers: Anthropic for the Claude entries, OpenAI for GPT, Google for Gemini, xAI for Grok, Moonshot AI for Kimi, Z.AI for GLM, plus DeepSeek and MiniMax. Two Model T entries are labelled as the platform's own. Note that none of the nine legal documents names any of these suppliers — the Terms of Service instead refer throughout to "our AI models."
No. Searched across all nine legal documents and the catalogue page, there is no non-affiliation statement, no trademark attribution notice, and no statement that those marks belong to their respective owners. The brand names are used heavily in marketing without any accompanying attribution language.
Three published plans: $29.99 monthly, $99 monthly including 1000 credits, and $149.99 yearly. Credits are a separate currency at ten US cents each, consumed at feature-specific rates. Critically, credits apply only to non-chat features and cannot be redeemed toward chat-text model access.
The Terms of Service contain no age clause at all. The Privacy Policy states the services are not intended for anyone under 18 and that such data is not knowingly collected, but describes no verification at registration. The one real check is narrower: under the 2257 statement, user-uploaded media must pass third-party 18+ age and consent verification through Didit before approval.
The Acceptable Use Policy prohibits CSAM and anything sexualising minors whether real or fictional, incest, bestiality, non-consensual sexual content and anything suggestive of those themes, realistic human faces or real likenesses without logged consent, harassment and hate and violence and terrorism, malware and phishing, and IP infringement. Adult sexual content as such is not on that list; the Content Removal Policy instead sets the line at content that "fails to meet standards for legal and consensual adult content."
As a general rule, no. Subscription payments are stated to be final and non-refundable, subject to company discretion or where law requires otherwise, and there is no proration for the unused portion of a term after cancellation. Cancelling stops future billing and preserves access until the end of the paid period.
Not in the sense of the operator holding nothing identifying. Accounts are created through Google Auth with the email stored, and third-party analytics collect log data including IP address, device name, operating system and usage statistics. The fair reading of the anonymity claim is that conversations are not tied to a public identity. Stated retention after account deletion is a 90-day target, and data may be stored outside your country.
Model pages state access as web, iOS and Android, but the site provides no store links, and a search of Apple's official lookup interface found no application published by Found Tech LLC. There is an "Uncensored AI" app on Google Play, but its developer details point to a different company in Nebraska using the uncensored.ai domain — a separate product, not this one.
Because there are two. This page describes the service at uncensored.com, operated by Found Tech LLC in Wyoming. A separate service at uncensored.ai, with the slogan "Look Into It" and a Nebraska operator, publishes the Google Play application. The domain chatuncensored.ai redirects to uncensored.com and serves its media assets, confirming which side of the split that domain sits on. Ratings and download figures for the Play app do not describe the service covered here.
