Dondi AI Review

Dondi.ai publishes a Sex Trafficking Policy. It publishes a 2257 exemption with a street address on it. It publishes a Blocked Content Policy that bans deepfakes of private individuals by name, and a removal policy that promises an answer within five business days.

It also registered its domain on 7 January 2026, dated every one of those documents 08 January 2026, and has not revised a single one since. The rulebook is excellent. It was written on day one, and the product is nine months old.

What is Dondi AI?

Dondi AI is an uncensored companion platform: pick or build a character, chat and roleplay without a content filter, generate images and short videos of them, and get voice messages back. The marketing site is WordPress at dondi.ai; the product itself runs on app.dondi.ai behind a login, which is where every button sends you.

The character roster is fictional and reads like it: Emily, 28, "your fit naughty roommate"; Kairi, 23, an art school dreamer; and a male cast including Alessandro Costa, a 28-year-old Italian chef, and Kai Chen, 26, a Shanghai bartender. Occupations and ages, no real people, and an 18-plus age gate on entry.

The operator is disclosed, though you have to hunt for it. The footer says "Powered by dondi.ai Labs Inc." The terms of service name no legal entity at all. The only address anywhere is on the 2257 page: 18627 Brookhurst Street 1300, Fountain Valley, CA 92708. Three documents, three different amounts of disclosure, which is untidy rather than evasive.

Policy check, and Dondi passes it more comprehensively than anything else we have looked at. Across the full marketing site, every feature page and 1.6 MB of application bundles, there is no image-to-image, no reference image, no identity adapter, no face swap, no nudify, no undress and no inpainting. The hits our scan returned were noise of the most boring kind: 1,429 matches on uploads are all wp-content/uploads image paths, 120 on uploadDate are schema.org video markup, and the handful of real upload strings are axios internals, onUploadProgress and an xhr.upload listener, sitting in the HTTP client.

Then the platform states the same thing. From its Blocked Content Policy: "Although AI content is not directly user-uploaded, certain outputs from AI prompts may still fall into prohibited categories and are subject to moderation and removal." A platform describing its own architecture accurately, and building a moderation policy around the risk that remains, is not something we get to write often.

What it prohibits is worth quoting, because most of this category does not bother. Section (g), Impersonation, Deepfakes, and Deceptive Content: "Content intended to impersonate, misrepresent, or falsely portray a real person, particularly public figures, private individuals, or celebrities, is strictly prohibited. This includes non-consensual intimate imagery (deepfakes)." Section (h) bans intimate content depicting anyone "without the clear, prior consent of the depicted party". Section (e) bans descriptions, names or characteristics that could be read as child-like.

And the 2257 exemption makes the correct argument for a product built this way: "dondi.ai operates as a purely AI-generated content platform. No real individuals are portrayed, depicted, or involved in the creation of any content available on the platform."

The features that actually matter

Twelve policies, and they are real documents. Terms of Service, Privacy Notice, Cookies Notice, Underage Policy, Content Removal Policy, Blocked Content Policy, DMCA Policy, Complaint Policy, 18 U.S.C. 2257 Exemption, Community Guidelines, Affiliate Terms and a Sex Trafficking Policy. Eromantic AI, which we reviewed yesterday and praised for its legal shelf, had no 2257 statement and shipped one policy as an unfilled placeholder. Dondi has the full set with nothing missing.

A removal process with a deadline. To report a generated face that resembles someone, you supply a description, a link or screenshot, the affected person's name and contact details, and supporting information. Dondi says it will "review all requests in good faith and respond within 5 business days". Compare ImScarlet, which reserves the right to remove a picture while explicitly declining to promise it.

A memory system as the paid hook. The platform sells persistent memory across sessions, and reviewers who tested it describe that as the main thing you are buying, alongside image generation, video and voice.

Everything costs a token, including listening. Both reviewers make the same point: tokens fund images, video and voice, and playing back a voice message spends one. The paid plan includes 200 tokens a month.

Pricing, without the bullshit

Dondi publishes no prices. We looked in the obvious place and the less obvious ones: there is no pricing page on the marketing site, no figure anywhere in 1.6 MB of application code, and app.dondi.ai/pricing renders a shell containing the single word "Dondi" because the real page is behind a login. That is now the fifth platform in a row to do this.

Two independent reviews agree on those two figures, one of them testing on 19 August 2026. A third tier near $29.99 appears in one source only, so we have left it out rather than guess. As always with an unpublished price: believe the checkout screen, not us.

Now the two clauses that actually cost you something, both verified in Dondi's own text rather than taken from a reviewer.

The refund window is 24 hours. Section 19.2: you may request a refund "within 24 hours of payment", provided you have not exceeded reasonable usage limits and have not "consumed premium features beyond minimal use". Token purchases get the same 24 hours. Section 15.3 adds that nothing is refundable at all if your account is suspended for a policy breach. A day is a deadline, not a trial, and the usage condition means the clock is not really the binding constraint: using the thing you bought is.

Your content may train the model. Section 7.2 of the privacy notice, headed AI and Technology Partners: your content and interaction data "may be processed using AI systems and related technologies operated by us or our technology partners", and that processing "may include AI model training and improvement". We looked for an opt-out and a retention period and found neither. On a platform whose entire purpose is explicit roleplay, that deserves to be the loudest thing on this page, and it is why a product with this much policy discipline does not score higher.

What users say

Almost nothing, and the reason is simple arithmetic. The registry record shows dondi.ai was created on 7 January 2026, so the platform is under nine months old. There is no Trustpilot profile. One trust scanner puts it at 61 out of 100, which for a young adult site means "no history" more than it means "bad history". We found no breach and no security incident.

The reviewers who did test it split the same way we do: good writing and genuinely persistent memory, a free tier too thin to evaluate anything, and paperwork that reads better than the product's age would suggest.

Which brings us back to the date. Every one of those twelve policies carries "Date of Revision: 08 January 2026", the day after the domain was registered. Whoever set this platform up did the governance work properly before launch, which is the right order and almost nobody does it. Nine months later, not one document has been touched. The policies are good. Nothing yet proves they are enforced.

Final verdict: who is Dondi AI for?

Get it if you want an uncensored companion from an operator that has written down, in specific language, what it will not generate and how to get something removed, and you will buy the annual plan.

Skip it if you want your explicit conversations kept out of a training set, you need more than 24 hours to decide, or you will not enter a card before seeing a price.

The recommended formula is the annual plan at $71.88, bought with a disputable card, on the understanding that what you type may train the model.

Dondi AI inverts the usual problem in this category. Most platforms here ship a decent product wrapped in paperwork nobody read, and we spend these reviews finding the placeholder text. Dondi shipped the paperwork first and did it properly: twelve policies, a correctly argued 2257 exemption, a real address, an explicit deepfake ban and a removal deadline in business days. What it has not yet earned is a record, and what it asks in exchange is your content for its training and one day to change your mind. 7.2 is a good score for a platform this young, and it is the training clause, not the age, holding it below the next mark.

Rating: 7.2/10