Dondi AI Review: Great Progress Bars, Forgetful Companion

The landing page claims "Join over 50 million users". The homepage adds "20M+ Monthly Messages" and "3,000,000+ Monthly Visits".

Put those side by side and the maths gets strange fast: fewer messages a month than users, and a crowd that visits far less often than it joins. I read them the way I read a used-car ad that says "one careful owner".

The part that sets Dondi apart is the game layer. Each companion carries a relationship Level with XP, milestones, a days counter and a visible Memories panel. It is a loyalty-card system bolted onto a chat window, and it is better built than the chat under it.

My verdict is short. Dondi is a polished storefront with a thin conversation behind it and a token meter that runs on almost everything. It suits someone who likes progress bars and short media clips, reads the terms first and pays by the month.

It does not suit anyone who wants a real free tier, a companion who remembers the last few minutes, or a straight answer on who runs the company.

If you are curious, I'd treat it like a rental rather than a purchase. One month tells you everything the app has to say, and the yearly deal only makes sense for someone who already knows they love it.

Key takeaways

• The relationship levels, milestones and days counter are Dondi's most distinctive feature and its best-built one.
• The homepage traffic and user claims do not add up against each other, so I'd ignore them.
• It is one browser app for chat, roleplay and pictures, with the pictures and video behind a paywall.
• The token meter touches messages, voice plays, images and video, so the sticker price is only the start.
• Read the company and refund sections before paying; they are the weakest part of the product.

The shape of those weeks was simple. I chatted, watched the relationship level crawl upward, spent tokens on voice replays and one animated portrait, and waited for the memory to catch up with the progress bar. It never did, and by the end I had stopped expecting it to.

I stopped after three weeks with 37 tokens left. My level is frozen at "Friend", like a loyalty card I'll never finish. Those leftover tokens bother me more than they should, the same way an unused gift card sits in a wallet for years because throwing it out feels like admitting defeat.

What the current facts are based on

The present-day facts in this review, dated 23 September 2026, come from the web version and a new free account. That split matters for reading the rest of the page. My story is a paid month in January; the current facts describe what a newcomer meets today on the free side, and where the two differ I give both.

The free-side facts are fresh while my paid-side memories are eight months old, and I'd weigh them that way.

Settings > Account holds the e-mail, the username, the reply length (a Premium option), push notifications, the token balance, a password change and account deletion. It is a short, tidy page. I'd rather have that than a maze of toggles, although a short settings page also tells you how little of the experience a user can actually adjust.

How it ended

There was no dramatic exit. The app did not break, the bill did not surprise me, and nothing offensive happened; the conversations simply stopped being worth the tokens. For a companion app, running out of reasons is a more honest ending than running out of money, and it is the one I'd expect most people to reach here.

Where it lives

The site is live, and it has two front doors. The marketing site at dondi.ai is a WordPress-style landing page, and the web app itself runs at app.dondi.ai.

I don't mind the browser-only setup. It means no app-store review, no icon on the home screen and no notification badge in the grocery queue, which some people will count as a feature. The two-domain split is more confusing: the brochure and the shop sit at different addresses, and the brochure makes promises the shop does not keep, as later sections show.

The age gate

The landing page opens with an adults-only age gate. Its text reads: "You may encounter mature themes, emotional intimacy, or adult-oriented conversations. Please confirm your age to continue."

That is standard for the category, and I don't hold it against Dondi in particular. I'd just not mistake it for a real age check. It exists so the company can say it asked.

Creating the account

The sign-up page at app.dondi.ai/auth/signup offers "Continue with Google". It also offers e-mail plus password, and the password only needs to be at least 6 characters. Sign-up requires ticking a box that confirms the user is an adult.

Sign-up to first chat takes under 2 minutes. I'd call that the fastest part of the product, and the one place where "no friction" is not a sales line. Nobody has to click through pictures of traffic lights, and I won't pretend to miss them.

A 6-character password floor is low for an account that will hold private chats and a saved card. I'd use a password manager and a long random string here, and I'd not reuse anything. The Google button is quicker, but it ties the account to a mailbox that probably holds far more of your life than a companion app needs to see.

The verification e-mail

A verification e-mail titled "Welcome to Dondi!" arrives within about 1 minute, sent from [email protected]. It uses a Firebase template signed "Your project-611394017529 team".

The link works, but verification is not required to use the app. I'd click it anyway. An unverified address with a typo goes unnoticed until the day a receipt or a cancellation reply needs to reach you, and that is the worst day to find out.

That signature is my favourite detail of the whole sign-up. It is the equivalent of a wedding invitation that still says "INSERT NAMES HERE" in the footer. Nobody at Dondi changed the default sender name, which tells you something about how much attention goes to the parts of the product that do not sell tokens.

Two quick questions

After sign-up, a modal titled "Tell us two quick things" asks for a name and a gender. The gender choices are Female, Male and Trans. The username is set automatically to the e-mail prefix.

That last one deserves a warning. If the e-mail address contains a real name, the username now does too, unless it is changed in Settings. I'd sign up with an address that says nothing about you, or change the username before chatting with anyone.

The menu after the first chat

After the first chat the menu shows Browse, Movies, Chat, Studio, Create, Community and Updates. It is a clean layout, with the discovery items tucked under their own heading. Movies sits second from the top, which says a lot about where the company expects the money to come from.

My overall read on the front door: quick, tidy and a little careless. The flow gets a reader from the landing page into a chat faster than most competitors, and it collects less than some. The loose ends, from the Firebase signature to the auto-filled username, are small, but they are the kind of continuity errors I'd catch in a first cut and fix before the client ever saw it.

That match row also offers the filters Mature, 18-19, Asian, Latina, Ebony, Petite and Curvy. It is a catalogue organised by type, the way a stock-footage site organises B-roll. I'd have liked filters for personality or setting instead; a list of body and ethnicity labels says more about the marketing than about the conversations waiting behind the cards.

How big the library is

The landing page claims "200+ AI Models". There is no reason to doubt the count, but the number matters less than the depth. A catalogue of two hundred faces is only as good as the few you would actually talk to twice.

A character page up close

A typical character page, for Sofia Thompson, 23, a "Private Chef", has 3 photos. It has a long bio and a list of traits. The bios are the best-written text in the app, and I'd read one before starting a chat, because they set up the tone better than the first reply does.

Three photos is thin for a product that sells its visuals so hard. Anyone choosing a companion by looks will run out of material before the choice is made, and the next photo is a paid one.

Community

The app also has a "Community" in-app feed. It is a browsing channel of its own, and it doubles as an advert for the paid image tools, since every post in it shows what somebody else generated.

My take on the library: it is well stocked and well labelled, and it is organised for shopping rather than for finding a good conversation partner. That fits the rest of the product, which is built around the catalogue and the checkout more than around the chat.

That is still true today. Custom character creation is a Premium perk. The chat header even reminds free users of it: "Unlock premium to create your own Sofia".

It is a banner that turns the character you are talking to into an advert for the one you are not allowed to make.

What the Create button does

Create leads to a page at /onboarding, which is a matching wizard rather than a builder. Its first screen reads "Step 1 of 4 What's your type? Petite / Athletic / Curvy / Thicc".

I think the wording tells you what the wizard is for. A character builder asks about personality, backstory and voice; a matching wizard asks what you want to look at and then points you at the catalogue. Neither is wrong, but only one of them is creation in the sense most people mean.

What the marketing promises

The homepage says "Browse companions or build one from the ground up". The site footer lists "Create Character". Both lines are true in the narrow sense that Premium users can build companions, and both leave out the price tag attached to the word "build".

My advice is simple. If making your own character is the reason you are looking at Dondi, budget for Premium from day one and treat the free account as a look around the showroom. If you want to write a character for free, other apps in the comparison table below let you do that, and I'd start there.

My opinion of the creation side is mild disappointment rather than anger. It is not broken; it is just priced like a premium extra in a category where it is usually the starter kit. I'd rate it the least generous part of the free experience, and that is a crowded field on this app.

The five-star comparison is not only a joke at my expense. In my trade the review is what you chase after the work is done; here the counter arrived first, and I was chasing a review for a job nobody had booked.

That motivation is the cleverest thing about the app. I knew full well the bar was a number the app invented, and I still sent the extra message to see it move. It is the same itch as a client portal that counts the revisions used: nobody asked for the counter, and everybody watches it.

I think the design is honest about what it is. It rewards showing up and sending messages, not good conversations, and it never pretends otherwise. The trouble is that the reward system is more consistent than the character it is attached to, which becomes obvious in the memory section.

What the replies are like

A short check of 3 messages ran on 23 September 2026. Replies came in about 5-10 seconds. They are short, with italic action lines.

Five to ten seconds is a fair wait: slower than the quickest chat apps, and quick enough that nobody puts the phone down between turns.

The italic lines do most of the scene-setting. They read like stage directions in a cheap script, a smile here and a head tilt there. Useful once, wallpaper after a dozen replies.

A sample reply reads: "Long walk by the river sounds nice. Ive been testing a new pasta recipe in the kitchen — flour everywhere, but the smell is worth it. You hungry?"

That is a competent reply. It picks up the topic, adds a detail from the character's job and hands the conversation back with a question.

It is also a template. Acknowledge, add a hobby detail, ask a question: the shape repeats, like a corporate promo that uses the same three transitions in every cut. After a while you can see the edit points, and once you see them you stop hearing the character.

Quick replies

The chat offers quick-reply suggestion chips. They are handy on a phone browser and a little sad everywhere else. When the app offers to write the user's half of the conversation too, it is fair to wonder who the conversation is for.

I'd leave them alone. Every message costs a token on top of the daily count, as the cost section explains, so a chip that fires off a generic line is spending money on the app's words rather than yours.

For quick flirty exchanges between other things, it works. For a long story or a character with real texture, I think the conversation engine is the weakest link in a product that spends its effort on the frame around it.

The game layer is what kept me coming back, not the replies. That is a strange thing to say about a companion app, and it is the most honest summary I can give of the conversations.

The unique moment came later. I hit the "Friend" milestone and got a small celebratory animation, and five seconds later she asked me who I was.

XP says friends, memory says strangers. I laughed, then felt slightly foolish for having waited for that milestone at all.

What the app offers today

The character page has a "MEMORIES - What Sofia knows" panel. The homepage says "she remembers your stories, picks up on your preferences". A visible memory panel is a good idea, because it lets the user see what has been stored instead of guessing.

In a short check on 23 September 2026 the bot was asked "do you remember where I walked today?" and answered correctly. Its reply: "You said by the river — but you didnt say which one. Was it the one with the old stone bridge, or the path that winds past the willow trees?"

That is a better result than my bakery moment, and it deserves credit.

What others found

The outside picture is split. Reviewers at intimeros.com found memory recall inconsistent. Reviewers at aicompanionatlas.com found recall correct in a short structured test, 3 of 3.

I think both findings can be true at once. A question about something said a minute ago is the easy case; a detail the character raised herself, two turns back, while the app is busy handing out XP, is the hard one. A short test measures short memory, and the failure in my own story was also short, which makes it more annoying rather than less.

My read

The Memories panel and the correct river answer suggest the plumbing is there. What I'd want before trusting it is consistency over weeks, not over three messages. Until then I'd treat "remembers everything" as a slogan and keep important details short, repeated and in the chat itself.

That is an editor's curse. A client would never spot three frames of fused fingers, and I could not unsee them. Still, it was the one moment in the app that looked like the future rather than a chat window with a price list, and I understand why Dondi leads with it.

What the character page offers

The character page has a voice preview, "Female Voice 1", and a Media Studio button. The site footer lists "Generate Image" and has an "AI Voice Chat" section. Whether that voice section means live calls or voice messages, and on which tier, is not stated on public pages.

That vagueness is a small thing, but it is the kind that costs money. I'd assume voice means pre-recorded replies with a play button, not a phone call, until a pricing page says otherwise.

Studio and the Community feed

Studio lets you pick a companion and describe a scene to get new photos or videos of them. A Community feed shows users' shared generations with view and share counts. The homepage sells it as "Unlock custom photos and video moments tailored specifically to you".

In chat, an offered photo arrives as a locked "Premium photo / Unlock this photo" card. It is the vending machine with the snack visible behind the glass.

Studio is the right idea, a scene described in plain words and turned into a picture, and the Community feed turns each result into a shop window. I'd keep in mind that anything shared there is seen by strangers. I think this is the part of Dondi that actually earns money, and the chat exists partly to set up these moments.

Movies

The unique feature today is "Movies": short episodic AI video series with episodes of 1-2 minutes, unlocked with Premium or Premium+. Movies holds 4+ series, including "The House on Wren Lane", "Ready Player Go" and "Blue Hour". Each series has 3-4 episodes.

Premium unlocks 1-2 episodes per series, and the rest need Premium+. That is the classic streaming trick: the pilot is included and the season finale is on another tier. I'd call Movies an interesting experiment and an odd fit for a companion app, like a wedding film company that suddenly sells a mini-series about someone else's couple.

My verdict on media: the video tools are the most impressive thing Dondi makes, and the most expensive. For a short clip, the quality is decent and the artefacts are the usual ones, hands first. For value, see the cost section before getting attached.

The numbers today

Free chat allows 5 free messages a day, shown by a counter such as "5 free messages left today". After a few messages the counter reads "2 free messages left today / Upgrade now". A new account starts with a balance of 50 tokens.

Settings shows that starting grant as "TOTAL PURCHASED 50". Nobody purchased anything, so the label is wrong, and I think it matters: a ledger that calls a gift a purchase is either careless or preparing the ground for something.

A secondary source, aicompanionatlas.com, dated 2026-08-21, reported an initial 45-token balance, flagged by that source as unconfirmed. The 50 in the account settings is the current figure.

The paywall itself was not rude, just abrupt. One minute I was getting to know a character, the next I was reading a price list, and her surname was still a mystery.

What the free side actually includes

The Free vs Premium table lists uncensored chat and image generation as Limited on free and Unlimited on Premium. Videos and voice messages are not included on free at all. Monthly tokens are none on free and 200 on Premium.

The word "Limited" is doing a lot of work there. The Studio page says "Image generation requires Basic or Premium", so free users cannot actually generate images. "Limited" in this table means "none, but we would rather not say so".

What the marketing says

The homepage says registration is free with no payment details: "Register for free ... Zero payment details required". It also says "Chat with 200+ AI Girls for free".

The landing page goes further: "unlimited messaging, photos, voice, and video are all included from day one. No throttling. No message caps".

What others found

Reviewers at intimeros.com found the free tier extremely limited, with premium upsells early in conversations. That matches my own first evening, and the numbers above explain why. Five messages is less a free tier than a trailer.

There is also a twist that the counter hides. The daily count says the messages are free, while the token balance still drops with each one, so the starting grant shrinks even on free days. I'd call that a hybrid meter with a friendly face, and the cost section sets out the numbers.

Who free is good for

Free is good for exactly one thing: seeing whether the interface and the characters appeal before paying. It is not enough to judge the memory, the conversation depth or the media, because none of those show up in five messages. I'd use the free account to browse the gallery, read a few bios and send the five messages to the character you would pay for, and then decide.

It is also a shame for Dondi itself. The relationship bar is the best thing the app makes, and five messages a day barely move it, so free users leave without seeing the feature that sets the app apart.

Paying monthly also meant I could walk away as soon as the app stopped being worth it, and that is exactly how it went.

The plans on 23 September 2026

Prices are in USD on app.dondi.ai/pricing, framed as "Premium 70% OFF". The Monthly plan costs $14.99 a month, which the page breaks down as $0.50 a day. That is the same price I paid in January 2026, so at least the monthly tier has not moved.

The 3 Months plan costs $35.97 every 3 months, or $0.40 a day, and is labelled "Save 40%". That works out to $11.99 a month.

The Yearly plan costs $71.88 a year, or $0.20 a day, labelled "Save 70%" and "+500 bonus tokens". That works out to $5.99 a month.

The yearly price is also the same $71.88 I declined in January. The pricing page spells out the catch plainly: "You'll be charged $71.88 today for 12 months." Payment is by card or Apple Pay.

What the discount really is

The "70% OFF" framing needs a note. The yearly plan's $5.99 a month is presented as 70% off, and the 3 Months plan's $11.99 as 40% off, both measured against the monthly price. That is a discount only in the sense that a bulk pack of batteries is a discount: you save per unit by paying more up front.

Card or Apple Pay is a sensible pair of options. I'd lean towards Apple Pay here, since it keeps the card number away from a merchant whose name changes from page to page.

What Premium adds

Premium lets you create unlimited AI companions. It promises unlimited photos and videos, marketed as adults-only image and video generation. It promises unlimited uncensored chat, voice messages and priority access.

Premium+

A higher "Premium+" tier is referenced on the Movies page, with one movie a month. No price is shown for Premium+. I'd never subscribe to a tier whose price only appears at checkout, and I would not trust a tier whose only visible benefit is one movie a month.

Is the step up worth it

Here is the money maths as I see it. The monthly plan at $14.99 is the only one I'd consider, because it caps the damage at one month if the app disappoints. The 3 Months plan at $11.99 a month saves a little, but it only makes sense if you already know you will stay.

The yearly plan at $5.99 a month looks cheap, and it is cheap per month. It is also $71.88 paid up front to a company whose name and jurisdiction change depending on which page you read, and whose refund window, as the terms section explains, is far shorter than the pricing page suggests. I'd say no to that trade.

My bottom line: Dondi's monthly price is ordinary for the category and its yearly price is aggressive. What decides the value is not the plan but the token meter, which is where the next section goes.

What things cost today

The model is hybrid: daily free messages plus a per-message token cost. Each chat message also deducts 1 token. The credit log records each one as "Chat message with Sofia Thompson -1".

In a short check the balance went from 50 to 47 tokens after 3 messages.

Playing a reply as voice costs 1 token, and the button says so: "Play · 1 token". A voice message costs 1 token, an image 5 tokens and a 5-second video 50 tokens. A Help Center article on tokens confirms 200 tokens a month on Premium, the 500-token yearly bonus, images from 5 tokens and video at 50.

Those numbers make the "unlimited" wording on the pricing page hard to take seriously. With 200 tokens, a Premium month buys four 5-second videos and nothing else, or a mix of messages, voice plays and a few images. Terms 18.4 says it plainly: "Some features of the Services may require tokens or virtual credits."

The per-message cost is the part that changes behaviour. When every line costs a token, longer messages start to feel like better value, and the conversation turns into correspondence. I think that is the opposite of what a companion chat should feel like.

A voice replay priced the same as a new message is the detail that bothers me most. A replay is not new work; it is the same clip played twice.

Token packs

The smallest token pack costs $5.99 for 150 tokens, one-time. The middle pack costs $15.99 for 500 tokens and is marked best value. The largest pack on the page costs $39.99 for 1500 tokens.

I'd not go near the big packs. At that level you are not paying for a companion; you are prepaying a meter.

Keeping track

The one good thing about the token system is that it is visible. Settings > Credits shows a detailed ledger per message. I respect that; plenty of apps in this category hide the meter and let the bank statement tell the story.

Then and now

The animated portrait in my story, in January 2026, came out as a two-second clip, and today a video is priced in 5-second units at 50 tokens. So either the clip length has grown since January or the unit was longer than the result, and I can't tell which from a single clip.

My advice on cost: read the Credits page after every session in the first week. It is the fastest way to see where a Premium month actually goes, and I'd much rather learn that from a ledger than from a surprise top-up.

I think of these timers the way I think of a car lot's "today only" balloons. They are part of the furniture, and the price under them is the normal price with a sense of urgency added. The friction section explains why these particular timers should be ignored.

What is real

The one bonus with substance is on the yearly plan. It adds "+500 bonus tokens free gift" in the first month, after the yearly payment succeeds. That is a real extra, and it is also bait for the most expensive commitment on the page, so I'd weigh it against the risk of paying for twelve months up front.

What is missing

There is no promo-code field. There is no daily check-in bonus. No referral or affiliate programme is visible.

That makes coupon hunting for Dondi pointless. Any site advertising a Dondi discount code is offering a code for a box that does not exist, and I'd close that tab.

By the end of the session, the only e-mail received is the verification mail, with no coupon. Some apps follow up a new sign-up with a discount a day later; Dondi does not, at least not within the first session. I think that is a small point in its favour, even if it leaves no deal to recommend.

My take

The honest summary is that the 70% banner is the price, not a discount on it. If you want the cheapest monthly rate, the yearly plan gives it, along with the 500 bonus tokens and all the risk of an annual commitment. If you want flexibility, pay the monthly price and ignore every timer on the screen.

That is a clear position, and I prefer clear to coy. A user knows what the app is from the landing page onward, and nobody wanders in expecting a study buddy.

What a user can control

Settings has no toggle for adult content. The only chat settings are "Reply length", a Premium option, and push notifications. In practice that means the user controls very little: there is no way to keep the chat on the tamer side short of steering it by hand.

Age rules

The company may ask for proof-of-age documents at any time. That clause sits in the terms and is not part of sign-up, where a single checkbox does the job. I'd take the clause as a warning that an account could be frozen pending documents, and I'd not keep a large token balance on an app that reserves that right.

Moderation notes

One gallery filter is labelled "18-19". Those ages are legal, and it is still a choice I'd rather a platform did not make, because it sells the edge of the legal line as a category. Staff Picks and Trending would do the same job without the label.

A Premium+ series in Movies uses bedtime-story wording in its title and in all three episode names; its content was not opened. In my view that naming is a lapse in judgement on an app of this kind, whatever the episodes contain. Titles with openly adult wording are also visible in the Movies list to anyone who opens it.

My read

The settings are sparse and the defaults are fixed, which suits people who want exactly what the landing page sells and nobody else. The catalogue labels above are the part I'd most want the company to clean up. They do not break any rule I can point to, but they make the product harder to recommend.

Who runs it

The app footer reads "POWERED BY DONDI.AI LABS INC.". The footer gives no address and no jurisdiction. The Terms on dondi.ai name a different operator: Techcommerce Unlimited, in Tustin, California.

The record lists the company country as the United States. The footer entity, Dondi.ai Labs Inc., differs from the operator named in the Terms, Techcommerce Unlimited.

So in September 2026 the question I could not answer in January has two answers instead of none. I don't find that much more reassuring.

My advice is to write both names down before paying. If a charge ever needs disputing, the bank will ask which company you dealt with, and "the one in the footer or the one in the Terms" is not a strong opening.

Where the documents are

The Terms are at app.dondi.ai/legal/terms, revised 08 January 2026. The privacy policy is at app.dondi.ai/legal/privacy. The Privacy Notice was last revised on 8 January 2026.

Both documents exist and both are dated, which is more than some apps manage. They were revised in the same month as my subscription, which may explain why I could not find the company details then; I can't say which version I read.

The same dates still stand in September 2026, so neither document has been revised since. I'd still read both before paying, because a company that names itself two ways may also update one page without the other.

Age and exemption

The Terms set the minimum age: "You must be at least eighteen (18) years old or the age of majority in your jurisdiction". They also claim an 18 U.S.C. §2257 exemption because all content is AI-generated. That exemption claim is a legal position, not a quality mark, and it tells a reader nothing about how the content is moderated.

There are no pro-rated refunds for unused time. Terms 18.3 allows them only where the law requires them. Asked to bet on which promise the support team will honour, I'd bet on the one written by the lawyers.

Tokens

Tokens have no monetary value and are non-refundable. They are also non-transferable. Tokens can be refunded within 24 hours only if unused. The Terms say "unused tokens expire at the end of the applicable subscription period".

That last clause matters more than it looks. It means a leftover balance is worth nothing once the period ends, which is how I'd read my own unspent tokens from January.

Chargebacks

A chargeback cancels the subscription and forfeits the tokens. That is standard, and I'd still only use one as a last resort, after a written cancellation request.

Data and training

The Terms allow AI training on user content: "Your Content may be used to train, improve, evolve, and enhance AI systems ... internally or with affiliated technology partners." The privacy policy lists AI training as a processing purpose.

It adds: "Your Content and interaction data may be processed using AI systems ... This processing may include: AI model training and improvement".

The privacy policy says personal data is not sold: "We do not sell your Personal Data". It is shared with service providers. Personal data is kept only as long as needed.

In plain terms, your chats may train models, possibly with partner companies, and there is no opt-out mentioned. I'd write every message on the assumption that someone other than the character may read it one day.

Ownership and disputes

On ownership the Terms say "You do not receive exclusive rights to any Output." So an image you generate and pay tokens for is not exclusively yours. For most users that changes nothing; for anyone hoping to use a generated picture elsewhere, it means the company keeps its own rights to the same output.

That combination is my biggest reservation about Dondi. Two company names, no named jurisdiction and a refund rule that contradicts the sales page add up to a set of terms I would not want to test in a dispute.

The timers

The countdown resets, so the urgency is fake. The fake-urgency countdowns come with "only 18 spots left" messages. A flea-market table with a "last one" sign and a box of spares underneath it is more convincing.

This is my biggest single complaint about the product's manners. A timer that restarts is not a sale; it is a nudge dressed as a deadline, and I think it quietly insults the user.

The free promise

The landing page claims unlimited free usage. That claim is not true for the free tier, which has 5 messages a day and no image generation. The gap between the landing page and the app is wide enough to drive a truck through.

The chat itself

Replies drop apostrophes, as in "Youre" and "didnt". It is a small thing, and it is the sort of thing a good subtitle editor fixes before anything else, because it makes a character look careless. Once you notice it, you see it in every reply.

After the 3rd message, the bot pushed an upsell: "Want me to send you a photo?" The offer arrives as a locked card, so the conversation turns into a sales pitch before it has turned into anything else. I'd call that the most revealing moment of the chat: the character's first initiative is a product placement.

The onboarding modal

The onboarding modal kept coming back until a gender was picked. It is a small nag, and it tells you the app treats that field as required even though it presents it as a friendly question.

Each of these is survivable alone. Together they form a pattern I'd call "easy in, slow out", which is the pattern I trust least in any subscription, from gyms to streaming apps. My advice is to budget the friction into the price: if leaving takes an e-mail and two days, act two days early.

Reviewers at intimeros.com say it works best for quick fantasy interaction. They list limited free messaging, frequent upsells and repetitive roleplay as the drawbacks. I agree with every word of that summary.

Who should skip it

It is not for people who want substantial free chat, since free means 5 messages a day with tokens deducted too. It is not for users who want only safe-for-work companionship.

It is not for users who need clear company and jurisdiction information. It is not for users who need a real 7-day refund. If any of those four matter to you, I'd look elsewhere before spending a cent.

My own verdict fits neatly between the two lists. I'd suggest Dondi to someone who wants a short, well-dressed experiment and reads the terms first; I'd steer anyone looking for a long-term companion to an app with better memory and plainer paperwork.

The billing descriptor is "Dondi". That is short and recognisable, which I appreciate; plenty of companion apps bill under a name you would never connect to the app. Anyone scanning a bank statement will know what this line is.

How cancelling works

Terms 18.3 on dondi.ai says "You may cancel your subscription at any time through your account settings."

The detail is less instant than that sentence suggests. A subscription can be cancelled through an account-settings option "which directs your request to our support team." It can also be cancelled by e-mailing [email protected].

The Terms promise a reply to cancellation requests within 24 to 48 hours. After cancelling, access continues to the end of the billing period. Settings > Subscription shows the plan and the billing activity.

What I'd do before paying

I'd set a calendar reminder for three days before the renewal date, not one. With a reply window of up to 48 hours, a request sent the evening before renewal may not be processed in time. I'd send the request both ways, through settings and by e-mail, and keep a screenshot of each.

I'd also check Settings > Subscription after the reply arrives. If the plan still shows as renewing, I'd write again the same day. Keep the e-mail thread; if a charge still appears, that thread is your evidence.

Deleting the account

A "Delete Account" button sits in Settings > Account. That is a proper self-serve option, and it is more direct than the cancellation route. I'd still cancel the subscription first and confirm it, then delete the account, because deleting an account with an active plan is the sort of shortcut that creates a billing mystery later.

The second is to pay monthly. The yearly deal is cheaper per month, but it locks money into a company with two names and a 24-hour refund rule, and a month is enough to know whether the app works for you.

Where to get help

The Help Center is at dondi.ai/help-center and has 24 articles. The token article there is worth reading before buying anything, because it lists the per-item costs the pricing page leaves out.

Telegram at t.me/dondiai is the only social link, on the site and in the app sidebar. No Discord, X, Instagram or app-store links were found. That makes Telegram the only public place to see other users talk about the app, and I'd not expect independent opinions in a company-run channel.

The app has a Feature Requests page. If something annoys you, that is the place to say so. I'd put in a request for an adult-content toggle and for a real self-serve cancel button; those are the two changes that would move my score the most.

Watch the name

An unrelated App Store app also called "Dondi", id 6498315021, by DONDITECH SAS at dondiapp.com, is a Spanish-language loyalty-points app, not this product. Dondi AI has no mobile app of its own. Anyone installing something called Dondi from an app store will get a Spanish-language loyalty-points app, not a companion.

Conversation quality lands in the middle. The replies are quick and on topic, but they follow a visible template, and the game layer is more engaging than the talk. Memory scores low because the forgetting in my story was fast and personal, even if a short check today came out better.

The free tier scores poorly: a handful of messages, no image generation and a landing page that promises the opposite. Value for money is weak too, because the token meter charges for almost everything the plans call unlimited.

Images and voice are the strongest row. The video tools impress, even with melting fingers, and Movies is an original idea, though both cost real tokens. Privacy and terms score lowest: two company names, no jurisdiction, training on chats and a refund promise the Terms contradict.

Cancelling is workable and slow, with a clear billing name. Content settings are simple to describe and hard to adjust.

Where Dondi is worse

On the basics, it trails. The free tier is thinner than most, character creation costs money where rivals often give it away, and the memory in my own story was weaker than the XP bar suggested. The terms are also among the least clear of any app I'd consider, with two operator names and no country for disputes.

The table below sets out when and for how long each app was in use, with its cheapest plan and its free offer. I'd read Dondi's row next to the apps with stronger memory before deciding.

My ranking instinct

I'd put Dondi in the "good frame, weak picture" group. If you want media and a gamified feel, it is worth a month. If you want the best conversations or the plainest paperwork, the table below has better candidates.

More on this site: the best AI girlfriend apps, free AI chat apps, the best AI companion apps, AI roleplay chat apps, all AI companion app reviews.

I don't regret the month. I got a clip with melting fingers, a lesson in rationing, and a frozen loyalty card with a few tokens still on it. I'd just not give it a second one, and I'd tell anyone curious to go in monthly, with a spare e-mail and their eyes on the meter.