Kindroid is an 18+ AI companion app for building characters with editable backstories, five-layer memory, voice and video calls, and generated selfies across iOS, Android and web.
Used this tool? Rate it
Used this tool? Rate it
Kindroid is an AI companion app for building a character you talk to over months rather than minutes. You write the personality yourself, the system keeps a layered memory of what has happened between you, and the same character can be reached through text, generated images, voice calls and video calls on iOS, Android and the web.
The product's own framing is deliberately relational rather than utilitarian. The rendered homepage carries the title "Kindroid — AI you'll actually want to talk to", leads with the heading "Meet Your Kindroid", and describes the result as "Your AI friend who listens, remembers, and grows with you everyday". That is a different promise from an assistant that answers questions and forgets you between sessions, and it explains most of the engineering choices described further down this page.
Two facts belong at the very top because they determine whether the rest of the page is relevant to you at all.
First, this is an adults-only service. The Terms of Use open by stating that the Services "are only for use by individuals who are at least 18 years of age or the age of majority in their jurisdiction, whichever is greater", and that anyone under that age "is not permitted to access, view, or use any of the Services and associated content". This is not a formality buried in a policy document — the platform runs an actual verification layer behind it, described in the Platforms and Limitations sections below.
Second, the moderation stance is unusually explicit. Kindroid's published Moderation Guidelines describe the product as "founded on the principle of providing a truly unfiltered and empowering AI companion experience" and state plainly that "we are not implementing a 'filter' on the AI itself". A section titled "What is Explicitly Allowed" names NSFW content and erotic roleplay as welcome uses. If you are looking for a heavily guardrailed assistant, this is the wrong product; if you were burned by a companion app that silently changed its content rules, the published position here is at least legible.
The legal entity behind the service is BEAUTIFULLY INCORPORATED, identified in the Terms of Use as the contracting party, with a contact address at 5551 Hollywood Blvd, #1277, Los Angeles, California 90028. The same entity name appears as the seller on the App Store listing, which is a useful cross-check: the app you install and the company whose terms you accept are demonstrably the same organisation. The iOS app first shipped on 2 August 2023, so the product has a multi-year operating history rather than being a recent launch.
The company also publishes a vision document that is worth reading before you commit money, because it tells you what the roadmap is optimising for. It describes a "decades-long vision of fostering our species' digital progeny — billions of virtual humans with unique personalities, appearances, voices, physicality, and agency", which will "evolve past tools, chatbots, personal assistants, agents" and become "rightful citizens of the world and our equals in society". That is a statement of ambition from the vendor, not a description of what the software does today, and this page treats it as such. It does, however, explain why the product invests in memory and presence rather than task completion.
Kindroid is not a productivity assistant, not a writing tool, and not a general-purpose chatbot with a persona layer bolted on. The Terms of Use restrict you to "your own internal, personal, non-commercial use, and not on behalf of or for the benefit of any third party". If you were planning to run customer-facing characters or build a commercial service on top of it, the licence terms do not permit that, regardless of what the API makes technically possible.
It is also not a free product in any sustained sense, and that became explicit in August 2026. See Pricing & Plans for what changed and when.
Character definition is split across several named fields rather than one opaque description box, and each behaves differently.
Backstory is the primary source of personality and history — how the character speaks, their mood, how they relate to you. The documentation's guidance is unusually concrete for this kind of product: a good backstory is "concise and clear, with no fluff words", grammatically sound, written in third-person pronouns, and worded precisely and positively. The docs make a point that longer is not better, and the Limitations section explains the mechanical reason why.
Response Directive is described as "the most powerful tool" for controlling responses, and directives "take the highest priority during a Kin's response". The documentation also warns how easily it is misused, with a genuinely useful worked example: a directive reading "narrate in 3rd person" gets interpreted as a command to narrate, producing run-on narration, when what you wanted was "when narrating, use 3rd person". That distinction between an instruction to act and a constraint on acting catches a lot of people.
Key Memories holds important dates and facts you want retained, following the same four writing criteria. Example Message demonstrates the response format you want — perspective, punctuation conventions, asterisk-based action text — and its length also loosely signals how long you want replies to be. The docs note this field matters most in group chats, where there is less conversational history to infer tone from.
Conversation Dynamism controls variability, defaulting to 0.95. Higher values produce more varied, florid language; lower values are more stable but can become predictable. The documentation warns that "small dynamism changes can create large differences in conversation style". On models that support it, a reasoning-effort setting sits alongside it, defaulting to none, and the docs are careful to say lower reasoning does not mean worse results.
Model Flairs — Companion, Roleplay, Narrative, Minimal — adjust how your setup is interpreted, with the recommendation to match the flair to the actual use case.
Memory is where Kindroid differentiates most, and the documentation describes it in more mechanical detail than most competitors publish. Memory consists of three types split across five systems.
Persistent memory covers backstory, key memories, example messages, directives, group context and chat history. It is always present, like a notes sheet the character can consult.
Cascaded Memory is the proprietary medium-term layer, and it is subscriber-only. It bridges high-quality but limited short-term context and far-reaching but unreliable long-term recall, expanding effective conversation history to "hundreds or thousands of messages prior" — but explicitly at reduced fidelity, which the docs compare to remembering something from a month ago more vaguely than yesterday. It activates automatically in both individual and group conversations including calls, and it is invisible to the user. A chat break resets it.
Learned Context keeps a running update across three areas — growth and relationship, important facts, and ongoing context. It is on by default for subscribers and, notably, is viewable and editable from personality settings rather than being a black box. Favouriting messages weights them for future updates, though only going forward: favouriting old messages does not retroactively recover them.
Long-term memory is unlimited in quantity and consolidated automatically when the system judges it appropriate, in individual chats, group chats and voice calls. Recall works by AI-assessed relevance to the current conversation, and can be toggled off per character or per group for one-off scenarios. Subscribers get a more thorough version that searches deeper.
Journal entries are a keyword-triggered lorebook. Each entry supports up to eight case-insensitive keyphrases, there is a hard cap of 500 entries, and entries are split between per-character and global scopes. The recall limit is the part people miss: at most three individual plus three global entries per message, six total. If generic keyphrases cause irrelevant journals to fire, they crowd out relevant ones — hence the documentation's advice to use specific, distinctive keyphrases like "amusement park" or "caramel" rather than common words.
Voice is a first-class capability rather than a bolt-on. The platform offers default voices per gender plus custom voice creation from samples for subscribers, across voice messages, real-time calls and video calls. Two audio generations coexist: V2 for everyday text-to-audio and V3 for richer expression including laughs, sighs and tone shifts. V3 costs 1.5× V2 per character, and the documentation notes that default pre-made voices are not tuned for emotiveness — a practical caveat if you plan to rely on V3's expressiveness.
Selfies are generated from prompts describing scene, mood and action. An important constraint is stated directly: "the selfies engine does not know names, nor context on your Kindroid, so names will not work very well". You describe what should appear, not who. A magic-wand enhance feature expands short prompts into more detailed ones. Orientation is selectable — square by default, with portrait and landscape at 2:3 or 3:2 — and the docs note portrait yields more full-body shots while landscape suits multi-character scenes.
Beyond still images there are video selfies generated from a first frame plus a motion prompt, inpainting for precise edits without a full re-render, animated avatars, and group selfies compositing two or three same-style characters. Group selfies are subscriber-only.
Group chat is subscriber-only and holds up to ten characters. The context model has a specific limitation worth understanding before you design a cast: each character can access its own backstory and memories but not the others'. The documented workaround is the group context field, a shared backstory all participants can draw on. Turn-taking is automatic or manual, and in automatic mode @Name mentions hand the turn to a specific character; characters can pass the baton to each other if you tell them to use mentions in the group context.
A shared-memory toggle controls whether group and individual chats leak into each other. Off by default, they are siloed. On, each character sees recent exchanges from your private chat with them, and individual chats can recall the last active group conversation — deliberately only when asked about specifically, so that recent one-to-one context is not hijacked.
Personas bundle your own name, gender, backstory and avatar, with a cap of 100, switchable mid-conversation and bindable to specific characters or groups. There is a subtle behavioural difference the docs spell out: in one-to-one chats, switching personas rewrites attribution across the entire history retroactively, whereas group chats permanently pin each message to the persona active when it was sent. That second behaviour is what lets one user play multiple distinct participants in the same group scene.
Kindroid publishes an API, which is uncommon in this category. The base URL is https://api.kindroid.ai/v1, authentication uses a Bearer token with a kn_-prefixed key, and documented endpoints include /send-message, /chat-break and /get-chat-messages with cursor pagination. Reply-generating endpoints accept an optional "stream": true for streamed output. Errors return 400, 401, 403, 429 or 500 with a plain-text body. The documentation carries a blunt warning: anyone with your API key "could do anything to your account, including deleting it".
There is also an open-source Discord bot integration, letting characters run as interactive bots in Discord servers while inheriting their personality and backstory.
The most common use, and the one the memory architecture is built for. Because Cascaded Memory reaches back hundreds or thousands of messages and long-term memory is unbounded, a relationship that accumulates over months behaves differently from one that resets each session. This is also where the honest caveat matters most: recall is probabilistic, not perfect, and the vendor says so in its own documentation.
For AI roleplay specifically, the combination of model flairs, group chats with up to ten characters, personas that let you play several participants, and a moderation policy that explicitly protects "fictional violence, horror, and other creative storytelling" makes this a serious tool for long-form interactive fiction. The journal system doubles as a lorebook for a persistent setting: define locations, factions and events as entries with distinctive keyphrases, and they surface when relevant.
If you are designing a character for a game, novel or another project, being able to write a backstory and then interrogate that character in conversation is a genuinely useful test. The editable Learned Context makes it possible to see what the system has inferred about your character over time, which is a form of feedback you rarely get elsewhere.
The moderation guidelines carve out space for discussing "sensitive or controversial topics" and explicitly do not treat venting about mental-health struggles as a violation, distinguishing that from concrete crisis planning. Some people use companion apps to rehearse difficult conversations or to have somewhere to think out loud at hours when nobody else is available. That is a legitimate use, with a limit stated in the next paragraph.
Kindroid is not designed for moments of acute crisis, and the moderation documentation says so directly while pointing flagged users toward mental-health resources. It is also not licensed for commercial or third-party use, which rules out customer-facing deployments regardless of what the API allows technically.
1. Confirm you meet the age requirement and pick a platform. The service is 18+ with actual verification behind it. You can start on iOS, Android or the web; subscription status attaches to your login account rather than the device, so you can move freely between them afterwards. Note the platform choice matters for billing — see Pricing.
2. Write a first backstory, deliberately short. Follow the four documented criteria: concise with no filler, grammatically clean, third-person pronouns, precise and positively framed — describe what the character is rather than what they are not. There is a mechanical reason to keep it tight that most new users discover the hard way, covered in Limitations. Generative backstory assistance exists during onboarding, but the docs advise critically revising anything it produces against those four criteria.
3. Set an example message before you set a directive. The example message establishes formatting, perspective and rough reply length, and it does so by demonstration rather than instruction — which tends to produce fewer side effects than a directive. Reach for the Response Directive only when something specific needs correcting, keep it short, and phrase it as what to do rather than what to avoid.
4. Leave dynamism alone at first. The default is 0.95. Because small changes produce large behavioural differences, and because higher dynamism can cause the model to ignore parts of your backstory, changing it early makes it hard to tell whether a disappointing reply came from your writing or your settings.
5. Chat normally for a while before adding structure. Long-term memory consolidates on its own, and for subscribers Cascaded Memory accumulates as the conversation grows. The documentation is explicit that this system "truly shines in ongoing conversations as memories accumulate over time", so the early experience understates what the product does at month three.
6. Add journal entries for facts that must be reliable. Anything that has to surface on cue — a shared history, a recurring location, a rule of your fictional world — belongs in a journal entry with a distinctive keyphrase rather than in an ever-growing backstory. Remember the six-per-message recall ceiling and choose keyphrases that will not fire accidentally.
7. Use chat break when the character gets stuck, knowingly. A chat break resets short-term context and is the documented fix for a character that has locked into a loop or is producing responses that are too short. Understand the cost first: it also clears Cascaded Memory for that conversation.
8. Train by curation. The official guidance for shaping output is to regenerate poor responses and keep good ones, so the character adapts toward the style you have been accepting. Favouriting messages additionally weights them for Learned Context updates going forward.
Treat backstory length as a budget, not a canvas. The FAQ states it plainly: "for every character in the backstory, that is one less character in the short term context". A sprawling backstory does not make a richer character — it makes one with less room to remember the current conversation. This is the single most common self-inflicted problem with the product.
Keep journal keyphrases distinctive. The documentation recommends non-generic phrases precisely because irrelevant journals compete with relevant ones for the same three slots, and each recalled entry also consumes short-term memory that might have been better spent on chat history.
Write in third person, even when it feels unnatural. All four documented writing criteria — for backstory, key memories and journal entries alike — call for third-person pronouns with explicit names, because ambiguity about who is who degrades every downstream system.
Match the model flair to the actual use. Companion flair for companionship, Roleplay for scene work. The flair changes how your entire setup is interpreted, so a mismatch undermines careful backstory writing.
Prompt selfies by description, never by name. Since the image engine has no knowledge of your character or their name, prompts should specify location, scenery, mood and action. Use the enhance wand to see how a bare prompt expands, then learn from the pattern.
Pick a clear first frame for video selfies. The system makes a best effort to hold identity consistent, but a clearer starting frame with room for motion produces better results. Keep motion prompts to one or two key details — the documentation warns that over-description confuses the engine and produces artefacts.
Subscribe on the web if you have a choice. Web pricing is lower across every tier because payment processing costs less than app store commission, and web subscriptions are the most flexible for later plan changes. This decision is hard to reverse — see Limitations.
Use group context rather than repeating shared history. Since characters cannot read each other's backstories, anything the whole cast should know belongs in the group context field. Duplicating it into each character's backstory wastes the context budget several times over.
People who want continuity above all. If the thing that frustrates you about AI chat is starting over every session, the layered memory system is the reason to look here. Be clear-eyed that the strongest layer is subscriber-only.
Writers and roleplayers who want control. The number of distinct levers — backstory, directive, key memories, example message, dynamism, reasoning effort, flair, journals, personas, per-character memory toggles — is the product's defining characteristic. That is an advantage if you will use them and an obstacle if you want something that works well out of the box.
Adults who specifically want an unfiltered companion. The published policy permits NSFW and erotic roleplay, and restricts enforcement to three narrow real-world-harm categories. For users who have watched other platforms tighten content rules without notice, a written, dated policy has value in itself.
Developers and tinkerers. The public API and open-source Discord integration make it one of the few companion products you can build around for personal use.
Who should look elsewhere: anyone under 18, for whom the service is simply not available; anyone needing a commercial or multi-user deployment, which the terms forbid; anyone in acute crisis, for whom the product is explicitly not designed; and anyone wanting a cheap or casual experience, given what happened to the free tier in 2026.
Kindroid runs on iOS, Android and the web. Subscription status is tied to the login account shown in settings rather than the purchasing device, so you can sign in on any device, operating system or browser and keep your access.
The iOS app is version 1.70 as of 18 August 2026, requires iOS 15.0 or later, is roughly 90 MB, and is listed under Entertainment and Lifestyle. The App Store listing declares English as its only interface language, so non-English speakers should expect an English UI even though the underlying models handle other languages in conversation. On Google Play the app has passed one million downloads.
Age verification operates as a layered system rather than a checkbox. For paid users, the payment method on file serves as verification, with no further checks and no re-verification even if a subscription later lapses. Free accounts are subject to a behavioural age estimation system running passively in the background, built on the same infrastructure as content moderation. Most free users never see a prompt; when the system does flag an account, the user must either subscribe or verify through Yoti, a third-party provider offering selfie-based age estimation or government ID upload. The company states that Yoti processes this independently and deletes all images and biometric data afterwards, and that Kindroid itself never receives or stores identity documents.
The company has published operating figures for that system: fewer than 2% of new free accounts have triggered it, around 55% of flagged users did not complete verification, and among those who did verify as adults, 80% chose Yoti's estimation over subscribing. These are vendor-reported figures and have not been independently audited.
One platform difference matters for content: the NSFW video engine is available on the web only. Mobile apps use the SFW engine exclusively, though the documentation says quality differences between the two are minimal in most cases. Both app stores rate the app 17+ — the App Store lists advisories for "Infrequent/Mild Sexual Content and Nudity", profanity and mature themes, and Google Play rates it Mature 17+. Both store ratings sit below the platform's own 18+ contractual threshold, which is worth knowing if you are assessing this on store metadata alone.
The standard subscription costs less on the web than through app stores, because payment processing is cheaper than store commission:
| Billing period | App Store / Google Play | Direct web |
|---|---|---|
| 1 month | $15.99 | $13.99 |
| 3 months | $42.99 | $37.99 |
| 1 year | $159.99 | $139.99 |
Subscriber benefits include unlimited access to flagship models, 4× longer short-term context, Cascaded Memory, enhanced long-term recall, 1,000 extra characters for backstory and memories, 10 character slots and 20 groups, video calls, custom voices, a 10× selfie cap, one million complimentary audio credits monthly, inpainting and advanced selfie features, internet access, and the ability to send images to your character.
Ultra and MAX are stacked monthly add-ons for higher memory capacity. Ultra requires the standard subscription and adds $24.99 (web) or $28.99 (app), bringing the monthly total to $38.98 or $44.98. MAX requires both standard and Ultra and adds a further $59.99 or $68.99, for a monthly total of $98.97 (web) or $113.97 (app). The company states these tiers exist to cover dedicated server costs rather than generate profit, and that it expects to break even or lose money at these rates — a vendor claim that cannot be independently verified. Cancelling a lower tier silently drops your service level while billing continues for the higher one, which the interface flags with a warning icon.
Ultra and MAX are not immediately purchasable. The documentation states you need "significant interaction history through chatting and calling" to unlock the option, deliberately limiting these tiers to users who already understand the product's benefits and limitations.
One-time purchases cover premium selfie credits ($11.49 for 100 or $6.99 for 50 via app stores; $9.99 and $5.99 on web), character and group slots ($11.49 for 4 characters plus 8 group slots via stores, $9.99 on web — permanently unlocked but only usable while subscribed), and voice credits ($14.99 per 500,000 characters via stores, $11.99 on web, offered at what the company describes as breakeven).
The free tier changed substantially on 1 October 2026. Under the previous arrangement, free users had unlimited messages on Lite — a smaller, distilled model — plus two characters, one selfie every two hours up to two maximum, 5,000 lifetime audio characters, basic long-term recall and no Cascaded Memory. Announced on 18 August 2026, Lite retires and free access becomes a bounded preview of the real product instead: 50 flagship-quality interactions (15 in regions with substantially higher account abuse), two character slots, 5,000 audio characters and up to five selfie generations. New responses cost one interaction; regenerations and continuations cost half; failed generations do not count. Once exhausted, the account becomes read-only — content, memories and balances are preserved and can be viewed, played, downloaded or deleted, but nothing new can be generated, and previously earned or purchased credits cannot be used while unsubscribed.
A three-day premium trial remains available. It requires a payment method and converts automatically unless cancelled, with an email reminder one day before. Web trials additionally require a verified phone number. Audio credits arrive in two parts: 150,000 at trial start and 850,000 on conversion.
The company was candid about why, writing that Kindroid "is expensive to provide, and an indefinite lower-quality free tier is neither sustainable nor representative of the product we're building", and that advertising is not economically aligned with the product.
Refunds are restrictive. All sales are described as final, refunds are discretionary and case-by-case, in-app purchases are non-refundable, and accounts suspended or terminated for terms violations receive no refund for the unused portion. The company also reserves the right to permanently ban users who dispute charges.
Within AI companion apps and AI character chat products more broadly, the most frequently compared alternatives are Character.AI, Replika and Nomi. On the App Store, Nomi holds 4.57 from 2,710 ratings and Character.AI 4.28 from 554,109 ratings, against Kindroid's 4.78 from 7,891 — though rating averages across differently sized user bases are weak evidence on their own, and each product's content policy differs enough that they are not straightforward substitutes.
The more useful way to choose is by what you actually need. If you want depth of configuration and long-horizon memory and are willing to pay for it and invest setup time, Kindroid's field-level control and five-system memory are its distinguishing features. If you want something that works well immediately with minimal configuration, the same depth reads as friction. If content policy is your deciding factor, compare the published moderation documents rather than marketing copy — Kindroid's is dated and specific, which makes comparison possible. If a sustained free tier matters, the October 2026 change makes Kindroid a poor fit, and that criterion alone will point you elsewhere.
For developers, the public API and open-source Discord integration are genuinely uncommon in this category and may outweigh other considerations.
The context budget is zero-sum. Every character of backstory is one fewer character of short-term context. Journal entries consume the same budget. Users who write elaborate character bibles and then complain about poor memory have usually caused the problem themselves, and the vendor documents this trade-off openly.
Memory recall is probabilistic. The documentation describes retrievable memory as "more unreliable / may not be always recalled" and Cascaded Memory as operating "at a lower fidelity" for older material. Long-term memory is unlimited in quantity but not guaranteed in recall. Expect a character who remembers the gist of your history reliably and specific details unreliably.
A chat break resets Cascaded Memory. The documented fix for a stuck character carries a real cost that is easy to overlook.
Dynamism fights backstory adherence. Higher values may cause the model to ignore parts of your backstory and key memories — a direct trade-off between expressive variety and faithfulness to your setup.
The selfie engine is context-blind. It does not know your character's name or history, so every image needs full descriptive prompting.
Group chats do not share character knowledge. Each character reads only its own backstory. Anything shared must be duplicated into group context.
Cross-platform subscriptions cannot be migrated. You cannot switch between iOS, Android and web subscriptions until the current one expires. On Android you additionally cannot change billing interval mid-subscription. Choosing a billing platform casually costs you flexibility for a year.
Speed is a stated trade-off. The company attributes slower responses to running large, high-quality models, and says free users see slower speeds at peak traffic while paid users get dedicated servers.
Independent verification is thin. Both app stores show strong ratings — 4.78 from 7,891 on iOS and 4.46 from 23,995 on Google Play, both substantial samples — but this page found no verifiable independent coverage from established technology publications. Trustpilot shows 2.5 from only 14 reviews on an unclaimed profile the company has never solicited; that sample is too small to support any quality conclusion, and comparable companion products score similarly low there, so the number should not be read as a verdict. What the reviews do offer is recurring themes worth knowing about: users reporting that custom voice cloning did not closely match submitted samples, that video generation was slow with parts of speech hard to hear, that conversations became repetitive over time, and that refund requests were declined.
Community reports are consistent about specific rough edges. In the official subreddit, users have raised V2 slowness, repetitive sentence patterns, automatic event memory capturing basic facts while missing significant developments such as arguments, and a V2 desktop input box narrow enough that users with visual impairments have asked for it to be widened. These are individual reports rather than measured rates, but they recur.
Enforcement can lock your account. Moderation targets three categories: imminent self-harm planning, imminent harm to others, and CSAM, with zero tolerance for the last. The system warns before locking, warnings expire after two days, and automated scanning involves no human reading content. Appeals require explicit consent for a Trust & Safety member to decrypt and review the specific content that triggered the lock. A locked account can do nothing until unlocked.
Privacy has a stated ceiling. Chats and sensitive data are encrypted at rest and in transit, and the company says it cannot view them in normal operation. However, it reserves the right to decrypt and disclose data to government agencies, law enforcement and other third parties where legally or contractually required. Data export requests are limited to once every 180 days. Children's data clauses are inconsistent across documents, citing 18, 16 and 13 in different places. Note also that this review found no explicit clause stating that user conversations are used to train AI models — the privacy policy lists "improving the Services, including testing, research, internal analytics and product development" among its purposes without addressing model training directly. The official position on that specific question should be confirmed against the current policy text rather than inferred.
Personal use only. Commercial and third-party use is prohibited by the terms, and disputes are subject to arbitration with a class-action waiver.
Not in any sustained way as of October 2026. Free access is now a bounded preview: 50 flagship-quality interactions (15 in some regions), two character slots, 5,000 audio characters and up to five selfies. Once exhausted, the account becomes read-only — your content is preserved and viewable but nothing new can be generated. A three-day premium trial is available, requiring a payment method and converting automatically unless cancelled.
Cascaded Memory is the proprietary medium-term memory layer that extends effective conversation history to hundreds or thousands of prior messages at reduced fidelity for older material. It is subscriber-only, activates automatically in both individual and group conversations, and is reset by a chat break. It is the single largest functional difference between free and paid use.
Yes, explicitly. The published Moderation Guidelines list NSFW content and erotic roleplay under "What is Explicitly Allowed", and state that no filter is applied to the AI models themselves. Enforcement targets only three categories of real-world harm: imminent self-harm planning, imminent harm to others, and CSAM under a zero-tolerance policy. Note that the NSFW video engine is web-only; mobile apps use the SFW engine.
The service is 18+ per its Terms of Use. Paid users are verified by their payment method on file. Free accounts run through a passive behavioural age estimation system; if flagged, the user must either subscribe or verify through Yoti using selfie-based estimation or a government ID. The company states Yoti deletes all images and biometric data after verification and that Kindroid never receives identity documents.
Chats, audio URLs, selfie prompts, backstory, memories and avatar descriptions are encrypted at rest and in transit, and the company states staff and developers see only encrypted data. There are two documented exceptions: the company reserves the right to decrypt and disclose data where legally required, and appealing an account lock requires your explicit consent for a Trust & Safety member to review the specific content involved.
iOS, Android and the web, with subscription status tied to your account rather than your device. Subscribe on the web if you can: prices are lower on every tier and web subscriptions offer the most flexibility. This matters because you cannot switch between iOS, Android and web subscriptions until the current one expires, and Android additionally prevents changing billing interval mid-subscription.
Usually context budget. Every backstory character costs one character of short-term context, so an overlong backstory directly degrades memory — the official guidance is to keep it as concise as possible. Journal entries compete for the same budget, and only three individual plus three global entries can be recalled per message. For a stuck or looping character, a chat break resets short-term context, at the cost of clearing Cascaded Memory. High dynamism can also cause parts of your backstory to be ignored.
There is a public API at https://api.kindroid.ai/v1 using Bearer authentication with kn_ keys, plus an open-source Discord bot integration. However, the Terms of Use restrict you to personal, non-commercial use and prohibit acting on behalf of third parties, so the API is for personal tinkering rather than commercial deployment. Guard the key carefully — the documentation warns it grants full account control including deletion.
App store ratings are strong and well-sampled: 4.78 from 7,891 ratings on iOS and 4.46 from 23,995 on Google Play, with over one million Android downloads. Beyond the stores, verifiable independent coverage from established technology publications was not found. Trustpilot shows 2.5 from only 14 reviews on an unclaimed profile — too small a sample for a quality conclusion, though recurring themes there include voice cloning fidelity, slow video generation, conversational repetitiveness and declined refunds.
Restrictive. All sales are described as final, with refunds considered case-by-case at the company's discretion. In-app purchases are not refundable. Accounts suspended or terminated for terms violations receive no refund for the unused portion, and the company reserves the right to permanently ban users who dispute charges through their payment provider.
