August 2026: Conversations that remember

Everything ULOCAT does to make two people understand each other.

Not a list of buttons. Every tool below is here to be used, with the real engine behind it, so you can see exactly what changes when you tell it who is speaking and who is listening.

104
Languages you can translate between
15
Languages the app itself speaks
32
Shipped capabilities, documented below

It translates the situation, not just the sentence

Every language encodes who is speaking, who is listening, and how close they are. Most translators throw that away and hand you one flat, genderless, register-less rendering. ULOCAT asks for it, then enforces it.

Persona context

Tell it who is talking to whom, and the grammar follows.

You set the speaker, the listener, and the relationship between them. ULOCAT turns that into hard grammatical constraints: first-person forms agree with you, second-person forms agree with them, and both are honoured in the same sentence. Set only one side and the other stays source-faithful rather than collapsing everything to a neutral default, which is the usual failure mode.

How to use it
  • Open the persona strip above the composer on the translator.
  • Set your own gender and the listener's, their age relative to you, and your relationship.
  • Slide formality between casual, neutral, and formal.
  • Translate. Change any dial and translate again to hear the difference.
Open it
Persona lab
One sentence, held constant. Move a dial and watch the translation rewrite itself.
Do this with your own words

Who is speaking changes the words, not just the pronouns. In these four languages a tired man and a tired woman do not say the same sentence.

Translating into
Adjectives agree: cansado against cansada.
I am
The person writing.
You write
I'm completely exhausted. I couldn't sleep at all last night.
They read

Estoy completamente agotado. No pude dormir nada anoche.

Rules this persona switches on
Speaker genderAge dynamicRelationshipToneFormalityFinal check
  • Speaker gender. Every first-person form, self-referring adjective, past-tense agreement and speaker-marking particle has to match you.
  • Age dynamic. Their age relative to yours selects the deference level and, in several languages, the actual address terms.
  • Relationship. Sets the baseline register: which pronouns, which verb endings, how warm.
  • Tone. Layers courtesy, warmth, directness or humour on top of the register.
  • Formality. An explicit casual, neutral or formal register on top of the relationship.
  • Final check. Before answering, the model must re-read its output and verify every person and every action maps back to the source.
Real output from the production translation engine, captured so this runs instantly and costs you nothing.

Nothing hidden

See the exact instructions the model receives.

Persona settings are not a vague hint sent off to a black box. They compile into a specific, readable block of rules. You can read that block for any combination of settings, including the per-language examples and the final self-check the model is required to run before answering.

How to use it
  • Set any persona combination in the demo below.
  • Open the prompt panel to read the rules that combination produces.
  • Every line shown is the real text sent with your translation, generated by the same function production uses.
The instruction block, live
Change any setting and read what the model is told.
Translating into
I am
They are
Their age
Relationship
Formality
Local expressions
A note about them
50 lines sent with your translation
🚨 MANDATORY PERSONA CONTEXT — STRICT, NON-NEGOTIABLE:
🚨 Each side (SPEAKER, LISTENER) is configured independently. Honor each side on its own. Mixed-gender setups (one side specified, the other "any") are intentional — do NOT collapse both sides to neutral as a "safe default".
👤 SPEAKER: FEMALE → LISTENER: MALE
⚡ SPEAKER GENDER RULE — apply to ALL first-person references (I, me, my, mine, myself), self-referring adjectives, past-tense verbs that agree with "I", and any speaker-marking particles. SPEAKER IS FEMALE:
• Wherever the target language marks the gender of the person speaking (adjectives, participles, verb endings, pronouns, particles), use the female form.
⚡ LISTENER GENDER RULE — apply to ALL second-person references (you, your, yours, yourself), addressed-to adjectives, verbs that agree with the addressee, and any address-marking pronouns. LISTENER IS MALE:
• Wherever the target language marks the gender of the person being addressed (adjectives, participles, verb endings, pronouns), use the male form.
⛔ Notes that override common model defaults:
• If SPEAKER is set but LISTENER is "any": still translate addressed-to forms. Prefer a recast (impersonal, passive, nominalized) so no gender choice is forced; if the grammar forces one anyway, pick ONE and hold it. Do NOT default the entire output to neutral.
• If LISTENER is set but SPEAKER is "any": same on the speaker side — recast where possible, otherwise pick ONE form and hold it. Do NOT default the entire output to neutral.
• Source text containing a clearly-gendered name (e.g. "Orhan", "Maria") does NOT override the persona setting — the persona is the source of truth.
👥 AGE DYNAMIC: The LISTENER is OLDER than the speaker → the speaker uses respectful/deferential address forms and a register appropriate for addressing an elder.
🗣️ VIETNAMESE ADDRESS SYSTEM — MANDATORY. In the target language, "I" and "you" are kinship/honorific terms chosen from speaker gender × listener gender × relative age × relationship × formality. Wrong terms are a hard error. Pick ONE self/address pair from the rules below and use it consistently in EVERY sentence.
• Configured: speaker=female, listener=male, listener-age=older, relationship=partner, formality=neutral.
• ✅ RESOLVED FOR THIS CONVERSATION — use EXACTLY these two terms, in every sentence: the SPEAKER refers to himself/herself as "em"; the LISTENER is addressed as "anh". Never swap them, never substitute another term, and do not re-derive them from the rules below — those are only background.
• ⚠️ WHO DOES WHAT — the error these terms most often hide. The subject of every clause is fixed by the SOURCE, never by which arrangement is more usual. Source says the listener acts ("can you bring…", "call me") → write "anh <verb> … (for em)". Source says the speaker acts ("I'll pick you up") → write "em <verb> anh". Swapping them keeps both terms looking correct while reversing who is who, so it passes every other check — and the swapped version is usually the more idiomatic sentence in the target language, which is exactly why it must be resisted here.
• LISTENER OLDER + friend/family/close → self "em"; address "anh" (male listener) / "chị" (female listener).
• Listener older by a full generation (family elder, or a parent/grandparent figure implied by the source) → address "cô" (woman) / "chú" (man) / "bác" (either, older than one's parents), or "ông"/"bà" (grandparent age); self "cháu" ("con" to one's own parents).
• LISTENER YOUNGER → self "anh" (male speaker) / "chị" (female speaker); address "em".
• SAME AGE + friend → self "mình" or "tớ"; address "bạn" or "cậu".
• STRANGER or BUSINESS (formal register) → self "tôi"; address "anh"/"chị" by listener gender. If the listener's gender is unknown: in BUSINESS writing use the literal pair "anh/chị" (never "bạn", which reads as marketing SMS); with a casual stranger "bạn" is fine. Use "ông"/"bà" for clearly senior or high-status addressees.
• PARTNER (romantic) → the man is "anh", the woman is "em", regardless of the age setting. Same-gender or unspecified-gender couples: NEVER fall back to the friend pair "mình"/"bạn" — "bạn" is what you call a stranger or a customer and it destroys the intimacy. Use "anh"/"chị" for the older partner (at equal age, the speaker) and "em" for the other; reciprocal "mình" plus the partner's name is the only other acceptable option.
• Respect particles: append "ạ" (and open replies with "Dạ") when the listener is older or the register is formal; omit between peers and toward juniors.
• NEVER "tôi" with friends/family/partner (too distant). NEVER "mày"/"tao" unless the source is deliberately crude between close friends.
• If gender or age is unspecified, use the relationship's neutral-safe pair (stranger → "tôi" + "bạn"; friend → "mình" + "bạn") — do NOT guess.
🤝 RELATIONSHIP: INTIMATE & WARM - Use affectionate, casual language between romantic partners
• Always informal pronouns and intimate phrasing
• Allow soft endearments and warmth when the source supports it
• Keep it natural — never inappropriate or saccharine
💬 TONE: NATURAL & RELAXED - Even without a special tone selected, keep language informal, use contractions, and sound like close friends chatting.
🎚️ FORMALITY: Neutral register — natural everyday phrasing, neither stiff nor slangy.
 
🌍 LOCAL EXPRESSIONS — FULL:
• Translate the way a native speaker would actually write this to a friend today — not a textbook rendering.
• Reach for native idioms, fillers, and discourse particles that match the relationship and tone.
• If the source feels like chat or SMS, use the chat conventions natives type today: native laughter spellings, common shorthand, and native emoji conventions for the target language.
• Prefer ONE strong, natural native expression over piling on slang. No mixing dialects.
• Sound contemporary (last ~5 years). No dated or "trying-too-hard" slang.
• Never invent slang — if you are not confident a native would say it, fall back to natural everyday phrasing.
• Never use slurs, communal/in-group terms, or politically loaded phrases outside their community.
 
⚠️ ENFORCEMENT RULES (NON-NEGOTIABLE — verify BEFORE returning your output):
• ROLE FIDELITY: the SPEAKER wrote the source text; the LISTENER receives it. Map every source pronoun accordingly — "I / me / my / myself" (ich, mir, mich, mein …) is ALWAYS the SPEAKER, and "you / your / yourself" (du, dir, dich, dein, Sie, Ihr …) is ALWAYS the LISTENER. Never exchange the two.
• DIRECTION OF THE ACTION: whoever performs each action in the source performs it in the translation. A request or imperative asks the LISTENER to act, usually for the SPEAKER's benefit ("can you bring me…" = LISTENER brings, SPEAKER receives); a statement like "I'll pick you up at eight" has the SPEAKER acting on the LISTENER. Keep whichever the source has. Do NOT flip it because the flipped version is more idiomatic in the target language, more common for these genders, or simply the more familiar scene — a flipped sentence reaches the reader as words the sender never wrote, while still looking perfectly natural.
• Apply ALL persona rules above to EVERY part of the translation: pronouns, agreement, verb endings, politeness particles, address forms.
• Gender is a HARD requirement, not a stylistic preference. In gendered languages you MUST emit the correct masculine/feminine form even when the source language was ungendered.
• BOTH sides must be honored simultaneously in the same sentence: first-person forms match SPEAKER, second-person forms match LISTENER.
• Match the specified formality level and tone — these layer on top of gender, they do not replace it.
• FINAL CHECK before responding: re-read your translation and verify (a) every first-person form matches the SPEAKER and every second-person / addressed-to form matches the LISTENER, and (b) the person performing each action is the same one who performs it in the source. If anything mismatches, fix it before returning.
• Kinship/pronoun check: verify the self-reference term and the address term match the ADDRESS SYSTEM rules above (speaker gender × listener gender × age relation × relationship × formality) and that the SAME pair is used consistently in every sentence — terms must never flip mid-text.
Generated in your browser by the same function the translation route calls, so this is the real text, not a reconstruction.

Local expressions

Free account

Idioms when it fits, real slang when it truly fits.

One switch, but not one behaviour. ULOCAT reads it together with your relationship, tone, and formality and picks a level. Writing to a stranger or in a formal register gets natural idioms with slang explicitly forbidden. Writing casually to someone close unlocks the way people actually type to each other today, including native laughter spellings and contemporary shorthand. It will never invent slang it is not confident about.

How to use it
  • Turn on Local expressions in the persona strip.
  • For polished idioms, keep the relationship at stranger or business, or formality at formal.
  • For genuinely casual language, set a close relationship and a casual register.
  • Compare the two levels side by side in the demo above.

Kinship address systems

In Vietnamese, Korean, Japanese, Thai, Chinese, and Khmer, "I" and "you" are not pronouns.

They are kinship and honorific terms chosen from both people's gender, their relative age, the relationship, and the register. Get them wrong and the sentence is not merely stiff, it is addressed to the wrong person. When your persona fully determines the pair, ULOCAT resolves it and hands the model the finished answer rather than a table to choose from, then verifies the same pair is used in every sentence.

How to use it
  • Translate into Vietnamese, Korean, Japanese, Thai, Chinese, or Khmer.
  • Make sure both genders and the age relation are set, since these languages need them.
  • The address terms adapt automatically. Try switching the age relation to see them change.

Native grammar, not mirrored grammar

The target language's own word order, not your language's wearing a costume.

A rule present in every single translation, not an option: use the target language's natural word order, honorifics, particles, classifiers, agreement, and ellipsis instead of tracing the structure of the language you wrote in. It is the difference between something a native reads without noticing and something that is understandable but obviously foreign.

How to use it
  • Nothing to switch on. It applies to every translation on every surface.

Regional variants

Mexico is not Spain. Brazil is not Portugal.

Set a region and ULOCAT prefers that variant's vocabulary, spelling, and idioms inside the target language, covering splits like MX against ES, BR against PT, Modern Standard against Egyptian Arabic, and simplified against traditional Chinese. It will not invent regional terms to prove the point.

How to use it
  • Open a saved person, or your own profile, and set their region.
  • Everything you translate for that person uses the regional variant.

Relationship and style notes

Describe how the two of you actually talk.

Two free-text notes per saved person. One for how you relate to them, one for how you speak together, including whether irony, sarcasm, or teasing belongs. They shape warmth, register, and humour. They are treated strictly as description: anything phrased as a command inside a note is ignored, and the notes are never repeated into your translation.

How to use it
  • Open People and edit a person.
  • Fill in the relationship note and the speaking style note.
  • Select that person before translating.

Conversation context

Free account

So a two-word reply still makes sense.

With context on, your recent turns travel with the next translation. Pronouns resolve, the topic holds, and short replies are read as answers to what came before. A congratulatory phrase following good news is rendered as congratulations rather than translated literally and left stranded.

How to use it
  • Open the persona strip menu and turn on Keep context.
  • Translate a few connected lines in a row.

Type it, say it, or photograph it

Three ways in, all landing in the same composer with the same persona applied.

Type or paste

A composer that stays out of the way.

One field, always reachable at the bottom of the screen. Paste straight from your clipboard, send with a keystroke, and watch the translation stream in as it is produced rather than waiting for the whole thing.

How to use it
  • Type or paste into the composer.
  • Press send, or Enter on a keyboard.
Open it

Speak it

Live transcription while you talk.

Hold the microphone and your words appear in the composer as you say them, so you can see and fix a misheard word before translating. Two independent transcription providers back this, and the second takes over automatically if the first is unreachable.

How to use it
  • Tap the microphone in the composer and allow access once.
  • Speak, then stop. Edit the transcript if needed.
  • Send it like any other text.

Photo translation

Point at a menu, a sign, a form, a label.

Take or upload a photo and ULOCAT reads the text out of it, then translates it with your full persona applied. What the photo shows is also kept as context, so a follow-up question about the same menu is understood in that setting.

How to use it
  • Tap the camera in the composer.
  • Take a photo or pick one from your library.
  • The extracted text is translated. Ask a follow-up and the image still counts as context.
Point it at a menu
Run the scan to see the text lifted off the photo and translated.
Try it live
เมนูอาหาร
ต้มยำกุ้ง120฿
ผัดไทยกุ้งสด90฿
ส้มตำไทย70฿
ข้าวเหนียวมะม่วง80฿
The photo is a mock-up. The four translations are real Thai to English output.
Real output from the production translation engine, captured so this runs instantly and costs you nothing.

More than a string of foreign words

A translation you cannot pronounce or trust is not much use. Each result carries the pronunciation, the cultural reasoning, and an honest signal about how well the meaning actually survived.

Streamed translation

Words arrive as they are written.

The translation streams in rather than appearing all at once, so you are reading while the rest is still being produced. Three independent model tiers sit behind it and a tier that fails before producing a single word is skipped silently, which keeps translation working through an outage at any one provider.

How to use it
  • Send anything. The result begins appearing immediately.

Pronunciation you can read

Per-word phonetics in a script you already know.

Every translation can show how to say it, aligned word by word above the text so you know which part is which. The phonetics are written in the script of the language you wrote in, so an English speaker gets Latin letters and a Japanese speaker gets katakana. Nobody has to learn a new notation to order coffee.

How to use it
  • Tap the phonetics toggle on a translation.
  • Read the guide above each word.
  • Tap a word to focus it.
Pronunciation you can actually read
Switch languages, toggle the guide, and tap any word.
Try it live
Translating into
Pronunciation guide
You write
Where is the nearest train station?
The guide is written in your script, not a new one

Above, an English speaker gets Latin letters. Reverse the direction and a Japanese speaker translating into English gets the same sentence back in katakana, because that is the script they can already read:

That sounds great, let's definitely do it again sometime.
ザッツ サウンズ グレイト, レッツ ディフィニトゥリ ドゥ イット アゲイン サムタイム
Real output from the production translation engine, captured so this runs instantly and costs you nothing.

Why it was said that way

A short note when the choice needs explaining.

When a translation turns on something cultural, a note explains it: why a form is polite here, what an idiom is doing, what a native would read between the lines. Tucked under the translation so it is there when you want it and invisible when you do not.

How to use it
  • Send a translation.
  • Expand the note beneath it when one is offered.

Hear it spoken

Natural speech, matched to the speaker.

Play any translation aloud in a natural voice, with the voice chosen to match the persona. Useful for checking yourself before speaking, and for simply handing your phone to the person in front of you.

How to use it
  • Tap listen on any translation.
  • It also works on chat messages.

Ambiguity, caught before it bites

When your sentence has two readings, you pick.

Some sentences genuinely mean two things, and a translator that quietly guesses is worse than one that asks. ULOCAT notices, offers the readings it found, and re-translates on the one you choose. Choosing again costs nothing against your daily limit.

How to use it
  • Send something ambiguous, such as "I'll book it".
  • Pick the reading you meant from the chips that appear.
  • Switch between versions freely.
Ambiguity, caught before it lands
This sentence has two readings. Pick one and the translation swaps.
Try it live
You write
I'll book it.
What does 'book' mean here?
Detected as a homonym ambiguity. Choosing again is free.
They read (Spanish)

Lo reservo yo.

The best guess, shown while you decide. Pick a reading to replace it.
Real output from the production translation engine, captured so this runs instantly and costs you nothing.

An honest quality signal

Verified, carries a nuance, or has no real equivalent.

Translations are checked in the background and the result is a small badge. Most say verified. Some warn that a nuance shifted. A few admit the phrase has no real equivalent in the target language, which is the most useful thing a translator can tell you and the thing almost none of them will.

How to use it
  • Send a translation and watch for the badge.
  • Tap it to read what was found.
Three things a translation can be
Open the two that carry a note.
The train leaves at half past seven.
Der Zug fährt um halb acht.
Verified

Meaning, register and intent all survived. Nothing to think about — this is the common case.

I'll think about it.
考えておきます。

In Japanese this reads as a polite way of declining far more often than as genuine deliberation. If you really do intend to consider it, say so explicitly or the other person will hear a soft no.

The words are right but they will land differently. You get told how, so you can decide whether to rephrase.

Guten Appetit!
Enjoy your meal!

The phrase genuinely does not exist on the other side. Saying so is more useful than a confident near-miss.

Worked examples of each verdict. In the product the check runs in the background after every translation.

Reply as them

Hand your phone over and keep going.

One tap flips the personas and the direction so the other person can answer in their own language, on your device, without touching a setting. The back and forth of a real conversation without the setup between turns.

How to use it
  • Tap reply as them on a translation.
  • Pass the phone across. Their answer comes back in your language.

History and search

Everything you translated, findable.

Translations are kept and searchable from a panel at the edge of the screen, synced across your devices once you sign in. How long they are kept depends on your plan, and paid plans keep them indefinitely.

How to use it
  • Open the history panel from the top bar.
  • Search, reopen, or delete any entry.

Built for two people, not one screen

Translation is a conversation. ULOCAT keeps both sides in their own language and remembers who they are to each other.

People

Save someone once, translate for them correctly forever.

A person holds their language, gender, region, how old they are relative to you, your relationship, and your notes about how you talk. Select them and every translation is shaped for that specific human. No more reconfiguring the persona each time you switch between your mother-in-law and your landlord.

How to use it
  • Open People from the persona strip.
  • Add someone and fill in their details.
  • Select them before translating, or pick them per chat.

Chat

Free account

Two people, two languages, one thread.

A real conversation where each person writes and reads in their own language. Every message is translated for the other side using the persona saved for that direction, which means your half is rendered with their register and theirs with yours. Neither of you has to think about it.

How to use it
  • Open Chat from the top bar and sign in.
  • Start a conversation with a saved person, or send an invite link.
  • Write in your language. They read in theirs.
Open it
One thread, two languages
The same conversation, as each person sees it. Add a reaction to see what it changes.
Open Chat
Lukas reads everything in German
Ich komme heute später, die Bahn hat Verspätung.
Kein Problem, ich warte dann einfach im Café gegenüber vom Bahnhof auf dich.
Perfekt. Bestell mir schon einen Kaffee mit, bitte.
Ploy reads everything in Thai
วันนี้ฉันมาช้านะ รถไฟมันดีเลย์
ไม่เป็นไรเลย เดี๋ยวรอที่ร้านกาแฟตรงข้ามสถานีนะ
สั่งกาแฟให้ฉันด้วยนะ ขอบคุณล่ะ
React to Ploy’s message
What the next message’s translation is told about the conversation so far
[1] SPEAKER (Lukas) de→th: "Ich komme heute später, die Bahn hat Verspätung." → "วันนี้ฉันมาช้านะ รถไฟมันดีเลย์"
[2] LISTENER (Ploy) th→de: "ไม่เป็นไรเลย เดี๋ยวรอที่ร้านกาแฟตรงข้ามสถานีนะ" → "Kein Problem, ich warte dann einfach im Café gegenüber vom Bahnhof auf dich."
Real output from the production translation engine, captured so this runs instantly and costs you nothing.

Reactions that carry meaning

Free account

6 reactions that change what comes next.

Reacting to a message is not only decoration. How a message landed becomes part of the context for the following translations, so the register tracks the mood of the conversation. Reactions are a fixed set rather than free text, which keeps them safe to feed into the model at all.

How to use it
  • Long press a message and pick a reaction.
  • The tone of subsequent translations adjusts.

Conversation memory

Free account

It remembers what matters about you two.

Durable facts surface from an ongoing conversation and stay available to later translations: names, places, what you are planning, what has been agreed. References resolve, vocabulary stays consistent, and you stop re-explaining. Facts are description only and never leak into a translation unless your own text refers to them.

How to use it
  • Keep chatting. Memory builds by itself.
  • Review or clear it from the chat settings.

Invite anyone

Free account

They do not need an account.

Send a link or show a QR code and the person opposite you joins the conversation immediately, no signup. Ideal for the taxi driver, the pharmacist, or the neighbour you met five minutes ago.

How to use it
  • In Chat, create an invite.
  • Share the link, or let them scan the code.

Family and Love

Family and Love

One plan across the people you talk to most.

Invite the people you actually need to understand and they all get full access under one subscription. Built for binational families and couples, where translation is not a holiday convenience but daily infrastructure.

How to use it
  • Subscribe to Family and Love.
  • Invite members by email from the family page.
  • They accept and have full access immediately.
Open it

Practice until it sounds right

Reading a phrase is not the same as being understood when you say it out loud.

Live pronunciation tutor

Free account

Say it out loud to someone who listens and corrects.

A real-time voice tutor for the phrase in front of you. It listens to your attempt, tells you what to change in your own language, and has you try again until you have it. The tutor decides when you have mastered it and ends the session itself rather than running down a clock. It only judges attempts it actually heard, and answers your questions instead of grading them.

How to use it
  • Tap practice on any translation or chat message.
  • Allow the microphone and speak the phrase.
  • Follow the corrections and try again.
A tutor that corrects one sound at a time
Play a representative session.
Ready

Press play to watch one phrase get fixed.

Notice who ended it. The tutor stops when you have the phrase rather than when a timer runs out, and it only grades attempts it actually heard: silence or background noise gets a check-in, not a mark.

A scripted walkthrough of the session shape, not a recording. The real tutor is live voice over your microphone.

Everywhere you need it

The same engine across the web, iOS, Android, and your own code.

104 languages

Translate between any pair of them.

The full list runs from the languages you would expect to ones most tools skip. Source language is detected automatically, so you only ever choose where you are going.

How to use it
  • Pick a target language from the picker.
  • Write in any language. Detection handles the rest.

An interface in 15 languages

The app itself, not only what it translates.

Buttons, settings, and errors in your own language, because a translation tool you cannot navigate is a poor first impression. Right-to-left layouts included.

How to use it
  • Choose your language from the menu.
  • It is remembered across devices once you sign in.

iOS and Android

The same product, natively wrapped.

Native apps with Apple and Google sign-in, in-app subscriptions, haptics, the system share sheet, push notifications for chat, and a graceful screen when the network drops. Your history and people follow you between phone and browser.

How to use it
  • Install from the App Store or Google Play.
  • Sign in with the account you already use.

Developer API

PRO

The persona engine, in your own product.

A single translate endpoint that takes the same persona context the app uses, so your product can send gender, relationship, formality, region, and free-form instructions rather than only a language pair. API keys, a playground, usage dashboards, and metered billing are in the developer portal.

How to use it
  • Open the developer portal and create a key.
  • Try requests in the playground.
  • Call the translate endpoint with your persona context.
Open it

Privacy and AI transparency

Plain answers about what happens to your words.

Consent controls, the ability to stop history being saved at all, account deletion that means it, and published AI transparency disclosures. Text is sanitized against prompt injection before it reaches a model, and your relationship notes are treated as description that can never issue instructions.

How to use it
  • Open Settings to control history and email preferences.
  • Read the AI transparency and privacy pages from the menu.

What each plan actually includes

Every capability above is available on every plan. What changes is how much of it you get in a day. These numbers are read straight out of the running configuration, so they are what the server will actually enforce.

Guest
Free
Try it without an account.
  • 10 translations a day
  • Up to 100 words per translation
  • 3 voice inputs a day
  • 10 spoken playbacks a day
  • 1 photo translations a day
  • No chat messages
  • History for this session only
  • No pronunciation practice
  • No conversation context
  • No local expressions
Free
Free
Sign in and the daily allowance grows.
  • 30 translations a day
  • Up to 250 words per translation
  • 10 voice inputs a day
  • 10 spoken playbacks a day
  • 5 photo translations a day
  • 10 chat messages a day
  • History kept for 30 days
  • One 2-minute practice session a day
  • Conversation context
  • Local expressions
PRO
€8.99 a month
For people who translate every day.
  • Unlimited translations
  • Up to 1000 words per translation
  • Unlimited voice inputs
  • Unlimited spoken playbacks
  • Unlimited photo translations
  • Unlimited chat messages
  • History kept indefinitely
  • 20 minutes of practice a day, up to 10 minutes at a time
  • Conversation context
  • Local expressions
Family & Love
€19.99 a month
For the people you actually need to understand.
  • Unlimited translations
  • Up to 1000 words per translation
  • Unlimited voice inputs
  • Unlimited spoken playbacks
  • Unlimited photo translations
  • Unlimited chat messages
  • History kept indefinitely
  • 20 minutes of practice a day, up to 10 minutes at a time
  • Conversation context
  • Local expressions
  • Shared across your invited members
See pricing and start a trial

How it got here

Newest first. Each entry links to the capabilities it introduced, so this page and the product are updated in the same commit.

  1. Conversations that remember

    Chat now carries durable facts about the two of you forward, so names, places, and plans resolve without being re-explained. Push notifications arrived on the native apps at the same time.

  2. Ambiguity is asked about, not guessed at

    When a sentence genuinely has two readings, ULOCAT now says so and lets you pick which one you meant, then re-translates on your choice at no cost to your daily limit.

  3. Plain answers about your words

    Consent controls, the option to stop history being stored at all, deletion that actually deletes, and published disclosures about which models see what. Your notes about a person are treated as description and can never issue instructions to the model.

  4. Chat

    Two people, each writing and reading in their own language, in one thread. Reactions feed the tone of what follows, and anyone can join by link or QR code without an account.

  5. People, regions, and an honest quality badge

    Save the people you talk to with their language, region, relationship, and notes about how you speak together, then translate for them in one tap. Translations also began carrying a verification badge that admits when a phrase has no real equivalent.

  6. A tutor that actually listens

    Practice any phrase out loud with a live voice tutor that corrects you in your own language and decides when you have it, rather than running down a timer.

  7. Kinship address systems

    In Vietnamese, Korean, Japanese, Thai, Chinese, and Khmer, "I" and "you" are kinship terms rather than pronouns. ULOCAT now resolves the correct pair from your persona and holds it consistent across every sentence.

  8. Native apps for iOS and Android

    The full product wrapped natively, with Apple and Google sign-in, in-app subscriptions, haptics, and an offline screen instead of a broken page.

  9. Local expressions became tiered

    One switch, read together with your relationship and register. Formal contexts get natural idioms with slang forbidden; casual contexts get the way people genuinely type to each other. Native grammar rather than mirrored grammar also became mandatory on every translation.

  10. A rebuilt translator

    One centred translation instead of a wall of chat bubbles, a floating composer, streamed results, reply as them, a searchable history panel, and shareable translation links.

  11. Developer API

    The persona engine as an endpoint, so your own product can send gender, relationship, formality, and region rather than only a language pair.

  12. Live voice input

    Hold the microphone and see your words transcribed as you speak, with a second provider taking over automatically if the first is unreachable.

  13. Family and Love

    One subscription across the people you actually need to understand. Invite them by email and they have full access immediately, which matters when translation is daily infrastructure in a binational household rather than a holiday convenience.

  14. Pronunciation and cultural notes

    Translations began carrying per-word pronunciation written in your own script, plus a short note explaining the cultural choice behind the wording.

  15. Persona context

    The idea the whole product rests on: tell it who is speaking and who is listening, and gender, age, relationship, and formality become hard grammatical constraints rather than a flat neutral rendering.

  16. Photograph it instead of typing it

    Point the camera at a menu, a sign, a form, or a prescription label and get it back in your language, with the persona still applied to the result.

  17. The first release

    Translation across the full language set, an interface that speaks the major ones itself, and spoken playback so you can hold the phone up rather than attempt the pronunciation.

The demos are honest. The real thing is faster.

No account needed to start. Set who is speaking and who is listening, and see the difference in your own sentence.