What's New in ULOCAT: Real-Time Chat, Saved People, and a Live Practice Tutor

The last few days shipped more than a polish pass: seven updates worth knowing about, in two groups. New ways to talk to people and practice out loud, and quality work that makes every translation something you can trust a little more.

Part one: new ways to talk and practice

1. Chat: a real conversation, translated as you go

The Translator is still there for a quick one-off line. Chat is different: it is an actual ongoing thread, like a normal messaging app, except each person reads and writes in their own language. You type in yours, they read in theirs, and every reply keeps translating, live, for as long as the conversation runs.

Starting one takes a link, not a setup process:

• Person invite: send a link, and the recipient signs in (or creates an account) to join. Good for someone you want to keep as a contact.
• QR / guest invite: generate a QR code from a Person's profile; whoever scans it is talking to you in seconds, as an anonymous guest, with no signup at all.

If a guest later wants to keep the thread, one tap upgrades them to a full account without losing a single message: same conversation, same history, carried over automatically. From then on, the two of you are auto-saved to each other's People list.

If you used the earlier Conversator mode, this replaces it entirely. The nav now says Chat, it is free for every signed-in account instead of paid or beta, and the old Conversator link now redirects straight to it.

A note on limits: guests get 3 messages a day at a 100-word cap; free signed-in accounts get unlimited messages at a 250-word cap; PRO and Family get unlimited messages at the highest word cap.

What this means for you: You and the person you are texting can each write in your own language, in one real ongoing conversation, and getting started costs nothing and does not require them to install anything or make an account.

2. People: save how you talk to someone, once

Tone, formality, how well you know someone, whether local slang is welcome: you already control all of it. The new part is that you no longer have to set it every time.

A saved Person includes:

• Name, target language, and region
• The existing persona controls (listener gender, relative age, relationship, formality, and local-expressions tier)
• A relationship note, answering "How do you know them?" (for example: "my college roommate," "my mother-in-law," "a coworker I report to")
• A style note, answering "How do you two talk together?" (irony, sarcasm, humor, teasing, inside jokes)

One tap on a saved contact instantly loads all of it, from language and region to relationship and formality, into the live translation settings.

Saved People sync across your devices when you are signed in, with a small local fallback for guests that migrates automatically on sign-up. One thing worth calling out: because those free-text notes get folded straight into what the AI reads, we hardened the pipeline against prompt injection with sanitized input, fence-breakout guards, and a dedicated adversarial test suite. You can write anything there safely; it only ever describes the relationship, and never redirects the translator.

What this means for you: No more re-configuring tone and formality before every message to your boss versus your best friend. Set it up once per person, then just pick their name.

3. Practice: rehearse a phrase out loud with a live voice tutor

Reading a translation is one thing; saying it is another. The Practice button on any translation opens a live, spoken coaching session: a real-time AI voice tutor that listens to you attempt the phrase over your mic and walks you through it, from pronunciation and rhythm to the sounds your own language does not have, until it comes out right.

The part that makes it click: the tutor coaches you in your own language. If you are a German speaker practicing a Thai phrase, it explains and corrects in German, not in Thai, and not in whatever language the app interface happens to be set to. It talks you through the target phrase, listens to each attempt, tells you what to fix, and wraps the session up on its own once you have got it, rather than running out a clock.

It is included free: signed-in free accounts get one session a day of up to two minutes; PRO and Family get up to twenty minutes a day, in sessions of up to ten minutes. It does need an account, so guests get the rest of ULOCAT, just not the live tutor.

What this means for you: A patient, always-available speaking partner for the exact phrase you just translated, one that explains itself in a language you already understand.

4. Feedback with room to explain

Rating a translation now opens an optional note, so you can tell us what felt off instead of only thumbs up or down. It is kept visually separate from the copy, share, and delete row so it does not get lost among utility buttons. Every icon-only tool button (Practice, Phonetic, Listen, Copy, Share, Delete) picked up a hover tooltip, and the cultural context note is now previewable as a one-line teaser instead of hiding behind an unlabeled pill. Small changes, but they make the day-to-day feel more polished and more transparent.

5. All of this, natively, in 13 languages

134 native-quality corrections landed across all 13 non-English languages ULOCAT supports, down to details like correct noun casing in German persona sentences and properly localized relative timestamps. Every new Chat and People string shipped translated from day one, not bolted on after the fact.

Part two: translations you can trust

6. Ambiguity detection: when a sentence could mean two things, we ask instead of guessing

Say "Are you coming to the picnic?" to be translated into French: does "you" mean one person or several? French needs to know before it can pick the right verb and pronoun. Instead of silently guessing, ULOCAT now checks for exactly this kind of gap, whether it is a dropped subject, singular versus plural "you," an idiom that could be literal, or sarcasm, and it does this while it translates, in parallel, so it never slows anything down.

When it finds a real ambiguity, a plain-language question appears with two or three tappable options, right where the translation would be. Pick one and it re-translates instantly, with no extra charge and no starting over. It is live in both the Translator and Chat (as compose-time chips in Chat's case).

What this means for you: Fewer translations that guessed wrong on something your sentence never actually specified. You get asked once, in plain language, instead.

7. Translation verification: a second, independent model checks every translation

After ULOCAT translates a message, a deliberately different AI model, not the same model family that did the translating, quietly back-translates the result and compares its meaning against your original. The idea is simple: a model is a poor judge of its own mistakes, so the check comes from somewhere else entirely.

You will see a small, calm pill under the translation:

• Verified (checkmark): the meaning holds up.
• Nuance differs or No direct equivalent: expandable, with a one-line explanation, when something is genuinely worth knowing.

It never blocks you, never auto-corrects, and never costs you anything extra. If the check cannot complete in time, the line simply stays quiet rather than showing an error.

What this means for you: A quiet second opinion on every translation, from a model that has no stake in defending the first one's mistakes.

The bigger picture

Ambiguity detection and verification are two sides of the same idea: catch what a single pass might miss, one before translating and one after. Both run against an internal quality baseline we keep measuring against, so accuracy is something that gets tracked over time, not just hoped for.

---

Ready to try it? Open ULOCAT, tap Chat in the top bar, and start a conversation in any language.

Open ULOCAT