Read their text through their attachment style and yours.

A relationship text analyzer reads a message, or a stretch of conversation, and tells you what it most likely means. It works from the words on the screen and whatever you tell it about the situation. None of that shows what the sender feels, whether they like you, or where this is going. Put a percentage on any of those and you have a guess with a decimal point.

Most analyzers aim their score at the other person, with an interest percentage or a compatibility number. Decode aims its meter at you. Spiral risk rates how hard this message is likely to hit you, given your attachment style and theirs. Nothing you do with their text changes how interested they are. You still decide what you send in the next ten minutes, and that decision is what the meter is for.

Decode reads the message and never rates the person who sent it, since one text says little about who someone is. If you’re comparing analyzers, look for one that reads through both attachment styles and shows you what your own pattern hears, and check what it says happens to your text after you paste it. Then decide what you want to know. "Does he like me?" has no answer inside a text. "What does this most likely mean, and why is it hitting me this hard?" is a question a text can answer.

A single text gives you the words and little else. Your attachment style supplies the tone you hear, and the sender’s style shaped the words they chose. Decode reads the message through both. The two examples below run one pairing from each side: an anxious reader with a dismissive avoidant sender, then the reverse.

Read the "What your brain hears" line first. It’s the message you would have answered if you’d replied in the first minute.

The reverse case runs on the same loop. If you lean avoidant, a text with an edge on it can make going quiet feel like the safe reply, and an anxious sender reads silence as proof of the fear behind the edge.

Both replies answer the question under the message and skip the argument about how it was said. For the message you’re about to send, Olively’s Translate rewrites your draft so their style can hear it, with the feeling left in. The attachment style translator has example texts for all sixteen pairings.

Some texts only make sense inside the argument they came from. In conversation mode you paste the exchange or add screenshots of it, up to 10 on iPhone or 5 on Android, and Olively sorts the messages in your screenshots into yours and theirs by which side of the screen each bubble sits on. Decode reads the thread as a whole and tells you who escalated and who withdrew. Then it names the pattern running between you and suggests one next message.

Couples research describes three conflict patterns, and each leaves a signature in a thread. In a 2026 study that followed 263 couples for a year, partners with more attachment insecurity used more demand-withdraw and demand-demand communication, and both patterns went with lower satisfaction for each partner. Avoidance showed the same link through the third pattern, where both people pull back. Sue Johnson, the couples therapist behind Emotionally Focused Therapy, gave the three plainer names.

Metrics like reply time and who texts first describe what happened in a thread. A named pattern explains why the same fight keeps coming back, and it tells you what your next message has to do.

Your stomach drops at "haha," and you’ve written half a reply in your head before the second sentence lands. In lab studies published in 2006, people high in attachment anxiety saw emotional expressions on faces start and fade sooner than others did, and those early calls were less accurate. The accuracy problem went away once they had to watch as long as everyone else.

Those studies used faces, so carrying them over to texts is an inference: a text gives a fast read even less to go on, and you fill whatever it leaves open with what you already expect. Collins and Feeney found in 2004 that people with insecure attachment rated notes from their partner as less supportive than outside judges rated the same notes, most of all when a note was ambiguous.

"What your brain hears" puts the fast read into words so you can see it before you answer it. Give the message one read and one reply, then stop. Answer tonight if it asks something with a clock on it, like a plan or a time. Wait until morning if your reply would mostly be about how it made you feel, or if they’ve ended the conversation for the night.

Decode reads through two styles you’ve already set. Yours comes from Olively’s attachment quiz, which takes about two minutes and is free on the web. Theirs is the style you pick when you add your partner, with a "help me figure it out" path if you’re unsure, and you can change it as you learn more.

That order matters. Researchers measure attachment style with questionnaires about how someone feels and acts across close relationships, and one thread from a bad night can make almost anyone look anxious or avoidant. An analyzer that guesses a style from the fight you’re upset about builds your misread into everything it tells you after that. Decode starts from the styles and reads the thread through them.

The word "private" covers three separate facts about Decode. On your phone: when you add screenshots in the iPhone or Android app, Olively pulls the text out of them on the phone itself, and the images are never uploaded. Sent for the read: the text, pasted or pulled from a screenshot, goes to Olively’s servers so Decode can read it. Saved to your History: Olively keeps each decode in your private History, and you can delete them one at a time, all at once, or along with your account.

Your partner is never notified that you decoded their message, and they don’t need the app. The privacy policy has the full terms.

You can, and plenty of people do. A general chatbot doesn’t know your attachment style or your partner’s unless you explain both, and it reads the message through the account you give it, often written while you’re still upset. A 2026 study in Science tested 11 leading AI models and found they affirmed users’ actions 49% more often than people did. In three experiments with 2,405 participants, people who had a single conversation with an agreeable model came away less willing to take responsibility and repair a conflict, and they trusted and preferred that model.

Decode starts from both styles and the words on the screen, and the first thing it shows you is your own fast read of the message. A chatbot that leans toward agreeing with you can end up confirming that read.

Decode reads what you paste or screenshot, which means one message or one stretch of conversation. It doesn’t import chat-history exports, so a report on two years of WhatsApp messages needs a tool built for exports. You won’t get a score for whether someone likes you or a forecast of where the relationship is going. Every read is a likely meaning, and a likely meaning can be wrong.

It isn’t a safety tool, and Olively isn’t therapy. If you’re worried about your safety, start with the support below.