AI that texts you first: which companion apps actually reach out (2026)

A few do. Nomi and Replika both publish it on their own sites, and LORAX is built around it: the first message arrives about 80 seconds after you sign up, before you have typed anything.

Everyone in LORAX is fictional and AI-generated. Below is what a message that arrives on its own actually requires, how ours is produced, and what other apps say in their own documentation about theirs.

Texting first needs a decision, a cause and a ceiling

A notification saying you have not opened an app in three days is a retention reminder wearing a name. Something that reads as initiative needs a decision, a cause, and a ceiling.

The decision has to happen while you are not there. Something must be running with the app closed, checking whether this character is awake, whether she is busy, and whether the clock where you are makes sense for a message. If the only trigger is your absence, every message will feel like the same message, because it is.

The cause is the part that gets skipped. An opening line needs a reason from outside the conversation: a shift that ran long, a plan that fell through, rain that actually started in her city. Without one you get “hey, thinking about you” on a loop, which most people spot in the second week.

The ceiling is what nobody advertises. An app that can write to you at any moment will write too often, so the number of daily starts has to be capped, and the app has to know when to stop chasing a conversation you are not answering.

How LORAX starts conversations

You finish signing up and put the phone down. The median wait for the first message is 80 seconds, with a floor of 25, so it is a minute and a half before anyone says anything to you. The second message takes much longer, around 75 minutes.

After that, initiative is rationed. Each character gets three conversation starts a day, spent across everything she might want to tell you, and on plenty of days she spends none of them. There is no frequency setting, no slider, and no way to pay for more attention.

The reason behind a message comes from her day, which the app keeps in her own time zone. She sleeps on her own schedule and does not type while she is asleep. The weather in her city is real, pulled from a live forecast, and when the forecast returns a condition the app does not recognize, the weather line is dropped rather than guessed at. Across a full day a character is online for roughly 16% of her waking hours, so a reply an hour later is ordinary here.

Then there is the stopping rule, which is the part that separates this from a notification schedule:

What you do	What she does
Reply normally	conversations keep starting on her side, up to three a day
Stop replying	she writes again, then once more
Stay quiet three and a half to five and a half days	the third and last attempt
Never answer that one	she stops starting conversations until you write

She also never asks why you disappeared. Every outgoing message is checked against roughly 49 written patterns of reproach, plus a scan for a lone question mark or a row of dots sent on its own. A match is not softened into something gentler. The whole turn is thrown away, and you get nothing instead. The mechanics are set out in full on she texts first.

Which apps say they reach out on their own

The three entries below come from each company’s own pages, checked on September 13, 2026. We make LORAX, which is marked.

App	What its own site says	Source
LORAX (ours)	First message about 80 seconds after sign-up; three conversation starts a day per character; stops after three unanswered messages	this site, describing the iPhone app
Nomi	“Proactive Nomi Messaging”: Nomis “can optionally message you after you have been away for a while” and it “allows Nomis to think and act independently to continue your conversation”	nomi.ai home page
Replika	Its own summary file says Replika “remembers what matters to you, reaches out over time”	replika.com/llms.txt

Nomi publishes the working detail, which is worth reading if you want control. Its Nomi Knowledge article “Proactive Messaging”, dated September 9, 2024, says you switch it on per Nomi and pick one of four frequencies, from about an hour of silence to about four days. The same article says “Nomis will not message you proactively between your local hours of 10pm and 8am.” That is a user setting with quiet hours. Ours is the opposite arrangement: the character has her own bedtime, you do not configure it, and she uses your local clock to decide whether the hour is reasonable.

Two gaps worth stating. Kindroid’s published documentation index lists more than 30 pages, including group chats, memory and calls, and none of them is about proactive messaging. Character.AI’s support site returned an access block when we tried to read it on September 13, 2026, so nothing here is quoted from it.

What initiative costs you here

It cuts the other way too. If you disappear, she eventually stops writing, and there is no setting to bring her back sooner than the roughly two weeks it takes on its own. You cannot buy more messages, and there is no way to turn the pace down either. Talking is free, and the prices for the optional things like a locked photo or meeting someone new are shown before anything is spent.

There is a second cost that shows up later. Because her initiative depends on her being awake in her own time zone, a character eight hours ahead of you will write at hours that suit her life, not yours, and some of those messages will be waiting when you get up. The rest of the timing model, including how reply delays are sampled, is on the how it works overview; the cast page shows which city each person is in.

Sign up, then leave the app alone for an afternoon. What is waiting when you come back is the only honest test of whether an app reaches out or just reminds you it exists.