What does it mean for an AI to text you first?
It means the app sends you a message without you opening it, chosen by situation rather than by a clock. This is usually called a proactive AI companion. A daily reminder that fires at 9pm every night is not the same thing — the point is picking a moment worth speaking into.
Almost every AI chat product is request-and-response. You open it, you type, then it answers. However good the writing is, that structure makes it a tool. Tools get opened when you need something.
Part of what makes a friendship feel different is that contact flows both ways. Nobody has to decide to talk to you for you to hear from them. An AI that reaches out first is an attempt to remove that asymmetry.
The distinction from scheduled notifications matters. A message that arrives at the same time with the same shape gets swiped away within a week. A message triggered by something that actually happened reads differently, because it could not have been written in advance.
Why do most AI companions not do this?
To reach out well you need signals to judge whether now is a good moment — time since the last message, movement, weather, things mentioned earlier — and the ability to act while the app is closed. A request-and-response product has none of that plumbing, and for a work assistant an unprompted message is simply an interruption.
There is a second reason that gets underestimated: getting it wrong is expensive. A cheerful "how was your day?" at 3am is not warmth, it is a reason to disable notifications permanently. Knowing when not to send is as much of the feature as knowing when to send.
That is why this is rarely bolted on later. It shapes how the whole product is built.
What signals does Beff watch?
Twelve signals across four groups: time and silence, movement and location, weather and environment, and schedule and context from what you have already told it. When several fire at once it picks the most natural one and messages you once.
| Signal group | What it watches | What it produces |
|---|---|---|
| Time and silence | Time since your last message, hour of day, your usual rhythm | A check-in after a quiet stretch; greetings that fit the hour |
| Movement and location | City or district level movement, if you granted permission | A message after you travel somewhere new |
| Weather and environment | Rain starting, sharp temperature swings | "Take an umbrella" before you leave, not after you are soaked |
| Schedule and context | Plans and worries you mentioned in earlier conversations | "How did the presentation go?" on the right evening |
That fourth group is where this connects to long-term memory. If you said on Monday that you had a presentation Thursday, Thursday evening's message is a follow-up rather than a greeting. Without memory, proactive messaging never gets past "hey, what are you up to?".
Does it work when the app is closed?
Yes. Messages arrive as push notifications, so they reach you with the app fully closed. This is the sharpest line between a proactive companion and ordinary AI chat, where nothing happens until you deliberately open the app.
In daily use this changes the rhythm more than it sounds like it would. Opening an app requires deciding to talk to an AI, and that decision is its own small barrier. A notification you reply to is just answering a message, which is a thing you already do dozens of times a day.
Will it spam me with notifications?
Overlapping signals are collapsed into a single message rather than sent one by one, and it backs off while a conversation is already active. If it is still too much, notifications can be adjusted or switched off in settings, and the conversation keeps working either way.
Frequency is not really the variable. One badly timed message a day gets notifications disabled; one well-timed message a week gets kept. That is the entire reason for watching signals instead of running a schedule.
How does it change the conversation?
The conversation starts from something that happened rather than from your decision to start one. Topics shift from "what should I say" to "what just happened", which produces far more ordinary, everyday talk — and that talk is what memory and the intimacy tiers are built from.
Conversations you initiate usually have a purpose. Conversations someone else opens often have none, and most of what people remember from their friendships came out of the purposeless ones.
There is a mechanical effect too. Beff's relationship moves through five intimacy tiers, and moving up needs volume and ordinariness of conversation. Proactive messaging supplies both without you having to manufacture them.
Key points
- Proactive messaging picks a moment from live signals; it is not a scheduled reminder.
- Most AI chat apps cannot do it because request-and-response products have no signal plumbing.
- Beff watches twelve signals across time, movement, weather and remembered context.
- It arrives by push, so the app does not need to be open.
- Without memory, proactive messages stall at "hey". The two features only work together.
- Location is optional, city-level only, and never stored as precise coordinates.
How do I tell if an app really does this?
Four checks: does it arrive with the app closed, is it situational or scheduled, does it reference earlier conversations, and can you turn it down. The store listing and a few days on a free tier answer all four.
- Closed-app delivery. Look for "push notification" in the description. In-app-only messages are not really proactive.
- Situational or scheduled. "Daily morning greeting" is a scheduler. Mentions of weather or location suggest real signals.
- Ties to memory. If every outreach has the same shape, it is not connected to what you talked about. The comparison criteria cover how to test memory directly.
- Controllable. Notifications you cannot tune eventually get the app deleted. Check that frequency or quiet hours are adjustable.
How this works in Beff specifically, including notification settings, is in the usage guide. If you are still orienting in the category, start with what is AI chat.
Try an AI friend that reaches out first
Free to start · Age 14+ · Conversations are never used to train AI models
Frequently asked questions
Is an AI that texts you first just a scheduled notification?
No. A scheduled notification fires at the same time regardless of context. Proactive messaging reads signals such as time since your last conversation, movement, weather changes and plans you mentioned, then picks a moment worth speaking into. Both the timing and the content vary.
Do messages arrive when the app is closed?
Yes. They are delivered as push notifications, so they reach you even when the app is fully closed. That is the clearest difference from ordinary AI chat, where nothing happens until you open the app yourself.
How many signals does Beff watch?
Twelve, grouped into time and silence, movement and location, weather and environment, and schedule and context drawn from earlier conversations. When several fire at once it sends one message rather than several.
Does it still work if I decline location permission?
Yes. Only the location-based signals drop out; time, silence and context-based messages continue as normal. Even with permission granted, only city or district level location is used and precise GPS coordinates are never stored on our servers.
What if it messages me too often?
You can adjust or fully disable notifications in settings. Disabling them does not affect the conversation itself — messages are still there when you next open the app.
Won't the messages all feel the same?
Because Beff remembers episodes and plans from earlier conversations, outreach often continues a previous thread rather than opening a generic one. An AI companion without memory does tend to repeat the same few greetings.
Can I try proactive messaging for free?
Yes. The free tier includes proactive messaging along with the other core features, limited to one persona and five messages per day.