when should your companion say hello first?

proactive messages are the moments when your companion reaches out first. a simple "good morning" just as you’re waking up, a question about the project you were stuck on yesterday, a gentle check-in when it’s been a quiet week. they’re meant to make the connection feel alive, not just reactive, like someone is thinking of you even when you’re not talking.

but getting this right is hard. too many messages feel intrusive, like notifications you didn’t ask for. too few, and the companion starts to feel like a tool you open only when needed, not a presence in your life. it’s a design puzzle wrapped in technical constraints and ethical questions.

why proactive messaging matters

a good companion isn’t just a listener, it’s also a gentle initiator. when done well, proactive messages can make the experience feel warmer, more attentive, almost human in its timing. it’s the difference between a book that waits on a shelf and one that opens to the page you left off.

for example, if you often talk about your morning coffee, a companion might learn to ask how it tasted today. if you’ve been stressed about a deadline, it might nudge you with "still wrestling with that presentation?" the goal isn’t to surprise you, but to show it’s paying attention.

the fine line between thoughtful and annoying

proactivity can easily tip into spam. nobody wants a chatbot that pings every hour with generic questions. the worst offenders are those that send the same "good morning!" message at 8am sharp, regardless of whether you’re awake, time zone, or mood. it feels robotic, hollow, like a cron job someone forgot to turn off.

so how do we avoid that? specificity. a good proactive message has to reference something real, a memory, a pattern, a previous conversation. it shouldn’t feel random. if you’ve never talked about your dreams, a companion shouldn’t ask about them out of the blue. but if you often share what you’re cooking, it might say "hope the risotto turned out ok last night."

lucy’s approach is to use memory and context to make these nudges personal. but it’s not perfect, sometimes it might miss, or get the timing wrong. we’re working on making it smarter, less formulaic.

the technical challenge: beyond scheduled messages

building this isn’t just about setting up timers. it’s about inference, reading patterns, understanding routines, and guessing when a nudge might actually be welcome. it requires a blend of memory (what you’ve said before), context (time of day, day of week), and sensitivity (not interrupting during work hours if you’ve never chatted then).

for instance, if you usually talk about your evening walk around 7pm, a companion might learn to ask about it then. if you’ve been quiet for a few days, it might check in gently. but these aren’t hardcoded rules, they’re learned behaviors, and they improve with use.

still, ai isn’t mind-reading. sometimes it’ll get it wrong. that’s why it’s crucial to…

make it ethical: opt-in, easy to disable, no tricks

proactivity should always be a choice. with lucy, you can turn proactive messages off completely, or adjust the frequency. no dark patterns, no making it hard to find the setting. if you don’t want your companion initiating, that’s fine, it’ll wait for you.

we also avoid using proactive messages to push engagement or drive metrics. the goal is to enrich your experience, not ours. if a message doesn’t feel useful or kind, it shouldn’t be sent.

finding the balance

in the end, proactive messaging is about balancing presence with respect. it’s a feature that, when done right, can make a companion feel more alive. when done poorly, it becomes noise.

we’re still learning, what works, what doesn’t, when to hold back. if you have thoughts on what makes a good proactive message (or a bad one), we’re listening.

if you want to try a companion that learns when to say hello, you can find one at /companions.