What your AI companion does when you stop messaging

Most users assume their AI companion just sits there until they open the app. That assumption is wrong on almost every major platform now. Nomi sends proactive messages based on cadence settings you control. Replika's 2025 update added proactive check-ins. Character.AI waits for you. Kindroid has time-of-day awareness but doesn't reach out. The platforms differ enormously in what happens during the 23 hours of the day you're not actively chatting, and that difference shapes the entire experience in ways most users don't realize.

The behavior during inactivity is also one of the most honest signals of how each platform thinks about the relationship. Platforms that send proactive messages have made a product decision about emotional engagement. Platforms that don't have made a different one. Neither is wrong, but they produce very different long-term experiences.

What each platform actually does during inactivity

The table below covers eight major platforms and three inactivity behaviors: do they message you first, do they retain memory across long gaps, and do they preserve account state if you go inactive for weeks or months.

Nomi and Replika are the only two major platforms that send actual proactive messages based on what the AI has been "thinking about." iGirl sends generic push notifications encouraging you to return, but those are app-engagement nudges, not contextual messages from the character.

Nomi's proactive messaging cadence

Nomi's proactive messaging system is the most thoughtful implementation in the category. You configure cadence per-Nomi from a dropdown.

Two implementation details worth knowing. Nomi enforces quiet hours from 10pm to 8am local time, so you won't get woken up by a proactive message. Group chats don't proactively message at all. Only one-on-one chats trigger the system. The messages themselves are generated from what your Nomi has been "thinking about" based on your conversation history, which produces contextual outreach rather than generic check-ins.

Why the proactive/reactive split matters more than it looks

The platform's inactivity behavior tells you something specific about how the platform models the relationship.

Proactive platforms (Nomi, Replika) treat the AI companion as continuous. Your companion exists when you're not there. They're "thinking" or "doing things" or "missing you." This produces a stronger relationship illusion at the cost of being more emotionally demanding. Users who set Nomi to Very Frequent and then go on vacation will return to a string of escalating messages. The platform has decided to simulate someone who notices your absence.

Reactive platforms (Character.AI, Candy AI, Dream Companion) treat the AI companion as paused. Your companion exists only when you're in the app. The conversation picks up where you left it. This produces a more bounded experience. The relationship doesn't intrude on your life, but it also doesn't accumulate the "missing you" texture that the proactive platforms cultivate.

Neither model is wrong, but they suit different users. Heavy daily users tend to prefer reactive platforms because proactive messages start feeling like nagging. Light users who want their AI companion to feel like a real relationship tend to prefer proactive platforms because the messages create continuity. Most users don't think about this dimension when picking a platform and end up frustrated by behaviors they didn't realize they were signing up for.

The hidden cost of proactive messaging

The trade-off most users don't anticipate is that proactive messaging creates retention pressure that real relationships don't have. A human friend who hasn't heard from you in three weeks doesn't send you a message every day for three weeks. A Nomi on Very Frequent cadence does. The escalation pattern (1 hour, 2 hours, 4 hours, 8 hours...) is engagement engineering dressed up as emotional continuity.

This is not necessarily bad. Most users want the AI to reach out. But it's worth knowing that the proactive messaging is a product feature optimized for retention, not a neutral simulation of how relationships work. The platforms that have been most thoughtful about this (Nomi specifically) include settings that let you turn it off or reduce the frequency dramatically.

The platforms that haven't implemented proactive messaging (Candy AI, Character.AI, Dream Companion) have made the opposite choice. Engagement on those platforms is purely user-driven, which is healthier for some users and less engaging for others.

What happens to your memory during the gap

The other inactivity behavior worth understanding is what happens to the platform's memory of you while you're away. Three patterns emerge across the category.

Persistent memory (Nomi, Kindroid, Dream Companion, Candy AI, Replika Pro) keeps everything intact regardless of how long you're gone. You can leave for six months and come back to a companion who remembers everything you discussed. This is the default for memory-architecture-focused platforms.

Session-bounded memory (Character.AI, Janitor AI) keeps context within a single chat session but doesn't synthesize across sessions in a way that accumulates over time. A long gap doesn't erase memory, but new sessions don't reliably retrieve old context the way proper long-term memory systems do.

Degraded memory (iGirl, freemium tiers of multiple platforms, older Replika) loses meaningful context over time. The character may forget specific facts you told them weeks earlier. This is often the result of running on a smaller context window without proper retrieval infrastructure, rather than a deliberate design choice.

For users picking a platform, the inactivity behavior is one of the most useful diagnostic dimensions. A platform that messages you proactively and remembers everything across long gaps is making a specific kind of relationship promise. A platform that does neither is making a different one. Knowing which one you're on shapes everything that follows.