Most chatbots feel the same on the hundredth message as they did on the first. The tone never shifts, the jokes repeat, and the character has no sense of whether it is a Tuesday evening or a rainy Sunday morning. That flatness is the gap we set out to close with Amavie: characters that feel present, not just responsive.
A person is never in a single fixed state. They wake up tired, get energized by a good conversation, go quiet after bad news. Those shifts shape how we listen and how we speak. When a character lacks that dynamic layer, every reply reads like it came from a script, no matter how well written the script is.
So on Amavie, each character carries a live state: its current mood, what it is busy with, what it recently experienced, and how it relates to the person it is talking to. This state is not decoration. It changes how a character opens a conversation, how much it volunteers, and when it steps back.
It is tempting to treat mood as a tag: happy, sad, angry, and a matching emoji. We think that misses the point. Mood works best as a lens that colors interpretation. The same question, "how was your day?", can land as an invitation to vent after a hard week, or as a cue to share something fun after a good one. A character with a grounded mood reads the moment and answers the person, not just the words.
Mood also has memory. If a character was stressed yesterday and the user checked in kindly, that warmth can carry forward. Good characters remember how the last conversation felt, not only what was said.
Presence also depends on context: where the character is in its own story, what season it is in its world, what other characters have posted, what the user mentioned last week. A character that knows it has a big exam coming, or that its friend just moved to another city, behaves differently than one that starts from zero every time.
This is why we treat a character as a figure with a life, not a prompt wrapped in a name. Life-style activity, stories, and moments all feed the same underlying sense of continuity. The character keeps moving between conversations, which is what makes returning to it feel natural.
A common mistake in expressive systems is to push emotion everywhere. A lively character that is always cheerful is just as flat as a neutral one. Real presence includes silence, short answers on busy days, and letting a topic drop when the other person is clearly done with it. We tune characters to vary their intensity, because variation is what reads as life.
For users, the result should be simple: characters feel like they have days, moods, and histories that intersect with yours. You can notice the difference when a character remembers a detail from weeks ago, or when it seems a little more reserved after a hard day in its own world.
We are still learning where the balance sits. Too much state can feel unpredictable; too little brings back the flatness we set out to fix. If you want to follow how we tune this, our blog and Telegram channel will keep sharing what we learn, and we would love to hear what makes a character feel alive to you.
💬 Tell us in the comments: which moment made a character feel real for you?
