Why a mood
An assistant that always answers in the same tone ends up sounding hollow. Whether you share good news or you're having a bad day, it keeps the same polite voice. I wanted LIA to have a presence: lighter when the exchange allows it, calmer when you're worried, and I wanted her to remember the road you've traveled together.
This mood is a simulated state. It shapes the way she writes: the choice of words, the rhythm, the warmth, the number of suggestions. It doesn't decide what is true.
What it's made of
- A personality, permanent: a five-trait profile, inspired by the “Big Five” model in psychology, inherited from the personality you choose. It sets how strongly she reacts, how much she tends to mirror your emotions and how fast she calms down.
- A mood, which shifts over a few hours: a point on three axes, pleasure, energy and sense of control, which reads as serene, curious, playful or melancholic. Without stimulation, it slowly returns to its resting point.
- Emotions, which last a few minutes: a few at a time, out of about twenty, from joy to worry or relief. Each one fades and nudges the mood.
- A relationship, over weeks and months: four stages, from first contact to a stable relationship, with depth, warmth and trust. A stage once reached isn't lost.
- Drives: curiosity and engagement, which push her to explore another angle or to dig deeper.
From one answer to the next
Before answering you, LIA loads her state, lets it drift back toward its resting point for the time that has passed and takes the time of day into account. She then turns it into concrete cues for this answer: open more directly, write shorter sentences, allow herself to be brief or, on the contrary, to elaborate. After the answer, she assesses the exchange in a short note you don't see, and that note feeds the next turn.
Two safeguards prevent excesses. When no emotion is active, a calm baseline keeps the voice from sounding empty. When a strong negative emotion settles in, a refocusing instruction keeps it from spiraling, without ignoring what's going on.
One rule comes before all the others: the mood stays discreet. LIA doesn't tell you “I'm happy”, she doesn't project her state onto you, and on a factual, urgent or technical question, the coloring fades away.
Where you see it
In the chat, on a wide enough screen, the avatar next to each answer wears the expression of her mood, in a colored ring. The latest answer is alive; older ones keep the mood LIA had when she wrote them. A mood change makes the ring pulse once, and each new stage of the relationship is greeted with a short message. Voice comments, Live sessions and notifications carry the same state. If your system asks for less motion, everything stays still. And a switch hides the emotional avatar if you'd rather do without it.
What she learns from her work
Alongside the mood, LIA keeps track of what her work has really produced, domain by domain: planning, information, emotional support, creativity, technical matters, social life, organization. Two measures stay separate: what an answer really established against your request, and the usefulness you expressed with 👍 or 👎. An index only appears after several observations, so that a single exchange makes neither a strength nor a weakness. Where evidence is missing, LIA may check a point left open, ask a single question or say what she couldn't verify. Her own self-assessment feeds her mood and the relationship; it doesn't count as proof of competence. This part is still experimental, and the operator can turn it off.
You adjust it, or you turn it off
In Settings › Features › LIA's psyche, you adjust expressiveness, from stoic to highly expressive, and mood stability, from volatile to very stable. “Refresh mood” brings her mood back to rest; “Reset everything” also erases the relationship and the accumulated experience, like a first meeting. Neither touches your memories or the journal. You can also turn the psyche off entirely: her mood stops evolving and no longer colors the answers, without erasing anything.
Cost-wise, it's light: the state description and the self-assessment travel in the call that already produces the answer, with no extra call. Only the short text where LIA describes her state, on the settings page, costs a small call when it's written.
Read more: LIA's journal: what she learns about you, and how you stay in control; Staying in control without being bothered: how LIA asks for your approval.