This is the week NEURA stopped feeling like a chat window and started feeling like someone in the room. I gave it a real personality and a smarter brain, and honestly the difference is bigger than I expected. Same product, but it finally acts like it knows me.
The personality came first, because a companion with no voice is just a search box that talks back. I wrote a built-in sass system inspired by JARVIS, wit and attitude that scale with a dial you set. Crank it up and NEURA gets sharp and a little smug. Dial it down and it plays things straight. That attitude shows up in two places: the way it talks in a full reply, and quick reactive quips it fires off in the moment. Getting the dial to feel like one consistent character at every setting, instead of two different bots, was the fussy part.
Then the brain. I added real tool calling, so NEURA can actually run actions and read back what happened instead of just describing what it would do. That one change moved it from talking about things to doing them, and it makes everything downstream feel real.
I also gave it semantic memory, so it pulls the right facts about you at the right moment instead of forgetting between messages or dumping everything at once. Paired with that, I wired in live context: it knows the time, what view I am looking at, and what just happened a second ago.
Put together, those pieces are the jump I have been chasing. NEURA went from answering questions to keeping up with me, holding a thread, having an opinion, and reaching for a tool when words are not enough.
Next week I want to push the memory harder and see how far the personality holds under pressure. But for the first time, sitting with it, it felt less like software I am building and more like something I would actually want around.