We built Aria around one belief: the most valuable thing an AI
can do isn't answer faster; it's remember longer.
Right now Aria tracks three things across every call:
Facts (your dog's name, your job, what you're working on)
Moments (emotional beats, hard days, wins)
@ritesh_p2 The continuity angle is really interesting, especially since most AI conversations feel disposable once the session ends. How do you balance long-term memory with privacy and user control? For example, can users easily review, edit, or remove memories that Aria has stored over time?
@nicole_hynek Memory review and editing are on our near-term roadmap, and here's how we're thinking about it: Aria's memory should feel like a trusted friend's memory, not a surveillance file. So the interface we're building lets you see what Aria remembers, remove anything that feels off, and correct things she got wrong. You stay in control of what persists.
The harder design problem is what to surface and when. Showing someone a database of extracted facts about themselves can feel clinical and strange, the opposite of what Aria is supposed to feel like. So we're being deliberate about how that UI works, not just whether it exists.
Privacy-first is not negotiable for us. Aria only remembers what you tell her directly. Nothing is sold, shared, or used to train models without explicit consent.