About
Novalis.
An AI that gets to know you, and never hides how.
The problem.
You’ve explained your situation to an AI before. Your project, your people, your constraints, why something matters. Then the conversation ended, and next time, you did it again.
Some AI now says it “remembers.” What that usually means is a pile of saved notes you never asked for, can’t fully see, and can’t correct without wiping the whole thing. That’s not memory. That’s a black box that happens to know your name.
What Novalis is.
Named for the poet who dreamed of a flower no one could forget, Novalis is built on one idea: an AI that’s genuinely useful over months, not just minutes, has to do two things nobody else is doing together: hold onto what matters, and never hide how or why.
How it works today.
Every conversation gets watched by something we call the Daily Observer, constantly reading for what actually matters: a decision you made, a person you mentioned, a constraint you’re working under. Most of what you say gets ignored on purpose. It’s not trying to remember everything, just what’ll matter next week.
Once a week, the Consolidation Engine goes through what the Observer flagged and decides what’s worth keeping, what merges with something it already knows, and what gets dropped. Your own brain does something similar overnight: most of a day gets forgotten on purpose, and a few things get filed away for good.
Ask Novalis why it said something, and the Why Layer gives you the actual answer: what it remembered, what it pulled in, what it left out. Not a fresh guess made up on the spot. The real record, every time.
Every Sunday, the Changelog turns that week into a plain-English summary: here’s what I learned about you this week. Got something wrong? The Dispute Function lets you correct it in one click, and it stays fixed.
Why this is different.
Every serious AI now claims some version of memory. Almost none of them let you see why it believes something about you, and none of them let you correct one specific wrong belief without wiping everything else out.
Novalis isn’t betting on remembering more than everyone else. It’s betting that being wrong about you, and admitting it, is the part nobody else bothered to build. A system you can correct is a system you can actually trust, and trust is the only thing that makes months of memory worth handing over in the first place.
What’s coming.
V1.5: Sharper memory.
Retrieval that weighs what actually matters over time, not just what you said most recently.
V2: A mind that meets you where you are.
Novalis starts noticing not just facts, but patterns: how you think, how hard you push, what kind of explanation actually lands for you. It adjusts: more grounding when you’re moving too fast, more structure when things feel scattered, your kind of explanation instead of a generic one. It can act on your behalf across the tools you already use, and watch for what’s useful before you ask. None of it happens in the dark. Every adjustment gets a reason, sitting in Personalization, waiting for you to keep it, tune it, or throw it out.
The one rule that doesn’t change.
Novalis is allowed to know you better over time. It is never allowed to know you better than you know yourself, without telling you how.
Join the waitlist, and you’ll be first in when the Open Beta opens.
Join the Waitlist