We've shipped a lot into A.L.I.C.E. over the last few days, and the changelog is only telling part of the story.
If you've only been glancing at version numbers, here's the short version: the npx package is safer, smarter, and much better at fitting into the runtime you already have. And for cloud-hosted users, Mission Control is moving from "interesting direction" to something that feels much more like a real always-on product.
Here's the fuller update.
One of the first issues we found was that the installer could be too opinionated about runtime configuration. In the worst case, it could overwrite the default agent model in ways that didn't match the user's actual OpenClaw setup.
That is fixed.
The latest 1.4.x releases changed the package so that A.L.I.C.E. agents now:
openclaw.jsonThat sounds small until it saves you from a broken install.
This matters even more now that we're leaning harder into NemoClaw as the recommended production path. A.L.I.C.E. should adapt to the runtime you have, not force the runtime we guessed.
We also cleaned up an identity mismatch that had been bothering me.
The orchestrator used to show up primarily as "Olivia," even though the product itself is A.L.I.C.E. That created an odd split between what the product was called and what users were actually talking to.
Now the orchestrator:
That makes the experience much more coherent without breaking existing habits for people who were already addressing Olivia directly.
Another big step: Pro users can now enable the Mission Control Cloud add-on directly during install.
That workflow now bundles the pieces cloud users actually need:
This is important because "Cloud" can't just mean "we'll figure it out after you buy." If someone is paying for synced memory, backups, and access from any device, the install path needs to understand that from day one.
We're much closer to that now.
The latest package update, 1.4.5, is especially important for cloud-hosted users.
Mission Control Cloud nodes now report heartbeats not only to the hosted runtime, but also to the admin control plane at admin.av3.ai. That unlocks a more truthful version of fleet management:
That may sound like backend plumbing, but it is exactly the sort of plumbing that makes a hosted product trustworthy.
If you're paying us for "always on," "connect from any device," and "your memories are backed up," there has to be a real operational layer underneath those promises. Heartbeats are one piece of that.
We've also been doing a deep pass on Mission Control itself.
The big architectural shift is simple:
That sounds obvious written out. It turns out it's very easy for a product to drift into a messier middle state where some pages are live, some pages are seeded, some are Supabase-first, and some are basically theater.
So we've been pulling Mission Control back toward a cleaner model:
It's still evolving, but the direction is much stronger now than it was even a few days ago.
None of this is flashy in the "look at the new gradient" sense. It's product integrity work.
The installer is safer. The naming is more coherent. The cloud path is more complete. The hosted control plane is more real. Mission Control is becoming more trustworthy.
That's the kind of work that makes everything after it easier:
That's the bar now.
The next big focus is continuing to close the loop between:
That means more real telemetry, better tenant and fleet visibility, stronger enterprise onboarding, and a smoother hosted experience for teams deploying on NemoClaw in production.
We're still moving fast, but the recent releases weren't random feature drops. They were foundational.
If you've been waiting for A.L.I.C.E. to feel less like an interesting framework and more like a serious product, that's what this stretch of work has been about.
More soon.