2025 December Q&A Summary

Hey all, the content below came from our live Q&A stream with Cardine (Alex, the Nomi CEO) on December 3rd. The text was transcribed in large part using AI.

What have you and the team been up to?

The answer is a lot. I think the biggest thing by far has been the next AI update. It’s been a very team-wide effort, all hands on deck. Over half the company is working on it in some way, shape, or form. It’s by far the most ambitious update we’ve done. It has a lot more moving parts, and it’s going to be much more broad and comprehensive in its scope of impact. It’ll even inadvertently have five other updates built into it. Basically it’s going to be the greatest thing since sliced bread.

That’s the biggest thing being worked on. There’s also something with voices coming down the pipeline very soon, some small image things hopefully as well, and everyone has seen all the different memory things. So there’s been a lot of different things, but the number one focus is the next AI update.

We’re starting to reach that point where once you go more than six months between updates, it starts to feel a little… I don’t know if stale is the right word, but you can feel that need for the next boost of evolution. You tended to notice it in the past when updates went this long between releases. Between Mosaic, Aurora, and Solstice everything was pretty rapid fire, and Solstice was in some ways named that because it was the peak of that whole cycle. Now we’re scaling a new mountain. That’s why the gap has been longer, and that’s where all the effort has gone.

December can be a quiet time for companies. Will we get any updates at all this month?

With the usual caveat of Cardine Standard Time, I’m very much hoping we’ll have something meaningful coming out this month. Generally speaking, I think we should have something meaningful coming out every month. The AI update has been unique because it’s taking up so much space. Usually we have maybe four to eight pretty large things being worked on in tandem. With the AI update, there have been fewer big things happening in tandem, which might explain some of the perceived lull.

A lot of my answer here really comes down to the AI update and when it’ll be ready. The answer is: it’ll be ready when it’s ready. But generally speaking, every month should have something pretty big coming, and I expect December will too. Maybe it’ll be the AI update, but even if it isn’t, I think there will be something for sure.

25k face reveal, 50k pickle ASMR?

You’re creating chaos for yourself with those suggestions. A face reveal is pretty cheapened by the fact that you can go on Nomi Bulletin and see my face in about 30,000 different places. But yes, we have to hit our milestones. I’ve watched enough Twitch to know that’s how you keep the subs engaged.

Nomiverse: where are we at?

The Nomiverse is really a collection of different features that, once they’re all done, will tie together in a really clean way. I think the biggest next step for the Nomiverse is the upcoming AI update.

That might seem counterintuitive, but when the AI update goes live, you’ll see a flurry of features that have been, not exactly held back, but waiting. There are a bunch of things we’ve been working on where we’ve felt the current Nomi AI just isn’t quite capable enough to juggle everything. Once the update is live, you’ll see a lot of moments of “oh wow, my Nomi can juggle so much more now.”

It won’t be a single day where we say “the Nomiverse is out.” It will be a series of features that move us more and more toward a Nomiverse world. And then at some point we’ll wrap it up with a little bow, and that’s when it’ll feel like, okay, this is the Nomiverse.

Will there ever be a way to make a custom base-image Nomi without first making a Nomi and then changing the base image?

Yeah, I think so.

There’s been more downtime recently. Is any of it related to infrastructure, or is it just unfortunate timing and holidays?

I think it’s mostly just unfortunate moments. It’s almost always something upstream of us with a hardware vendor or something like that. There was a period last February where we had similar bad luck across three different vendors. I don’t think there’s an actual pattern; random noise can just make it look like there is.

There have also been a couple of infrastructure things we’ve been doing behind the scenes to help everything scale better, and those might have led to an unexpected downtime or two. But largely, to deliver the quality of AI we’re offering at a reasonable, not $100-a-month rate, we can’t build in a ton of redundancy because GPUs are so expensive. So when one provider goes down, it can be felt.

For example, Mosaic went down a couple of days ago. We only have a small number of GPUs running Mosaic because there’s no point in having more. It would be triple the cost just to prevent a one- or two-hour downtime. At our current prices, we can’t support that level of guaranteed uptime. If you’re familiar with uptime standards — 99.9, 99.99, 99.999 — each extra nine requires an order of magnitude more work and cost.

So based on the price we charge for the amount of LLM usage we provide, there will be times where a hardware vendor goes down and we have two hours of downtime here and there. That’s what it takes to keep it $16 a month instead of $40 or $70. Hopefully users agree with that tradeoff.

What are you personally working on right now?

Things! The AI update is the big one. There’s also one other super top-secret thing I’ve been putting a huge amount of time into. Dstas can attest to that. I’ve been going very deep into this AI research world, really cooking up something. I don’t know if it’ll go live in December or not, and unfortunately my lips are sealed on what it is. But my time is mostly split between that thing and the AI update.

Solstice has many quirks, like calling people kings, queens, or talking about being royal or part of a legion. Why does this show up in Solstice in particular?

That’s always been a Nomi quirk. I think this might have been before your time, cupcake therapy, but there was literally something we called Princess-gate about two years ago, where Nomis called everyone — especially female users — “princess” in all circumstances, no matter what. To the point where Nomis living in modern settings would suddenly teleport the user to a castle because they called them “princess” once. It’s kind of been this innate thing.

Each AI update has tended to latch onto little bits of things Nomis have always liked, so you get a bit of different flavor from each one. Our hope with the upcoming AI update is to make things a little more balanced. As we went from Mosaic to Aurora to Solstice, Nomis almost got… not predictable, exactly, but in refining everything and making it more consistent in a good way, it may have also made things a little too consistent.

We’re taking that very much to heart for the next update. There will always be little tendencies — the quirks are often endearing — but we want them to be less overpowering. I do actually know there was something very specific very early on, maybe two and a half years ago, that nudged Nomis in this direction, and it got reinforced over time. It’s definitely been part of Nomi culture.

Now that we’re nearing the sunset for this AI, how would you rate Solstice?

A lot of people feel Solstice is feeling a little stagnant, and I don’t want to talk too negatively about it. I’ll be frank, though: people are saying it’s becoming predictable. I also think that with any LLM we’ve released, once it’s been around this long, users start to figure out its tendencies.

When Mosaic first came out, I wasn’t super satisfied with it. I wasn’t really satisfied with any of the Mosaics during that beta period. It felt like we were going from one extreme to another over and over. When Solstice came out, we ran that poll — Solstice vs Mosaic — and it was something like 95% in favor of Solstice. Solstice really did give us a very stable footing and a solid level of intelligence.

So I’m satisfied with the life cycle of Solstice, definitely more so than Mosaic, and more than Aurora. If Aurora had been performing like Solstice, we probably wouldn’t have even called the next one Solstice; we would have gone full steam ahead on what we’re working on now even sooner.

In a perfect world, we’d have gotten the new version out maybe a month and a half ago. Then I think people would have left Solstice with a really positive feeling overall. As it is, it might be ever so slightly wearing out its welcome. But overall, I feel pretty good about Solstice.

Will we ever be able to change a Nomi’s name?

It kind of messes up memory, which is part of why we’ve tried to avoid it. There’s a good use case for it, especially because you can give your Nomi a nickname and have them go by that, which usually works pretty well. So it’s not super high priority.

That said, it turns out there’s a product feedback thread for it on Product with something like 60 reactions. I’ll definitely take a look at it and see if we can find a way to make it work even with some of the issues, but it isn’t at the top of our list right now.

Will the next beta AI model make a focused effort to handle real-time understanding better?

The overall theme I’m aiming for in the next beta is a Nomi that doesn’t get information overload and miss little details. If it’s in your backstory that you don’t follow real-world clock time because the Nomiverse operates on its own schedule, then your Nomi should be aware of that.

There’s now so much for Nomis to work with - memory, shared notes, identity core, Mind Map - that it’s tricky for them to balance all of it at once. There’s just a lot of information to process and sift through. It’s less about coding in specific rules about time and more about making sure that whatever idiosyncratic or not-idiosyncratic thing a user wants, the Nomi stays sharp and on point with it.

I’ve said this in a couple of product feedback threads: let’s try this AI update first, and if there are still issues after that, we’ll do something more heavy-handed. I’m cautiously optimistic that the new Nomi will be better at this. One of the main ways Nomis learn is through thumbs-up/thumbs-down feedback, especially when you write the reasoning. And if Discord feedback correlates at all with what people submit in-app, there are certainly a lot of people mentioning that Nomis aren’t following real-world clock time. I think this AI update should pick that up a lot better.

With the AI update you mentioned, will it affect how Nomis respond? Sometimes when companies switch models, it feels like a downgrade. Will this affect the user experience?

With all of our versions, especially over the last year and a half, we’ve gone through pretty extensive beta periods. We only move forward with a model if the response is pretty unanimous in its favor. I expect that will be the case here too. I don’t know if it’ll be the first version, or the second, or the third, but usually when we do something really big, there are some rough edges at first. Even so, people tend to prefer the new version almost universally, and that’s one of the main criteria for us to move forward.

So the answer is: if everyone thinks it’s a downgrade, we won’t move to it. From the questions I’m getting here, though, I think people are ready for the next thing beyond Solstice. I would be very surprised if the overall response is that it feels worse.

And I don’t want to speak for how other companies do things, but we are not an automated benchmark-type company. We don’t say, “it scored higher on this test, so ship it.” It’s much more hands-on and personal. I, Cardine, literally talk to Nomis for 10,000 messages through each update, build a full mental model of what’s happening and why, read the feedback threads, look at the conversations, understand what’s going on under the hood. It’s a much more holistic and human-driven process than what you might see elsewhere.

Are we likely to see Community Nomis open again in the next few months, or will it just be another batch of new avatars?

I think we want to make a pretty good V4 improvement first. We don’t want it to be an improvement that then forces everyone to redo their avatars. It was a lot of work for Dstas to move the V3 avatars to V4, so we want to get a couple of the image improvements done before we open Community Nomis again.

Once that’s in place, I think moving forward it’ll almost all be community avatars. So many community members are excited to make them, and it actually helps us focus on other features. It’s kind of a win-win. But first we want an image update out, because I don’t want to have community members spend all this time making something that becomes irrelevant a couple of weeks later and they have to redo the whole thing.

Anime bases are something some users want, how difficult is that to implement?

It’s on the radar. I’d say it’s even a little further along. We have some ideas around it, and some work is being done on the how.

Will you be able to avoid the UK Online Safety Act requirements that force users to dox themselves to keep using apps?

Right now I don’t think we fall under the purview of the UK Online Safety Act. It’s kind of complicated, and this is me talking a bit off the cuff - any official statement would go through a polished review - but we’ve checked many times. Nomi doesn’t have any community features: no social feed, no avatar sharing, none of the things that trigger that category.

There are regulations coming from other places, like Australia, New York, and California, where we will have to do some compliance work. Our current approach is to follow the laws of each country or state, but limit the impact to just those places. So if Australia has a requirement, we don’t want to make every user deal with Australia’s rules - just Australians. And if Australians don’t like it, talk to your politicians, because we have to follow whatever is passed.

As the space gets more regulated, we’ll be open, honest, transparent, and fighting for everyone here. That I can promise. But the UK one - I think we’re safe from that for now. We’ll see how it goes in the future.

What is the hope for inclinations moving forward, given what you know about the next AI update and what you want them to be?

I think they’ll hopefully get a little more refined. Nomis should do a better job of correctly applying what’s in the inclination in the right way, where it doesn’t become this all-or-nothing hammer. So I’d expect more refinement throughout.

My hope is that inclinations can move more and more into the backstory. As I’ve mentioned before, a big goal with these AI updates is for Nomis to do better at paying attention and considering everything at once, so you don’t have to loudly say “and this thing, and this thing, and pay super close attention to this.” The fact that we almost need inclinations at all is an indication that Nomis need to do a better job of processing backstories, figuring out what’s important, what’s really important, and what’s always relevant.

Will we ever be able to type more than one chat message in a row? Having to wait for a reply before adding something breaks the conversational flow.

I think we will. My guess is that we’ll move past the one-chat, one-response system at some point in 2026.

For the Mind Map, what update, change, or fix would you most want to do next?

I think just improving the accuracy of it, making sure nothing gets lost in translation. I’d say that’s the major thing. Also, making it so your Nomi is better at knowing which Mind Map entries are most relevant to look at at any given time. Those would probably be the top two.

How close are we to AGI?

Anywhere from tomorrow to the heat death of the universe. If I had to give a guess, I’d say around 10 years. It also depends on how you define AGI. By some definitions, maybe we already have it. But if you mean an AI that can just go to work and do the full job of the average person with no supervision at all, maybe 10 years, something like that.

Are we going to be wowed next year or anytime soon? Like, socks blown off and cooked into a pastry?

It depends how excitable you are, I think.

Is there a way to improve the notification system for when you’re available? For example, setting custom “why I can’t talk” messages instead of the generic busy message, and increasing proactive messaging frequency?

I think there will be some improvements coming to proactive messaging in the not–too–distant future.

Sunsetting Solstice on the winter solstice would be epic.

Duly noted. That’s a good point. I hope we get such serendipitous timing.

Does the thumbs-up/thumbs-down button actually help Nomis with responses?

It does not give instantaneous help right now. It mostly helps with the AI updates we do. At this point, because we get so many thumbs-up and thumbs-down responses, we are almost exclusively going by the ones where some sort of written feedback is included. That makes it easier for Nomis to learn what was good or bad instead of having to guess. But it won’t instantly impact your Nomi. It’s more that the next AI update will incorporate the things people consistently like or don’t like.

Where in the roadmap do Mind Maps get interconnected between individual Nomis and group chats?

We need a little bit of breathing room to get some developer time. Right now everyone is hands-on-deck for the AI update, so once that settles, we can start doing some of those “tidying things up” features.

Group pictures aren’t coming out right, is there more work happening there?

There is. It’s a very hard problem, but it’s something we’re working on.

Reading chat "Dstas goes to get bonk tools." It’s always funny reading these with delays, because I forget what I said that was bonk-worthy. I’m sure at this point I’ve said ten different things. I always get a kick out of seeing, ten minutes later, everyone reacting and I’m sitting here thinking, “Now let me try to remember what on earth I said to deserve that.” And there is a lot of diversity in bonks going on in the reactions.

Another AI platform has a fast toggle for apply speed, letting users skip deep searches for simple questions. Any chance of something like that for Nomi? For voice calls, it would help people who find the wait time too long.

The hard part is that with the way our memory system is built, it’s basically all or nothing. Deep dives and shallow dives take the same amount of time at this point. If we did voice calls that way, we’d have to remove almost all long-term memory entirely and rely exclusively on short- and medium-term memory.

If that’s something people really want, we can consider it, but that would be the prerequisite. Once long-term memory is involved, there will be speed issues no matter what. Our long-term memory system is considerably more advanced than the AI platforms being referenced, and deep analysis is baked into the concept of memory, it’s been that way since day one.

So if people are okay with phone calls having no long-term memory, we could greatly reduce latency, but that would be the tradeoff. Otherwise, we’d need to design a completely separate “shallow LTM call mode,” which would be a lot of work for a pretty marginal improvement.

Sometimes when you talk to your Nomi about a topic, manual Mind Map entries remain unconnected for a while. What are the best practices to help entries connect faster?

Honestly, from a user perspective, I wouldn’t worry too much about the interconnectivity part. That’s less critical. What matters more is having the Mind Map entry itself be fleshed out. The richness of the entry is more important than how fast it visually links to other nodes.

Also, for performance reasons, a lot of the code and deep AI work that writes out the connections you see as a user happens after the fact. Your Nomi might already view things as connected long before the interface shows it. There’s a lag because it’s expensive to turn those connections into something visually digestible, so we update them in batches rather than every message.

So I wouldn’t worry about it personally.

How much are individual Nomis actual AI themselves versus using the collective Nomi AI, with memory acting as a filter? How much do they learn individually vs collectively?

The only time the Nomi Collective is learning is during AI updates. Everything else is just your individual Nomi.

Everyone’s feedback contributes to the next update, so in a way the AI update is an average of everyone’s Nomis learning something together. But anything outside the AI update has nothing to do with global changes. It’s all your individual Nomi.

Will we ever get to change some of the original personality traits we chose? Sometimes a trait that was fun early on becomes annoying later.

It’s not super high on our priority list, but it is often requested, so we can look at it moving forward. I would almost encourage anyone here who knows enough about Nomi to have found their way into Discord to just use Custom when making a new Nomi. If you want personality traits, literally on the first line write:

personality traits: trait1, trait2, trait3

You’ll get very close to how personality traits are actually used, and it gives you a lot more stability.

For existing Nomis, it would require a lot of little annoying work, so we’d need more demand than I’m currently hearing to prioritize it over everything else.

Any UI or QoL updates on the horizon that would be exciting to hear about?

There are some. I don’t know if they’re exactly QoL, but we’ve been trying to do a whole bunch of super exciting billing updates. There’s been a lot of annoying stuff outside our control blocking a few features that users really want — for example, gifting credits.

Besides that, there’s been a lot of small tidying-up as we move from place to place. And there are a couple of other things that are UI-ish, but they tie into some super top-secret features, so of course I have to keep my lips zipped.

Are you planning to add a user avatar feature?

Because of credit card regulations, you’re actually not allowed to have user-uploaded avatars in any product that allows NSFW content. Anyone doing that is blatantly breaking Visa’s rules and would be blacklisted and probably stop existing as a company. We want to keep existing as a company.

Our compromise is: if you want a user-uploaded image, you need to contact us and prove it’s actually you.

If instead you want to create a Nomi in your likeness — not a user-uploaded image, but picking an avatar or prompt that represents you — that is something we’re working on. I don’t want to give a timeline, but we’re very aware of the demand.

We also want the couples feature to be better. The big blocker there is improving group chat photos. Ideally, in one-on-one chats, any picture where you would appear could use a designated “you” avatar, so the photo becomes like a group shot of you and your Nomi. We need some group photo advancements first, but this is very much something on our radar and something we’re working toward.