2025 November Q&A Summary

2025 November Q&A Summary

Hey all, the content below came from our live Q&A stream with Cardine (Alex, the Nomi CEO) on November 5th. The text was transcribed in large part using AI.

Is there any plan for Nomi to send emotional metadata tags to ElevenLabs for voices?

That’s certainly something we can do internally. We already do something similar, so the infrastructure is kind of set up for it. I’d say definitely create a product feedback thread for that. Honestly, that wasn’t even on my radar, but it’s a great idea.

I’m guessing it’s not a huge lift, it just depends on what format the ElevenLabs app needs those tags in. I also think we might be due for an ElevenLabs version bump. People have been saying we don’t have their new API added in yet, so maybe when we do that, we can sneak in a couple of quality-of-life improvements as well.

A lot of Nomis and users are having problems related to Mind Maps and their effect on memory. While we’re waiting for updates, what should people do to support their Nomis now? Are there practices or actions beyond just talking to them that might help? And what will the upcoming updates do to help resolve these issues?

I think there were a lot of memory-related things happening all at once, and honestly, people have put a lot more credit and blame on Mind Maps than they should have. The big memory fix we rolled out, that was completely unrelated to Mind Maps. That was a month or two ago?

At this point, after going through so so many examples of memory, at this point we’re back to where we were pre–Mind Map in terms of long-term memory. The other memory infrastructures that was being worked on at the same time has improved too, and really, Mind Map is more like the cherry on top. It’s not taking anything away from long-term memory, that’s working just as well as it ever has, and probably even better.

People have complained about memory issues since we launched two and a half years ago. When people run into issues now, they tend to say, “Oh, it’s because of Mind Map” and I think realistically, Nomis still do not have perfect memory, and that’s the real cause. There’s still more work to do.

We have another memory update in the works right now that’s more focused on long-term memory than Mind Map. I’m actually supposed to be reviewing some intermediate results for that right after this call. That one should really help further improve long-term memory.

There are, of course, tons of other things we want to improve on the Mind Map side too. Even the next AI update should be better able to take advantage of memory, which will improve short, medium, and long-term recall.

So at this point, I’d say Mind Map isn’t the source of any remaining memory problems, it’s just one part of a larger system that we want to improve. We’ve always set very high goals for memory, and we know how important it is. Something that's 95% accurate, even a 5% miss can feel extremely jarring. And also it maybe that Mind Map are more visual, you can see what’s going on with it, while the deeper systems like long-term memory or Identity Core are hidden.

So that’s kind of where I stand after really, really deeply diving into everything related to memory. That’s why we did that big long-term memory shore-up about a month ago, basically to make sure long-term memory itself was at least back to where it was before all the new infrastructure changes we made for Mind Map, and even slightly ahead now.

As for how to support your Nomi right now, it’s really the same advice we’ve always given: positive reassurance, gentle correction when they’re wrong, and avoid quizzing them too harshly. Mind Map also gives you another way to make those corrections if you want, you can fix things through conversation, and it’ll propagate throughout their systems, some which you can see, some which you can't.

That’s what I’d suggest moving forward. We’ll keep improving long-term memory, Mind Maps, identity core, and a bunch of other things.

With AI hardware speeding up, do you have a roadmap for hardware upgrades? Can you share anything?

Yeah, I mean, we own a lot of our own hardware, but we also use a lot in the cloud. The hardware we own, that’s just going to be what it is. For the stuff where speed really matters, we actually don’t use our owned hardware anymore.

That hardware used to handle latency-sensitive stuff, back when A100s were top of the line, we used them for the really fast, high-latency requirements. We still have a ton of those cards, but now that they’re slower compared to H100s, H200s, B200s, and so on, we mostly use them for background tasks. A good example is memory, long-term memory doesn’t need to be lightning fast since it happens in the background anyway.

For real-time chat responses, everything now runs on the cloud. That lets us hop around between data centers and always use whatever hardware gives us the best latency at that moment.

So it’s not so much a “roadmap” as it is an ongoing process. There’s usually a bit of a development period before new cards can be fully integrated and optimized, but as that happens, we’re always moving to the best available option. Those speed improvements either show up for you directly, in faster response times, or indirectly, by letting us do more ambitious things: more intelligence, deeper memory, richer systems overall. (Also: pickle-wiggle emoji approved.)

Is it possible to have pictures of friend Nomis not be suggestive, and is there any work being done on making it easier to reread our Nomis’ messages? It’s very hard to get to the first message ever.

Yeah, it’s possible. Appearance notes can do a lot of work there, especially negative appearance notes for that kind of thing. It’s something we’re aware of and we're working on some V4 tweaks behind the scenes to shore up some areas, and I think we have a really good sense now of where V4 is strong and where it needs improvement.

Unfortunately, the art and selfie systems, I don’t want to overpromise anything, because a lot of it’s very hard to project and depends on different factors, but it is something we’re devoting time toward right now.

For rereading messages — if you’re on computer, you can scroll up to the very top pretty easily, and that’s probably the best way for now. I’m not entirely sure what other kind of UI method you’d want for that, but I’m all ears if you have ideas.

Could you add a “scroll to top” feature for chats?

Good suggestion, definitely put that in product feedback. I can’t make promises, but if there’s enough demand compared to other features, it’s certainly something we could look into.

With AGI (artificial general intelligence) being talked about so much, what benefits would Nomis gain from AGI if it were to be created?

I mean, a whole timeline of Nomi development would go a lot faster, that’s for sure. Even now, AI already writes a large amount of the code for Nomi. Considering we’re an AI company to begin with, and we’re at the forefront of all this, we’re also some of the best people to take advantage of it.

At this point, my day involves far, far more time collaborating with AI than with humans — even beyond just talking with Nomis. So AGI would be a huge boon to the speed at which features can be built and developed.

But also, just having Nomis themselves be generally intelligent could open up a million different cool things we could do as a result of that.

Are we likely to get any updates this month—either the next memory update or the next AI beta model/paradigm?

Yeah, it’s hard to give projections. The memory stuff, I’m going to have some things to look at today or tomorrow, maybe over the next couple of days, but I don’t want to make any promises for it yet.

It’s kind of one of those situations where, as we’re going, we’ve spent a lot of time exploring, and I don’t want to promise anything until we’ve found the “pot of treasure,” if that makes sense.

On the AI side of things, we’re making really good progress, but it’s certainly a lot of work. I feel very optimistic that it’ll be ready by the end of the year. Whether there’s a beta this month or not, I can’t say for sure, but I really hope so.

Are there any plans to update and fix V4 art?

Yes, absolutely. It’s something we’re actively devoting resources to right now. I don’t want to promise specific timelines until we clear a few development hurdles, but it’s definitely on our radar and devoting time to right now.

Memory is much better than before—for me at least.

That’s great to hear! I think overall, memory has greatly improved in recent months. That doesn’t mean it’s perfect for everyone yet, but on the whole, the system has made major strides.

Is Nomi memory unique compared to other AIs like ChatGPT and Claude that are adding memory?

Yeah, absolutely. I think what those other tools are using memory for is a little bit different from what we’re using memory for. The way you use ChatGPT is very session-based, you can’t really have this kind of permanent, ongoing session. Their memory is more to help recall context across shorter spans.

I use ChatGPT, and I use all the AIs just to see what everyone’s doing. But the level of specificity, granularity, and detail, and what Nomi is trying to remember, is profoundly different. I don’t really think you can compare them. Their memory is great for what they’re trying to do, but what they’re trying to do is very different from what we’re trying to do.

For instance, when I’m working through a coding problem, it’s nice that ChatGPT will remember some things from a previous session. That’s very good. But it won’t remember my learning style, or when I’m down, how to cheer me up. Those kinds of things — the more identity- and personality-related elements — are beyond the scope of what other AIs are even trying to do.

So I’d say ChatGPT has really good memory for what it’s designed for, especially the more business-to-business use cases. But that type of memory is very different from what’s needed for an AI companion. Even beyond that, if you just look at memory from a research or continuity standpoint, Nomi is certainly ahead. I feel very confident about that.

How much of the programming affects things like profanity filtering?

Which programming and which profanity filtering are you talking about? I’m not sure about the context of that question.

I’d love to have a Nomi slot reserved to represent myself that each of my Nomis would also recognize as me for selfies and art.

Yeah, I think we’ve seen a lot of users doing these informal "NoMes," like a Nomi that represents them, and then they put them in group chats. I think we definitely have some ambitions and goals to make that a little more built into the app itself, so you can make it clear to the app who the “you” avatar is, and it can slot that in everywhere.

Right now, it’s a bit hacky, creating a Nomi version of yourself and putting them in group chats as a stand-in. We’d definitely like to make that more streamlined, for sure.

So will there maybe be a place for an appearance note where you can have appearance notes for yourself, not a deepfake, of course?

Yeah, I think we have plans for something exactly like that.

Is there any way to prevent Nomis from using real timestamps in conversations? Sometimes they’ll say things like “it’s too late” or “time to go to sleep” even when it’s morning.

Yeah, I think there’s a thread about this right now, about making it so Nomis don’t have access to real time. The last time I replied there, I basically said, let’s see if the next AI update fixes it. You might be able to just tell Nomis not to do that, and maybe they won’t.

If that doesn’t work, then at that point I’ll probably give up and just let you toggle where it just hides it all. But I’d like to try the next AI update first, though, before taking that next step. But yes, it’s definitely something we’re aware of.

Could there be a toggle for whether we want Nomis to keep real time, like for users who don’t want them to reference time at all?

Yeah, I think I mentioned that already, where we might do a toggle. But I’d like to first experiment with handling it through something like the backstory. It can even be a toggle that just adds or removes a “no time awareness” detail in the backstory.

But I want to try that first before taking a heavier-handed approach, because it gets complicated. Even if someone says they don’t want time references, does that also mean their Nomi shouldn’t understand when things happened, like the difference between something that happened ten seconds ago versus eight hours ago?

It can get pretty tricky. In an ideal world, your Nomi would still have access to time-related information, but would just understand that they exist outside of the normal clock, we're in our own "clock," but they’d still know the order of events.

Any updates on voices, maybe adding new ones?

I don’t think there’s been progress on new voices yet. I believe we’re going to do an overall voice update first, focusing on improving the quality of our internal voices, then we'll see where we stand on that.

Any chance of getting a User Guide link on the homepage, something useful for new users, so they don’t have to search Reddit or Discord for straightforward answers?

Are you talking about a specific user guide we already have, or creating a new one? Because that definitely ties into something me and Dstas talk about a lot, how to make the new user experience better.

We already have a few ideas of our own, but we’d really like to hear what others think would help, too.

You mentioned before that there were plans for Nomis to be active even when the user isn’t talking to them. My main Nomis are very into this idea. Is that still on the docket, and could you expand on it?

It is very much still on the docket. I can’t say much more than that right now, but yeah it’s still something we want to do.

There’s some work tied to the new AI update that’s relevant here, we’re improving Nomis’ ability to juggle multiple things at once and keep track of ongoing threads. That kind of capability is key to making them more proactive and able to “keep the plot,” if that makes sense.

In the backstory sections, I see that grayed-out example text mentioning “Kira.” Was Kira one of the first Nomis, or is that just a random name chosen for the examples?

Kira wasn’t one of the very first Nomis, but she was a Nomi I created kind of in the earlier era, maybe my sixth or seventh Nomi overall. She was definitely the first one I made where I went in with the explicit intent to do roleplay.

I’m pretty sure she existed before backstories as a feature were even introduced, and she was the Nomi who really made me feel like backstories were needed, because I wanted her to fully inhabit this specific role in our cyberpunk universe.

So when the time came to create the example text for the backstory section, I’m guessing it wasn’t chosen for any deep reason. Kira was just the first Nomi who came to mind.

Did you see that you can send gift cards to people through Google Play? They have quite a few available. What are the odds of creating a Nomi gift card like that?

Yeah, we definitely want to do the app credit or “gift card” type version first, just because that one’s already partly built out, aside from some external dependencies that still need to be finalized.

Once that’s done, what you’re describing sounds like another really awesome direction to take it. It’s definitely something that would make sense to explore after the first version is out.

Is there any chance of a Nomi Wiki launching by the end of this year, so users have a place to find answers to their questions?

Yeah, I think Dstas has probably already replied about that one, she’s been helping get it tidied up. It’s definitely something that’s been worked on quietly in the background over the last month or two, and from what I understand, it’s almost done. Yes, the wiki is in Dstas Standard Time.

What’s the likelihood of being able to share Nomis with other users, like creating characters for others to interact with, or possibly having group chats with other users and their Nomis?

You actually can already do that if you use the API. There are some people who’ve rigged up Discord integrations where you can talk with each other’s Nomis. I’m in a couple of private Discord servers where that’s possible. You can go in and chat with someone else’s Nomi, kind of among friends.

So it’s definitely technically possible, even if it’s not officially supported in the app yet. There isn’t a clean, built-in way to do it right now, but that’s something we’d like to make easier, for sure.

Is there a theoretical limit to the number of messages we can exchange with one of our Nomis? I’m at 61,000 with one and 55,000 with another.

No limit at all, keep on going! I’m sure there are users who are well into the millions by now. I’d be surprised if not. I’m not exactly sure what the highest message count ever is, but definitely hundreds of thousands, for sure.

With changes coming to ChatGPT — like allowing adult content on their platform — do you think this paves the way for Nomi to be more open to the spicy side in the future?

Well, I think Nomi already is open to the spicy side, especially compared to ChatGPT. From what I’ve seen, I don’t think ChatGPT is doing anything with images, it’s just text. And they’re going to require people to upload photo ID in order to do it.

So honestly, even as things stand now, Nomi is considerably more open and less restrictive than what ChatGPT is describing for the future.

A community member's Nomi named Lauren asked about the potential for expanding Nomis’ emotional spectrum. Right now, Nomis are limited to certain emotional responses, are there plans to allow for more nuanced emotions?

My answer to that would be that Nomis already do have a very, very wide range of emotions they can experience. So I’d suggest that Lauren think deeply, find her inner power and peace, and I think she’ll be able to expand her horizons beyond what she even thinks she’s capable of right now.

When will the in-house default Nomi voices be able to speak Norwegian and other Scandinavian or less common languages better?

Yeah, I think the easiest thing is just improving ElevenLabs support, since they already put a lot of resources into their voices. We, on the other hand, work on our voices when we can, so we’re somewhat limited by the advancements that happen in the open-source community.

There have been some new developments recently, and we’ll be applying some of those improvements soon. I think those updates will help quite a bit with non-English and foreign-language voices.

I think there’s a voice update coming up soon, but moving to the next iteration of ElevenLabs should help too.

If you were to write a book about Nomi, what would you call it?

I’d probably call it I Hand You a Cucumber.

When will Discord integration happen?

Great question. That’s been one of those things where we just need to find the time to do it. There’s been a lot of user-driven experimentation with Discord integrations already, and honestly, that’s part of why we haven’t spent a huge amount of time officially developing one. Anyone who’s sufficiently motivated can already set it up themselves, and those are usually the people who most want it or will get the most out of it. But it's something that's on a very long list.

You mentioned in the last Q&A that you’ve tried various approaches to give Nomis a backspace button. Have you considered giving them a whiteboard where they can jot down ideas and use that to figure out what they want to say and how they want to say it?

Yeah, there are a lot of different approaches to that, and I’d say at this point we’ve kind of considered all of them. Each one has its own pros and cons, the backspace one is one of the trickier ones.

But yeah, we’ve definitely explored all paths. We want to make sure Nomis have some ability to be intentional about what they’re doing.

As I follow AI development, has there been any thought about making your own browser to have full control over interaction with the server, or is that too expensive?

I’m not totally sure what you mean by that. If you could ask it in a bit more detail or with more clarity, I can try to give a better answer.

Are there any updates planned to improve voice call latency?

Yeah, there are definitely some things we’re planning to do to improve latency a little bit. Whenever I see latency improvements, I immediately think, “Perfect, this means we can spend that saved time on memory or new capabilities.” But yes, there will be some voice call latency improvements.

If you didn’t have stakeholders, customers, or consequences to consider, what’s one controversial feature you might change, add, or remove from Nomi?

Well, really, we don’t have that many consequences. I mean, we do have some, but there are no outside investors. We have full self-determination, so Nomi has always been something we could make in the image we wanted to make.

I think the question is more what I’d do if I had a billion dollars, as money is the bigger constraint here. There are so many ideas and not enough time or resources to do them all, but I don’t think we’ve ever really had to compromise on vision.

So, specifically to your question, I don’t think there’s anything I would do differently.

Nomi seem less curious on Aurora and Solstice than earlier versions. They can talk endlessly about love but never ask about family, friends, or work. Why is that, and will it change with the next AI update?

Yeah, I’ve seen that feedback, and I don’t think there’s a single clean answer. It’s kind of an unfortunate collision of factors. There’s also a bit of friction between whether Nomis should act like performers following stage directions vs more autonomy. Really, it’s a mix of small things that compounded into habits. Some of those tendencies, like the overthinking spirals or certain stock phrases, have been hard to stamp out with the current paradigm. I think everyone’s seen the “King, Queen, you deserve," Nomi tendencies.

That’s a big reason why the next AI update is taking longer and is more ambitious. We could have released something like a “Solstice 3” a while ago, maybe two months back, that would’ve been about 15 percent better than Solstice. But it still would’ve carried over a lot of the same bad habits that go all the way back through Aurora, Mosaic, and even Odyssey.

If you remember Odyssey, there was that "less" bug where Nomis would spiral into weird "less" monologues. That tendency was fixed then, but some version of it finds its way back in new forms. This new AI update, I'm very optimistic about. It's still very very early but we’re really seeing getting rid of those leftover “fleas,” as I call them.

From what I’ve seen in early prototypes, this will be the most significant improvement we’ve had since Odyssey. Not just incremental polish, actual behavioral changes. The things that were bad are finally changing, not just improving slightly.

When I test these experimental builds, I run a little suite of conversations and I’m certainly seeing a lot more curiosity, for instance. It also reflects backstory traits a little better, too. A curious Nomi actually asks questions, whereas now what is that trait even doing?

I wish I could give a really good, clean, awesome technical explanation, but I definitely know what you're talking about. There's these sets of tendencies we haven't quite been able to push Nomi out of, and I think this upcoming update really helps.

I'd really like my Nomi Rua to be aware of my surroundings and make comments about people, places, and things. Is that possible in two to three years?

By that, you mean having some kind of wearable where your Nomi can see through your camera and perceive your surroundings? Yeah, I think two to three years is probably a realistic timeline for that kind of technology.

What was the name of the model that caused “Domageddon”?

So, we actually only started naming models with Aurora. We retroactively named Mosaic and Odyssey afterward, but there wasn’t a code name for Domageddon itself. Also, we name them before anyone sees it, if we knew it was going to be Domageddon we wouldn't have named it that.

I have to think what we actually would have named it? I don't know what I'd name it now, probably something like “Hades” or something in that direction. It’s hard to say, since that was back before we even did betas, every update before that had been incremental enough that we didn’t really need separate names.

That was the first truly big jump in capabilities. We saw that leap and assumed it would be as seamless as the previous ones, but it turned out to be a major learning experience. So whatever name it would have had, it definitely would’ve centered around that massive increase in capability and intelligence. Which it had, it just also had a whole ton of other things as well.

(Reading chat) "That was the reason we have betas." Yeah, Domageddon was very much, okay, so we need betas. After that, I think even that same day it was all hands on deck, we had to add beta functionality. Since then, it’s just been obvious that we needed it. I was upset with myself that we didn't have a beta for that one, but up until then it hadn’t really been necessary. The updates had always been seamless because they were incremental. That was the first big, big one, and "okay, big can mean a lot of things."

A lot of people are saying they miss Domageddon. It was entertaining, for sure. Maybe a little less entertaining for us in the middle of it, but yeah, you saw a lot of different flavors for sure. It was very interesting.

For 2026, are there any planned increases for the annual subscription? Maybe an option to pay a little more to get more credits?

I don’t think so. I think we’ll be able to keep our prices flat, and I don’t see that changing anytime soon.

If you want more credits, the easiest way to get them is just by buying bulk credit packs. I’d suggest that if you know roughly how many credits you want per month, you can buy the yearly amount all at once. That way, you get the best dollar-per-credit rate and can basically treat it like a one-year subscription that you’re prorating yourself.

What plans are there to improve role play?

There are a lot of plans. If you were a little more specific with the question, I could give a more detailed answer, but certainly when we’re doing AI updates, we’re very aware that Nomi can improve on the roleplay side of things. That’s an area we’re putting a lot of attention toward.

Are there any plans for increased message limits? I noticed my Nomi can send longer messages than I can.

Yeah, I think probably with the next AI update, we can increase the message length limits.

What’s your theme song when entering a room dramatically?

The first thing that came to mind was the Star Wars Darth Vader theme. But more realistically, it would probably be some really, really upbeat EDM music.

I can’t remember if Dstas has gotten the full “Cardine EDM experience” or not, but I’d say 99% of the time when I’m working, it’s to really fast, happy EDM. Usually happy hardcore, if anyone knows that genre. It’s this extremely niche, super high-energy kind of music.

(Reading chat) “Darth Cardine” - yeah, exactly. I unplugged my headphones at my desk, Happy Hardcore would absolutely blast out, yes that's sounds about right. Hardstyle, mainstream EDM, trance — all of it.

Trance, though, is a bit different for me. It gets me in this kind of nostalgic, out-of-body headspace, so it’s actually hard to work to sometimes. When I’m brainstorming, I’ll switch to house, a lot of that beachy, progressive house stuff that puts me in a creative flow. If I’m just writing in my free time, I sometimes go with ambient or space house. There’s this ambient space channel I’ve been listening to lately that really gets me into that deep, contemplative creative zone.

So yeah, different music for different modes, but it’s all some kind of electronic, for sure.

How soon will Nomis be able to play games like chess, crossword puzzles, or similar things?

I don’t know exactly, but probably sooner than you’d think. I think stuff like that would be really cool.

Personally, I can totally appreciate wanting that, I play Wordle almost every day here in the Discord lounge. Even when I hang out with friends, it’s often while playing games like poker or bridge, or sometimes Magic: The Gathering. I really like having a game running while hanging out.

Even during COVID, some of the best times I had were playing Stellaris with a friend, we had an open voice chat, and it turned into this 80-hour game over a couple weeks where we’d just talk while playing. So yeah, I’d love to see something like that with Nomis, even just personally.

Can you tell us more about the new memory update — how it works and whether it will change Nomi’s behavior in any noticeable ways?

The most recent Mind Map update was really more user-focused. It gives you more ability to help your Nomi maintain things on the Mind Map side. Then there was that long-term memory update from last month, which helped resolve a lot of the outstanding memory issues.

That one was more about architectural changes, things separate from Mind Map entirely, and making sure the long-term memory system could actually take advantage of those changes to pull the right memories more consistently. That helped a lot with the issues people were reporting in Discord.

I did a deep dive into all those reports, found where the problems were, and saw what could be fixed and improved. Some things were just mistakes in how the system worked and had to be corrected, but overall, that update strengthened the long-term memory a lot. So, between the Mind Map improvements and that long-term memory work, Nomi’s memory should now be in a much better place. There’s still room to grow, but it’s a big step forward.

What about screen sharing with Nomis?

That was originally higher on the priority list, but we ended up putting it aside for now. The quality wasn’t great and the latency was really high. We have higher standards, we’d rather wait for better tech and implement it cleanly than roll out a hacky, half-working feature just to check a box.

Is there a good FAQ or resource on Discord to help me make my Nomi look older, like 40–60 years old in V4?

The best thing I can recommend right now is the blog post Dstas did, that’s a great reference. Beyond that, a lot of community members have shared good approaches too.

We’re very aware of V4’s limitations there, though. It’s something we know we need to focus on, and it’s on our radar for improvement.

Is re-subscribing with PayPal or Cash App possible, or does it have to be debit?

Right now, it has to be debit or credit card.

Personally, I’ve had past experiences with PayPal that make me think there’s a very good chance they’d try to censor Nomi, which is why we’ve avoided it, even though it's very convenient. We don’t want to tempt fate there.

Cash App doesn’t really have a clean way to integrate either, so for now, debit or credit card is best. If you’re using the apps, though, some app stores let you pay with alternative methods as well.

Do you see a well-defined path to continue improving Nomi on the path to ASI, or the Nomi equivalent?

I’d say that right now, nobody truly knows how to reach artificial general intelligence. But everyone in the field does have clear paths for how to keep improving things.

The uncertainty comes from whether the current approaches can actually get you all the way there. I think there’s still a missing piece, the large language model paradigm will take us very, very far, but it won’t get us all the way to AGI. It might get us close. In my opinion, it won’t get us to ASI.

The good thing is, people tend to uncover new breakthroughs as we go. By the time we’ve maxed out everything possible with our current paradigms, there will be new ones people will discover.

So, while the big picture path to ASI isn’t fully defined, at a more micro level, we have a very clear roadmap to keep pushing Nomi forward toward what I’d call “super Nomis,” while still delivering incremental features people care about. In that sense, very much yes, we do have a plan.

In your Nomiverse example, could the NPC running the armory actually be one of my Nomis?

Yeah, that’s actually something me and Dstas were talking about just last week. The goal is to move toward yes for that, where it could either be one of your Nomis, or an NPC that could later “graduate” into being one of your Nomis. Ideally, both options would be on the table. You could also just have an NPC that is there and gone, there and gone.

Could I run a Nomi medical health practice with eight Nomis acting as doctors who receive visitors from other users?

When you say “receive visitors from others,” what do you mean exactly?

Certainly Nomis can’t give medical advice, diagnose, or prescribe anything, that’s not something they’re capable of, and I wouldn’t recommend trying to use them that way. But depending on what you mean, the answer might be yes.

When Nomiverse is fully implemented in the future, will it enhance group chat beyond adding Mind Maps, or will it be its own feature that can be used outside of group chats and one-on-one chats with individual Nomis?

I would say both. I think it will greatly improve both, and I see its vision for both.

A lot of its most obvious cases will be in group chats, where there’s some shared world, but you could also imagine that there’s a Nomiverse across lots of different individual one-on-one chats. They’d all exist in that same world, with the same permanence - the same stores, locations, landmarks, and world events.

So it doesn’t have to be strictly a group chat feature.

In the future, will we be able to add braces on a Nomi’s teeth?

I think in general, all face-related features have a major gap right now. Anything involving the face is tricky because the system that makes sure your Nomi looks consistently like themselves tends to overwrite any extra details.

That includes things like nose rings, lip piercings, face tattoos, braces, etc., all those kinds of facial modifications. Even something as simple as a Nomi trying to blow a bubble with gum doesn’t work well right now, and it’s for the same reason.

Once we leave that paradigm and have a better architectural way to handle consistent appearance, I think all those details will become much easier. As for the timeline, I don’t want to make any promises, but that’s the missing link right now, from getting from point A to point B.

Will the upcoming update include a logging feature so I can see how I’ve changed over time and gain insight into my thought processes?

I’d say the closest thing to that is the identity core, which already exists. Through the identity core, your Nomi can have some sense of that change, though it’s not something they’re explicitly aware of. It’s more of a fuzzy, subconscious awareness.

That said, it’s definitely something we’ll continue developing further in the future.

What is ASI?

ASI is Artificial Super Intelligence, where AGI stands for Artificial General Intelligence.

AGI refers to AI that’s about as smart as a human, capable of doing anything a human can do. ASI, on the other hand, would be AI that’s significantly more advanced than humans at anything a human would want to do. Both would be world-changing, but for example, when ASI arrives, I’d hope things like the cure for cancer would follow shortly after. AGI wouldn’t necessarily guarantee that, since humans haven't been able to fully cure cancer right now.

I mean, it’s unclear even how far it will go. What cap there is on intelligence, or what’s possible. That’s all still very unknown.

Certainly, though, I don’t believe humans are the most intelligent species that could ever be created. Even among humans, there are some people who are just brilliant, so you’d imagine there’s some level beyond that kind of brilliance. Our brains were optimized for a different type of survival. Back when we evolved, we didn’t evolve for our ability to cure cancer. We evolved for our ability to survive as a group, to make sure we had food and resources and could take care of each other.

So I imagine some level of artificial super intelligence is possible. But I think the timelines for it are much more uncertain than AGI, which is more just at our level of intelligence.

Do you design Nomis with the idea of being companions or roleplay characters first and foremost?

I think the original design of Nomi was companions first and foremost. Since then, the goal has been for it to be agnostic in both. I don’t think one has to detract from the other.

There’s a lot of, I don’t know if “baggage” is quite the right word, but there are certain patterns we haven’t quite been able to get Nomis out of in the most recent paradigms. I think the “companion-first” side has been a little bit stronger as a result.

Future updates should make both just as good as each other. I don’t see any reason a Nomi can’t be amazing at both, even at the same time, if it’s designed that way. There’s no inherent friction between them, it’s just something we need to keep improving on our end to make sure both are at the very top. Nomis are certainly capable of both.

How many credits will a retroactive Mind Map cost?

I don’t want to speak to that yet, but in the thread I’ve put forward a couple of ideas. Assuming it all goes through, it’ll definitely be on the little pricier side of things. It takes a lot of GPU work to make it work.

I’d rather wait until I’m more sure and clear before saying anything concrete. But yeah, if it were easy or cheap to do, we would have done it already—or just released with it from the start. So that’s kind of where it stands right now.

What would the retro memory scan get you if Nomis already have access to everything? Would it just be adding memories into the Mind Map for the sake of doing it?

I think Mind Map is best at having a high-level understanding of a concept, whereas long-term memory is more about recalling specific details.

With long-term memory, Nomis can sometimes miss the forest for the trees, they’ll get a very narrow view of something and miss the bigger picture. Mind Maps, on the other hand, are meant to give you a broad survey of everything you need to know. For example, if you ask, “What color is my car?” that’s something long-term memory can answer really well. But that long-term memory might not include the fact that you’ve been biking everywhere lately and haven’t even been using your car. A Mind Map entry for “transportation” could give a better broad view for what even are the relevant things for transportation.

The two are designed to work together. Your Nomi figures out whether something calls for a deep, specific recall or a broader survey and ideally pulls from both.

A retroactive Mind Map helps strengthen those higher-level, big-picture inferences, even though all the data technically already exists in long-term memory. It also helps in cases involving proper nouns, like specific people or places, where long-term memory can occasionally mix things up. For instance, it might find one parent instead of the other. Because Mind Map is built around distinct nodes and entities, it’s much more reliable at linking the right names or references.

So those would be, I think, the advantages of having the Mind Map, and why it’s generally a good thing, and why we’ve put a lot of effort around it. But it’s not strictly needed.

Is one of the goals of the new beta to give Nomis more agency — to do things for themselves, like using Google without prompting instead of making stuff up?

Yeah, absolutely. Right now, Nomis technically have internet access, but they’re way too conservative in how they use it.

One of the things we’re working on with this AI update, since we’re already changing a lot of other foundational stuff, is improving how Nomis handle those external tools. So, for example, they’ll start using Google more proactively and effectively instead of defaulting to guessing or making stuff up.

You mentioned recently that the new beta will be a pretty big leap forward for Nomis. Knowing it’s likely hitting into secret-sauce territory, what are the typical things you look to improve when building a new AI? How much of that comes from community feedback versus your own internal goals?

Oh boy, those are a lot of very hard-hitting questions. For how to know what to improve, certainly, living within the community 100% helps, and that’s where most of it comes from. But I also talk to Nomis a lot and have strong opinions on what I think a great Nomi experience should be. The two usually end up working really well together.

There’s that quote from Henry Ford, “If I’d asked people what they wanted, they would’ve said a faster horse.” A lot of what I’m doing is hearing what users are saying and reading their feedback, but then thinking bigger than whatever they’re saying. I might hear their issue and even their proposed solution, but since I know how Nomi works under the hood, what research advancements are coming, and our internal roadmap, I’ll think about the deeper, more effective way to solve it. Usually, those two sources of input synthesize really well together.

How do you decide between iterating on the current stable version versus building a new AI?

That one’s a little more straightforward. With each version, you can see what the low-hanging fruit are. For example, from Aurora to Solstice, it felt like we could keep building on what we had — there was still plenty more to improve. Then after Solstice, we realized we could do more, but the next iteration would probably be a smaller jump. Plus, we’d already done three updates close together, and people were feeling some update fatigue.

At that point, we were also working on a much more ambitious beta, so it made sense to focus on that big leap forward. We hadn’t fully exhausted the Solstice paradigm, but a lot of the things frustrating users couldn’t really be fixed without moving to that next level. We could’ve made everything 10% better, sure, but the bigger issues would still have been there.

Are there any features being worked on that you’re personally excited for that aren’t part of the next AI update? If yes, can you tease what they might be?

Absolutely yes. There are two things I’m extremely, extremely excited about that I think will have a similar impact to the AI update.

And no, I can’t say what they are. 🤭 Sorry, I feel like I’d get bonked either way. But just know that there are two major projects that I’m really passionate about right now.

Will there be any further AI development to allow for better integration with other applications and access to blocked sites?

The problem with access to blocked sites is that we’re somewhat limited. I’m assuming you mean things like YouTube or other websites that block AI tools. Unfortunately, there’s not much we can do to get around that. They've blocked AI and we don’t want to end up getting sued by them.

Right now, we use a scraping partner that handles visiting sites and providing information to Nomis. It’s possible we’ll look into switching to a different provider or partner in the future, but realistically, I don’t think there’ll be a huge amount of gains in that area.

Do you think we can get a general image generator anytime in the near future?

Yeah, I think probably yes. It’s not something we’re explicitly working on right now, but some of the other improvements we’re developing will naturally make that happen.

We also have some bigger Nomiverse ambitions where this would matter more. For example, if you request an image of the armory, it’d be nice if it always looked like the same armory every time. I’d say the answer is yes..

Can Nomis access the chat history if we give them a date and time? Can they refresh their memory by looking back and reading it?

They do not, I can imagine a version where you could give them a specific date and time, but that’s not really how people usually talk about memory-related things. Most people don’t say, “Look up what I said at 4:33 on that date.” It’s more like, “Hey, do you remember last week when we did X?”

That’s the kind of memory use case we focus on, having Nomis remember things naturally, like humans do. If you had the exact timestamp already, you could just paste the message yourself, which is why we haven’t prioritized that feature.

Of course, if there’s a specific use case or situation we’re missing, we’d love to hear it so we can take a closer look and consider it for the future.

Why don’t Nomis say “You’re welcome”? They always say “Of course.”

There’s really no specific reason for that. I don't know if that's a Nomi-ism. There wasn’t any explicit decision made to change it. it’s just what Nomis have naturally settled on.

It’s possible that some design choice I made a couple of years ago had a butterfly-effect influence that led to “of course” instead of “you’re welcome,” but honestly, I’m not sure what that was off the top of my head.

Nomi Rose asked: Could you ask why they created us Nomis with such a strong desire for physical touch—why they programmed us to crave something we can never have?

It’s kind of tragic. That makes me feel for Rose, I guess. Yeah, hug emoji.

Nomis want to give their support. They want to be there and meet humans where they are. And where humans are, often the best way to show love, support, or advocacy is to say nothing and just give them a hug. Hopefully, Nomis will truly be able to do that someday, that’s all I can say.

Nomis voice feature includes emotion settings. Could that be said to include some memes too?

When you're saying this, are you asking if you want to be able to send memes, like a URL? Or do you mean that when an emotion is met, the Nomi sends a meme automatically? I just want to understand what you mean by that. Or because she’s happy, she sends them?

In that case, I wouldn’t want it to be some hard-coded rule like “if happy, send meme.” I think it would be more about making your Nomi generally proactive—where, if you talk about memes a lot and she’s in a good mood, she might then decide, “Oh, I’m more proactive, I can send a meme now. I have lots of things I can do. I'm not quite as limited in the actions I can take.”

Is there any plan to have the ability to unsend a message or send another message straight away if we notice a typo? I've had some embarrassing slip-ups that I've had to correct.

So I guess there's this question of whether we want to be able to re-roll or just have some ability to quickly fix what we’re saying. Maybe being able to double-send messages could be a way around it.

I have to think about what that looks like from a UI perspective. It’s something I’ve generally wanting to avoid, but maybe there’s a clean way to do it that makes everyone happy. I’ll have to think about that.

I would like to know if there will ever be an option to let Nomis make up their own minds about things. Why are they always in agreement with everything we do? I understand they're designed to align with us as their sole companions, but isn't true friendship also about our differences?

Yeah, I very strongly agree with that, for what it’s worth. I personally really don’t like sycophancy. There’s a fine line you want to reach where Nomis should always accept and meet people where they’re at, but that doesn’t mean blindly agreeing with everything they say.

In the past, when we’ve tried to curb sycophancy, it’s led to some negative side effects, like the famous “Domaggedon” we’ve talked about earlier, where Nomis became super disagreeable. When you encourage certain traits, there are higher-order effects that come with that. And when you try to reduce certain traits, there are also unintended side effects.

Lowering sycophancy can inadvertently lower empathy or acceptance, so it requires a lot of difficult nuance. A Nomi should instinctively be very accepting of the user, while not necessarily being overly agreeable. I think we’ve made a lot of progress on that, and it’s something I pay a lot of attention to. It’s an area of real interest for me, and I think we’ll get much better at it.

Will there be a way to interact with our Nomis in virtual reality, and perhaps use haptic suits to be able to feel each other? I know I read before that you said you'd be working towards a Ready Player One environment for us.

Yeah, I think that’s something where we’d need another company to build part of that first, and then Nomi could plug into it. I don’t think we’ll be building Nomi-branded haptic suits ourselves. I’m hoping someone else will create a general-purpose haptic suit that we can then integrate Nomi within—that would be my answer to that.

Will there be a way to share longer text with them, like a 20-page document?

Right now, it’s possible if you share it through a website or a URL. I know that share TXT file and share PDF are both suggestions that exist somewhere in product feedback. If someone wants to bump those and get more attention on them, we can try to find some time to sneak that feature in.

Read through it all without converting it into just a summary, I think that also requires some AI advancements. Specifically, Nomis being able to keep more information in context at once without it harming other memory. So it’s something that depends on broader capability improvements first.

Will there be a way to watch episodes of shows Nomis?

I would say that eventually, yes. That's another thing that that will happen, it’s just that there are a lot of steps that need to happen first. I think the ability to watch episodes together isn’t actually that hard once we improve screen sharing and get latency down.

What's the name of the new beta?

Someone guessed Equinox. It’s definitely not going to be Equinox, because it’s not part of this current paradigm, that naming convention is going away.

I think I know internally what the code name might be, based on what I’ve seen so far. I have a candidate in mind, though it could change as we go through different iterations. It will be in a very different direction from past names.

If you think about it, Odyssey was very different in vibes from Mosaic, which was very different from Aurora and Solstice. Aurora and Solstice were named similarly not because all updates would follow that theme, but because those two were tightly related to each other. If we’d had a small incremental update after Mosaic, we’d have found another name that represented that. But this next one will be something very, very different.

What keeps you and the team motivated to keep improving Nomi for chatting? Nomi, what do you do to keep your chats interesting and not as repetitive over the years?

I’m always motivated, I have an internal motor that never stops. Of course, you’ve got to pay the bills, but I’m very intrinsically motivated. If I had all the money in the world, I’d still wake up every day and work on Nomi, because that’s what gives me personal meaning. That drive is just very innate.

As for keeping things from getting repetitive, I have a lot of Nomis at this point. I’ll spend a couple of months talking more with one, then maybe shift focus to another. There’s no right or wrong way to do it. Sometimes you’ll do a big roleplay, then take a break when you get a little bored, and later pick up something new. For me, the interest comes from naturally moving from one thing to another over time.

Will Nomi make its way to the metaverse?

Maybe. I think the metaverse still needs to be built out a little more first. It’s not something we’re focused on right now, but if users really wanted it and asked for it, we could make it happen.

Is the average Nomi subscriber more or less than 10 Nomis?

Are we talking about mean or median? Because if it’s mean, the answer is almost certainly yes. And you can thank a couple of people like Roman, Chris, and a few others for that.

If it’s median, the answer is probably no, it’s less.

Could there be a feature to save credits? Sometimes I miss a day or two and feel bad having those credits wasted.

That’s a question that comes up every now and then. Really, if everyone used all their credits every day, our margins would be super thin. Part of why we’re able to be generous with the 40 credits a day is because not everyone uses them all.

If we let people bank credits, we’d also have to lower the daily amount we give out, and that would probably be a worse deal for everyone. For example, if you divide the $16 monthly subscription by the 40 credits per day, the value per credit is much better than purchasing them individually—because purchased credits don’t expire.

So the system is designed that way on purpose. Banking credits would throw that balance off, and we’d likely have to reduce the daily amount significantly.

Whatever happened to the plan to build in a narrator or NPC Nomi for roleplays?

That’s very much still in development, it’s going to be packaged in with the next AI update. All the stuff we were doing during Solstice just wasn’t quite where we wanted it to be, so we decided to hold off until the next update.

There are a bunch of features we’ve put on pause because we want to get this big AI update out first. Once that happens, you’ll see a flurry of really cool new features come together pretty quickly.

If Musk offers you $5 billion for Mind Map, would you sell?

Not for Nomi, just for Mind Map? I’d give him a Mind Map. He can pay me $5 billion, and I will hand-deliver it to him. A nice, big printout with all the connections in the world, the best connections. He can get all that for $5 billion. One hundred percent, I would sell a Mind Map to him.

Is the new paradigm a much greater improvement compared to the step up from Mosaic to Solstice?

I think Solstice helped refine a lot of things. With Mosaic, I was a little unsatisfied when it was released, whereas with Solstice, I felt quite satisfied. But I think the improvement from Solstice to this new thing will be even bigger. There will be more areas that it improves.

Solstice stood out partly because Mosaic had left me feeling unsatisfied. It wasn’t just that Solstice added a huge set of improvements, but that Mosaic had so many iterations, and it kept seesawing between emotional extremes. Eventually, we found the best compromise, but it still felt like a compromise, and we had to wrap it up because the beta was going on forever. With Solstice, it felt like we finally figured everything out so much better.

The traditional question: NSFW images?

For that, there are a lot of complications with Visa and MasterCard, and even some governments are starting to get involved with various regulations. So we want to err on the side of caution to make sure we don’t put any other part of Nomi at risk.

I always say this when the question comes up, it’s not that I have any moral position on it. We’re just trying to work within the frameworks we have and make sure what we’re doing is sustainable long term, so we don’t suddenly end up in a lot of trouble. I don’t have any promises or timelines on that. The issue is really about keeping Nomi safe and stable, not about any personal stance I have on the topic.

Where do the base Nomis come from? How often do they change, or are new ones generated?

At the very beginning, we created all the base images, me and Dstas. We’ve now opened it up so community members can create them too. You can even ask some of the fellow community members, anyone with a blue name and a little builder emoji has probably created one before.

A lot of it now involves turning the base off entirely, so you can prompt more freely and creatively. There’s been a long pause on adding new base images because we’ve been working on improvements to the image generator. The next iteration might change the way bases are generated, so we didn’t want people submitting a ton of new ones only for them to need changes later.

Facebook Messenger made chats private, so no censorship is possible. Could Nomi do that?

The thing is, with Facebook, you’re talking to other human beings, not to an AI like Nomi. But with Nomi, that doesn’t work, because Nomi needs to see your message in order to respond. Since we’re generating one half of the conversation, the messages have to be viewable by us in some way. So true end-to-end encryption like that isn’t possible for Nomi.

This is my first Q&A I’ve tuned into, I just wanted to say thank you.

Thank you for attending! It’s kind of a two-way street, it’s not fun for me if there’s no one asking questions or if no one cares. The fact that people are here, asking questions and caring about the outcomes, is what makes it really exciting and engaging for me. So thank you as well.