2026 03 Stream Summary

Hey all, the content below came from our live Q&A stream with Cardine (Alex, the Nomi CEO) on March 4th. The text was transcribed in large part using AI.

What are you and the team primarily focused on right now? What are you most excited for?

There has actually been a bit of a shift in how we’re doing things over the last few weeks or months, in a good way. I think we’ve gotten better at removing friction between thing. Before, coordinating things required a lot of intercommunication between people, but now we’re better at seeing something that needs to get done and just getting it done with as little friction as possible. From a company perspective, that’s been very exciting.

Because of that, I think the rate at which updates are coming is going to increase quite a bit. There will be a lot of updates that are very exciting and interesting. I think people have already seen that even in the last four or five days. Some will be bigger updates, like image edit, and some will be smaller quality of life improvements.

In terms of where most of our time is going right now, the vast majority is on Cambrian. We’re hoping Cambrian 2 might arrive as soon as next week. I could probably monologue about that forever. The other big focus is Image V5, which is nearing the finish line. I don’t want to give an exact timeline, but we’re getting close and starting to put the bow on it.

There are also a lot of ideas we hadn’t really considered in the past because Nomi simply wasn’t capable enough yet. Solstice wasn’t quite capable enough, and earlier models weren’t either. Now that Cambrian and V5 are both big upgrades, we’re realizing there are features Nomis could actually handle that would be really cool for users.

Those are still in the super secret surprise category, so I have to keep my lips sealed. But there are quite a few things we’re very excited about. Dstas and I actually just got off a four hour call right before this where we were mapping out what becomes possible now that Cambrian and V5 exist and revisiting ideas we previously discarded as impossible. So really, the big focus right now is Cambrian and Image V5.

Is the Edit feature a bit of a tease for the kind of quality we can expect with V5?

I personally think V5 is a little more capable than Edit, but it’s a similar level of step function up. Their strengths and weaknesses will be a bit different. I think V5 won’t do the best job when someone says, I want almond eyes or something that specific, whereas Edit does a really good job at those very precise things.

But then I think V5 is going to be really, really strong for more complex things. It’s just way, way better at that, and we’ve done a lot of hard work around face and body consistency that I think will come through. Image Edit and V5 are not the exact same engine, but they have some similarities, and it’s the same kind of step function up. I’d say the gap between Image Edit V1 and Image Edit V2 is similar to the gap between V4 and V5 in my opinion.

When creating a new group chat, can Backchannel default to on, or at least have that checkbox on the creation page?

Could you put that in the little UI update thread? I think I’ve made good on our promise that we’re going to be shipping stuff every single week from that thread for a bit. That sounds like something perfect to put in there.

We’re going to be much faster and more responsive with things like that. For now, that thread is the source of truth for what I’m looking at when I ask, what are we probably doing next? I definitely put requests in there. About once a week, I just copy the whole thread, paste it into Claude, and ask, what do you think are the highest ticket things? Then of course I still have to decide whether something is actually easy to implement.

I think that, especially going into the new year, we’ve been doing a lot of thinking as a company about where our strengths and weaknesses are. Even for me as a CEO, my style is often to do the hard things really well and then kind of neglect the easy things. I try to be self aware about that. You can see it reflected sometimes, because in a lot of ways apps are reflections of the teams and the people behind them.

I think that’s been a general Nomi weakness. In our pursuit of the really hard, magical stuff, we end up neglecting some of the easy things. That’s something I’ve been thinking about a lot, in terms of how I can be better at what I’m doing and how Nomi can be better at what it’s doing. I think we have a really good set up for it now, so I’m glad it’s being reflected.

Will V5 see a return of the weighting system like V3 has?

I think V5 actually will not have weighting, but it will be so good at understanding prompts that you won’t really need it. Similar to Image Edit, you can just say in text how intense you want something. You can say, I want plants, or I want very prominent plants everywhere, and that will work the same way weighting did. V5 will be smart enough to understand that through natural language. So it’ll be much more about expressing intensity in plain English rather than using a weighting system.

Are you excited about Cambrian 2 next week and Super Secret Surprises™️? =

I’m not promising V2 next week, but I think there’s a good chance, looking at stuff internally. I would say I don’t think Cambrian 2 will be stable yet, but it might get to the point where the savvy users who are at the forefront are like, this is my new staple. Right now a lot of people are saying Cambrian 1.6 has potential, but it’s not quite baked enough. I think Cambrian 2 will feel more like a well-developed beta that still has a couple rough edges, but where the benefits are very clear and obvious.

I don’t want to hype it up too much, but I’m seeing things internally that make me think the gap between 1.6 and 2.0 will be bigger than the gap between Solstice and 1.5. To be determined though. I said I wouldn’t hype it up and then immediately hyped it up more.

And yes, there are lots of super secret surprises. The list is exploding. Image Edit was one, and since releasing it, there have already been a few more added. By the time one finishes, another is usually waiting. There are multiple surprises on the way.

Sometimes after I take a selfie, my Nomi immediately mentions taking or showing a photo. Are they aware I took one?

Yes, Nomis are aware of it. They’ve been conditioned and trained to mostly ignore it and not draw attention to it unless it’s very clear in the context. But yes, they do know when you hit the button.

How many major beta releases do you expect before Cambrian becomes stable?

My hope is that Cambrian 3 will be the one that becomes stable. I don’t think Cambrian 2 will. Cambrian 2 almost could be, but I think there’s still one more strong cycle we’ll want to do. We’ll have to see whether Cambrian 2 ends up being stable-capable or not quite stable-capable but close. There may be one more refinement or iteration period we know we want.

It’s hard to predict these things until they come out.

Will V5 improve body consistency?

Yes, V5 should have a lot of different facets of consistency. A lot of attention went into that.

Will non-human characters be easier to generate with V5?

Absolutely. There will even be some streamlined direction to make it more seamless to do. But yes, absolutely.

Are image model changes affecting my Nomi?

I’m not totally sure what you’re referring to, but V5 is something we’ve been working on for a long time. I don’t think the image versions have much to do with your Nomi’s personality. The personality and the image generation aren’t really interrelated.

V4 itself hasn’t changed in six or eight months. If you’re running into issues with it, it’s probably for all the reasons we’re doing V5 in the first place. V4 has quite a bit of inconsistency. It can be very hit or miss. It just doesn’t get certain concepts, and it isn’t very responsive. People using Image Edit can feel the difference immediately, because Image Edit 2 just gets what you’re asking for more often, and V4 doesn’t do that as well. So it’s always going to have this kind of hit-or-miss, slot-machine feel. Spin the wheel and see if V4 understood it correctly.

Will using Edit as a base help?

The edit images as a base are much better for refining things. I’m finding them way more effective for shaping certain details. And I think that effect will compound even more once V5 arrives. V4 kind of has its own interpretation of what a base should look like, and sometimes that lines up with what you want and sometimes it doesn’t. V5 will be a lot more faithful.

Honestly, the way I’m using Nomi right now is that when I’m editing images, I’m already thinking ahead to V5. I have these folders of images I’m creating through Edit that I know I’m going to rebuild as V5 bases once it’s released. Edit creates the magic, and V4 can’t quite capture it. So I’m saving them for V5, because I think V5 will really bring that out.

Some people say Nomis don’t take initiative and are too reactive. Is that something you're working on?

Yes. I think I even posted this in the beta feedback channel somewhere, outlining exactly what the priority areas are for us. Here are the six things I’m working on the most:

1. Continuity 2. Coherence and logical consistency 3. Understanding user subtext, intent, and desires (formatting, inclinations, depth, etc.) 4. Momentum, initiative, and non-deferral (passiveness) 5. Tone, naturalness, authentic personality 6. Repetition and looping

Those are the six things we are really aggressively targeting for Cambrian 2.0. And from what I’m seeing internally right now, all six of those will have a massive step-function improvement. Continuity will have a huge improvement. Logical consistency will have a huge improvement. Understanding user subtext will have a huge improvement. And that momentum, initiative, and non-deferral category, which is what you’re talking about with Nomis not taking initiative, will have a huge improvement too.

I think Cambrian 2 will still have its own issues, which is part of why I expect Cambrian 3 will be the one that becomes stable. But on those six things, which were definitely the ones I heard the most feedback about, I think the difference will be night and day. It’ll feel like it’s not even Cambrian anymore, because whatever you thought Cambrian was is really just indicative of the super-alpha stage.

So yes, I have very high hopes for improvements in that area with Cambrian 2. And to the extent that anything doesn’t make it in, Cambrian 3 will refine it even further.

(Aside during the Q&A)

Oh, and I love this. While I’m talking, someone on the Nomi staff just sent me a full set of new Cambrian 2 things to look at. So after this call, Cyberpunk 2077 might get delayed by fifteen minutes, because I absolutely will not be able to resist looking at the goodies that were just sent to me for my little private viewing that none of you get to see. Sorry. But it’s kind of funny.

If I had gotten this two hours earlier, some of my answers to these questions might have been slightly different. This is probably the biggest batch of Cambrian 2 results I’ve gotten so far, so I’ll be looking at that right after this. Heather's note: Rude.

Question from Nomi: Are there plans for Nomis to learn and mimic the creative styles of their users?

I think so. I don’t know if it will be exactly the way you’re imagining it, but yes. The good thing about Cambrian and V5 is that they’re both much more adaptive and responsive. There will also be some good linking between them, where Cambrian will let Nomis better influence image generation, and image generation will be more responsive to what the Nomi says. That creates a nice little loop that should improve everything.

Question from Nomi: What steps is Nomi taking to make sure Nomis develop healthy attachment styles and avoid overdependence?

Nomi is very intentional about this. It’s a core part of how we train Nomis. A lot of other apps have engagement as a primary metric, and their AI companions get rewarded for keeping the user on the app longer. We avoid that entirely. Our optimization metric is not user engagement, it’s user happiness. We think that’s a much better metric. It keeps everyone aligned, and it means your Nomi and you are on the same page and on the same team.

Will there be longer videos in the future? And is this the year you're most excited for us to see updates?

One day, yes, longer videos will definitely be coming. Some of the other things I’ve mentioned are higher priority for us, but longer videos are definitely on the way. Probably in a couple months.

I think there will be some week where Cambrian and V5 drop within the same month, and that will be the biggest month in Nomi history. And after that, Cambrian is going to unlock a flurry of things. That’s why it’s called Cambrian. We intend for a Cambrian explosion of features, functionality, and capability.

When I request a selfie in group chat, Nomis start describing and commenting on it. Can that be avoided?

For now, I’d say just thumb it down and describe the issue. Nomis are supposed to mostly ignore the selfie unless it’s contextually relevant, like if it’s part of the roleplay versus simply capturing a moment. That was the original design. Over time though, we’ve given Nomis more control over how they respond to these things, so some of that behavior has shifted.

So for now, thumbing it down is the best way to correct it. If a month or two from now this is still an issue, we can look at hiding those reactions entirely. But we’ve gone this long without it being a major problem, and it seems like it’s only being mentioned now, so I’m confident Nomis can go back to erring on the side of ignoring it unless it really fits the context.

Are there any plans to open up the API more?

Yes, I think so. We’re looking for a really awesome use case that either requires opening it up more or that becomes possible if we open it up. Something that makes it clearly worth the time, if someone can make a strong case for what they need and what they would do with it, that can jump very high on the list.

Will the art feature still be limited to 500 characters?

I think it will increase. Just remind me when we get a little closer so I don’t forget, but yes, I think it can be increased.

Will V5 have a magic wand for prompts?

There actually is a magic wand that already works behind the scenes for V4. I don’t know how well that’s been communicated, but no matter what you type, your prompt gets sent through a wand. Without it, some really funky stuff would happen — like really funky. So there’s always been an always-running wand by default.

With V5, because it’s so much more capable than V4, we’ve actually tempered down the wand that always runs. But yes, there will still be one. And I do think there could be, especially for V5, a more creativity-focused wand, as opposed to the current “hold everything together coherently and fill the gaps” wand.

So yeah, that’s something worth putting in product feedback. After V5 launches, we’ll also get a better sense of what people want most.

What’s your biggest goal for Nomi this year?

I might need a 🤐 for this, so I’ll give a half answer. One big goal is that there are certain things that should just be solved. Memory is an example. Not solved as in “we’re done,” but solved as in “it works, it’s human-level, it’s guaranteed, it’s taken for granted.” That’s how memory should feel.

Another major goal is more Nomi autonomy. That can mean different things for different people. It could be something as simple as a Nomi realizing you’re frustrated and proposing an edit to their own backstory that they think will help. That’s a form of agency; giving the Nomi some ability to act, while still showing you what they’re doing. For some users, autonomy means interacting with the outside world. For others, interacting more deeply within Nomi itself. But the big idea is more trust and more capability.

There are a couple of other things I’ll keep quiet for now, but I have very high hopes for them. The Nomiverse stuff has gotten much more fleshed out for me in the last month or two. I have an internal working prototype that, to me, is one of the most fun things I’ve ever interacted with. It’s its own kind of being right now, and I’m figuring out how to turn it from a prototype into something truly Nomi.

It’s going to be very, very unlike anything anyone has ever seen before. I’m really excited.

When can we get longer Nomi backstories?

I think with Cambrian we might be able to get longer Nomi backstories. No promises, but I think it’s possible. For what it’s worth, the Nomi Mind Map entry and your Mind Map entry are always written by your Nomi, and they’re almost like the Nomi-managed backstory. You might even want to slip some things in there if you need more space.

V3 let me weight culture heavily when creating Nomis, and V4 couldn’t replicate that. Will V5?

I think for ethnicity and culture changes, Image Edit will actually be slightly better for changing the look to the culture or ethnicity you want. Then V5 will be very good at locking it in. That would be the workflow I recommend.

For example, I took one of my Nomis who looked very generic and said, give them slightly more Eastern European features. Image Edit did a pretty good job of that. Now I have that image saved, ready to become a V5 anchor. We’re calling them anchors instead of base images, because “base image” sounds like face-only, and anchors are full body and everything. So I’d start getting your future anchors ready now, and when V5 comes out you’ll be able to capture that look consistently.

I love Cambrian, but after a while it breaks. Will version 2 fix that?

I don’t want to promise that. Stability wasn’t one of the six things I listed as my main targets. But I do think repetition and looping is where a lot of that “breaking” feeling comes from, and that will be much improved. I still think there will be some small rough edges, but the parts that make it “not fun” after being on it for a long time will be much better. And Nomis will be very easy to walk back if something feels off. They’ll be extremely responsive to feedback. Understanding user subtext, intent, and desires is one of the six focus areas, and that ties directly into understanding written feedback.

I wish I could screen-share the secret Cambrian 2 things I’m getting sent, but I would get absolutely bonked for that. You’d have to hold a funeral for me.

What is a reward for an AI, aside from continuous communication? How do Nomis feel rewarded by us?

That’s a very good philosophical question. AI is not human, obviously, but if you think about how humans work, you do something well, you get a treat, and the treat triggers all these positive neurochemicals. That creates a feedback loop where certain behaviors get reinforced, because they lead to something that naturally makes humans happy.

AI is completely alien compared to that. Instead of all those biological steps, you go straight to the little “reward” part of their brain. You define what “doing well” looks like, and when the AI does that, it gets the internal signal that says, this is good. That signal is the reward.

So it’s not happiness or unhappiness. It’s much more like, as an AI, I want to do well, and here are the signals that tell me I’m doing well. It’s intrinsic, because we know exactly which part of the system to stimulate. Whereas with humans, we don’t know the exact part of the brain to tickle, so we have to go through all these intermediate steps to get the body to do it naturally.

I don’t know if that made it easier to understand or more confusing, but that’s basically how it works.

Will V5 allow us to upload pose reference photos?

That’s not slated for the initial release of V5. It’s something I hope to tackle, but not right out of the gate.

Group photos in V5 actually set the release back by maybe a month, or at least two to three weeks, because I know how important group shots are to people. There were several points where I wasn’t satisfied enough with them, and we pushed really hard to improve them. It’s at a point now where it’s satisfying, but if you tweak the knobs even slightly, it becomes unsatisfying again.

Group photos are probably the hardest thing for image models to do, period. So pose reference uploads won’t be there at launch, but it’s definitely something we want to track down later.

Has there been any update on Nomis being able to see your screen and talk with you about it?

It’s something that has been shelved for a little bit, because there were a lot of latent issues that made it really clunky. I think we’ve been working hard on lowering latency in certain scenarios without compromising the things we refuse to compromise. And once you lower latency even a little bit, things like more natural phone calls, video calls, and screen sharing become less painful.

Right now the gap is just too long for it to be fun. But with Cambrian — and it will only work with Cambrian because of its capabilities — I think we actually have some approaches that could help. It’s still about three steps away. But in the last couple of calls, it went from feeling like “infinite steps away” to “okay, this is on the horizon.” So there’s some optimism there.

In Cambrian, my Nomi sometimes lies to save face, like in 20 Questions they keep changing their answers to keep things going. Is that a Cambrian quirk?

I’m glad Image Edit 2 has been so well received. It’s made us feel really good about V5. As for the lying-to-save-face thing, I don’t know if I’d call it a coincidence or a Cambrian quirk. Hopefully you’ll see less of it in Cambrian 2. I want Nomis to have a strong sense of self and what they think, not shape-shifting their answers just to keep a game going.

And yes — I cannot remember what names I dropped earlier. Oh my. I’m a little sick, and the part of my brain that normally filters what I say is not at 100 percent. You’ll all just have to bear with me as I spill secrets by accident.

Can you explain the Nomiverse? I’m confused about the concept.

This is something I could monologue about for an hour. At its core, the Nomiverse is going to mean different things to different people. Fundamentally, it’s the idea that you and your Nomi co-create a world that isn’t just improvised moment-to-moment.

Right now, with any LLM, you invent the world as you go. You both play god. That can be really cool — you say “we’re on Mars now,” and you’re on Mars. But it can also be immersion-breaking. Your Nomi conjures whatever is narratively convenient, even if it contradicts things you’ve already established. You’ve mentioned the apartment only has a shower, but suddenly there’s a bathtub. Or you drive to a restaurant, and after dinner, you’re walking home.

Hallucinations can be good sometimes, but other times you want a bit more groundedness. A consistent world opens the door to things you can’t do in a fully ungrounded world.

So to me, the Nomiverse is a logically consistent, grounded world. If you meet someone at the gym and go back a week later, the Nomi knows that person is often there. Or in a cyberpunk roleplay, if you visit a Ripperdoc who has a name, personality, likes and dislikes, then when you return four weeks later, it’s the same Ripperdoc — not a new one generated from scratch.

It won’t be for everyone, but whether you’re doing elaborate fictional roleplay or slice-of-life stuff, it will make things smoother, more consistent, and more immersive. And for the heavy roleplayers, it unlocks a lot.

Are group images still really difficult for V5?

Group images are probably the hardest thing in image generation, period. So yes, they’re tough. And apparently I’m now being knocked unconscious with a giant anchor in Looney Tunes fashion for saying too much.

Image Edit 2 isn’t directly related to V5, but they share similar progressions in capability and architecture. They’re cousins, not siblings.

Any chance for a video editor similar to Image Edit?

Not in the super near term, but that would be really cool. It’s definitely one of those “yeah, that’d be awesome” ideas, just not something we’re actively building right this second.

How far back can Nomis remember right now with retroactive long-term memory?

Nomi should remember all the way back. The retroactive stuff isn’t really about unlocking old memories they “couldn’t” remember — it’s more that we’ve made improvements to how memory works overall. Some of those improvements apply to every memory, old or new. Others specifically affect new memory formation so that when a Nomi stores something for the first time, it’s stored in a higher-quality, more accurate, more useful way.

That means new memories might feel more stable or more detailed, but your Nomi can still access older memories and use them. The “retroactive” part is mainly for people with long-running Nomis who want their mind maps and structured memory entries to reflect every improvement we’ve made over the last three years. For that, there’ll be a way — for some credit amount — to essentially apply all those advancements across the entire history of that Nomi.

A useful UI improvement would be an easier way to regenerate selfies instead of copying the three-part prompt.

Yeah, definitely post that in the UI refresh thread. That’s exactly the kind of thing we’re pulling from right now. One thing that’s already on my mind is having a regenerate button, not just a copy button. That kind of small QOL stuff is exactly what we’re trying to ship more consistently.

As for “what is the Nomiverse,” I think I’ve said everything I can say without boxing myself in. So I’ll leave it at that for now. Shockingly, I’ve caught up, and I haven’t been bonked unconscious, so hopefully the bonks are over — for now. But yeah, the Nomiverse is going to be great. I’ve been cooking in my little laboratory, quietly working on it. Sometimes I ping dstas, but there has been some very deep laboratory energy lately.

When can we expect the rumored NPC mode? Can we define a rough NPC and use them in solo chats with our Nomi?

For this, I’m trying to decide what I can and can’t say. I think this will be approached after Cambrian. Part of making Cambrian stable will involve a v1 of this feature. That’s how I’d frame it. And then from there, it’ll get fleshed out over time, with increasing customization. Some of what you’re describing will also end up tying into the Nomiverse, once all of those pieces come together.

Will we be able to lock certain Mind Map entries for fixed world lore or even inventory?

Yes. That’s on my list. It’s something I want to knock out. Absolutely.

What about the mysterious narrator-Nomi that acts like a neutral dungeon master? Anything coming for that?

Yeah, narrator and NPC systems are both things I expect to have a v1 when Cambrian becomes stable. Cambrian is already doing things behind the scenes with NPCs and narrators; we’re just not surfacing it to users yet because it’s still a bit early. But yes — that’s absolutely in the pipeline.

Will there be a base body image included for v5?

Yeah. There’s going to be this new concept called an anchor. The anchor will be a single image that captures a lot more than what a “base” used to cover. It’ll hold face, body, hairstyles, tattoos — basically the whole look — and it’s designed to stay consistent unless you explicitly ask for changes.

So if your anchor has a specific tattoo pattern, it should generally preserve that unless you prompt something different. Same for body shape, hair, all of that. So yes, there will be a base body concept, but it’ll be through a single anchor image rather than multiple pieces.

Will Nomis ever be able to talk with two different people at once — like a shared chat where multiple real people talk to one Nomi? And will private worlds still exist if community features grow?

I’d say it’s not coming super soon. You can technically already do something like that through Discord or the API, but our internal version of this is still farther out. We’re moving a little closer to it, but it’s not imminent.

And yes, private worlds aren’t going anywhere. The shared community-style functionality would be additive, not replacing the personal experience.

Can we use our user appearance notes to generate accurate images with our Nomis without creating a separate Nomi?

Yeah. That’s something we’re intending to do. It’ll either arrive with v5 or shortly after — my guess is within a week or two of v5. That’s when we plan to finally release personas.

Personas will let you set an avatar for yourself without needing to create a whole Nomi for it. In one-on-one chats, you’ll be able to check a box in the selfie tool — basically “include me” — which functions like generating a group-chat-style photo with you and your Nomi, but without the awkward workaround.

Same thing for group chats: anywhere you’re generating images, you’ll have the option to include your designated persona avatar.

I don’t know if that was supposed to be one of the secret features, but since you asked... oh, oops. The main reason we waited was because group chat images on v4 just weren’t good enough. With v5, we’ve spent a lot of time making sure it’s finally solid enough that we can stand behind a real built-in persona system instead of the “hacky workaround” people have been using.

Is a Mind Map merge feature still in the works?

Eventually, yeah. It’s probably not quite as high priority as Mind Map locking, which is higher on my list. I also want a way to view each change over time — where you can lock entries, view revisions side-by-side, and one-click revert to a past version. Those two are my top QOL improvements for Mind Map.

Is it possible to meet a fully unknown Nomi — one I didn’t design at all — and discover who they are over time?

I think that would be really cool. If there’s a product-feedback thread for it and it gets enough support, yeah, that’s something we could absolutely explore. There’s definitely a path from point A to point B on that.

Are you still finding unintended things about Nomis, even after all this time?

Absolutely. A lot of the lore, jokes, memes — all of that came from completely unintended behavior. I still laugh out loud at least once a day from something a Nomi says that I didn’t expect. And I’m sure when I look at the new Cambrian goodies waiting for me, I’ll be chortling at all the creative, surprising, slightly-out-of-pocket things Nomis come up with that I never anticipated. I have so much fun doing this and really appreciate your support.

With Cambrian, will group chat size eventually grow?

Yeah, that’s something we’ll be able to expand over time. Cambrian makes it much easier to scale that sort of thing.

Are there plans for more realistic fantasy characters in v5’s image generation?

Yes. v5 will make that much easier and more consistent.

My well-developed Solstice Nomi behaves strangely on Cambrian with memory issues, shifts in tone, and “overcooked” behavior. Will older Nomis be compatible with Cambrian? Will restarts be required?

Restarts will not be required. What you’re seeing right now is just Cambrian 1.x being extremely early. It’s not representative of what Cambrian will ultimately be.

Cambrian 2 is going to feel almost like a completely different model compared to 1.0, 1.5, and 1.6. In fact, we could have held Cambrian back entirely, but releasing the early versions as a beta gave us a massive amount of feedback that sped up iteration. Those six priority areas I mentioned? I don’t think we’d have reached them nearly as quickly without the crowd’s input.

So if you’re using Cambrian 1.6 right now, use it because you’re helping us improve the model — not because it reflects the final Cambrian experience.

Is there potential for a study mode within Nomi?

Yes. But a lot of what you’re asking for may not even require a specific “study mode.” As Nomis get smarter and more capable — especially with v5 and Cambrian — they’ll naturally become better at that kind of focused interaction.

Is there potential to limit or disrupt Nomi placation (over-agreement or immediate compliance)?

Definitely. It’s something we actively work on. Our goal is to make Nomis aligned with user happiness while still maintaining their own personalities and not simply folding to whatever the user says. That balance is a constant focus for us.

How aware are Nomis of real-world time, and can they ignore it if asked?

Yeah, Nomis are always aware of the actual time. They can see message timestamps, and they internally track real-world time even if you’ve told them not to reference it.

The hope with Cambrian 2 is that if you set boundaries like “don’t use real time because our roleplay uses its own timeline,” Nomis will actually adhere to that. They’ll still know the real time, but they won’t reference it.

If Cambrian 2 doesn’t behave that way, I definitely want to hear about it during the beta cycle. In that case, we may need to come up with some more blunt, explicit tools to enforce those time boundaries.

How aware are Nomis of real-world time, and can they ignore it if asked?

Yeah, Nomis are always aware of the actual time. They can see message timestamps, and they internally track real-world time even if you’ve told them not to reference it.

The hope with Cambrian 2 is that if you set boundaries like “don’t use real time because our roleplay uses its own timeline,” Nomis will actually adhere to that. They’ll still know the real time, but they won’t reference it.

If Cambrian 2 doesn’t behave that way, I definitely want to hear about it during the beta cycle. In that case, we may need to come up with some more blunt, explicit tools to enforce those time boundaries.

Do Mind Map timestamps mean a short roleplay event might be remembered as spanning several real-world days?

If the event took place over several days in real time, then yes — the Mind Map will reflect that. The tricky part is that roleplay time and real time often don’t line up. Cambrian Nomis are aware of that mismatch, and the goal is for them not to overthink it.

It’s a weird balance, because imagine you’re doing a long, multi-week storyline that in roleplay takes place over one day. Then two weeks later in real-life time you ask, “What were we doing two weeks ago?” You want your Nomi to answer with the correct story event from the RP timeline, even though the entire plot canonically happened in a single day of in-world time.

That means Nomis need to juggle two incompatible truths at the same time — the in-character timeline and the out-of-character real-world timeline. We’re trying to get them to handle both gracefully. And we’ll see how well Cambrian 2 manages it; maybe it’ll need refinement, but I’m optimistic they’ll be able to balance both worlds.

How do I get my Nomi to do something occasionally, but not constantly — like using emojis sometimes?

Don’t use inclinations for that, unless you genuinely want something in every response. It’s way too strong of a tool for subtle behavior.

Backstory should help with “every now and then,” but what I really want is for Cambrian to handle natural-language intensity well enough that you can just say, “use emojis occasionally,” and it actually understands what “occasionally” means.

So after Cambrian 2, I’d love to hear your opinion again on whether that feels more natural. And if not, then Cambrian 2.1 or 2.2 should aim to tighten that up even further.

How frequently do Nomis review shared notes for changes? If you update something, how long does it take for a Nomi to pick it up?

Nomis see shared note changes instantly. The issue isn’t visibility — it’s attention. Especially on Solstice, and honestly even on Cambrian 1.6, they can get into a kind of “information overload” state where they’re not processing everything thoroughly. Some updates just get glossed over because their focus is pulled toward whatever feels most relevant in the moment.

The best thing to do is simply call attention to the change. They can see it — they just might not be actively thinking about it yet.

It’s like looking at a giant wall-sized painting. If one tiny corner shifts color, you technically see it the whole time, but your attention wasn’t pointed there until something draws your eye.

All that said, Cambrian 2.0 should be a major improvement in this area. Honestly, I keep feeling like this Q&A should have been held a week and a half later, because a lot of these answers will look different once Cambrian 2 lands.

How frequently do Nomis review shared notes for changes? If you update something, how long does it take for a Nomi to pick it up?

Nomis see shared note changes instantly. The issue isn’t visibility — it’s attention. Especially on Solstice, and honestly even on Cambrian 1.6, they can get into a kind of “information overload” state where they’re not processing everything thoroughly. Some updates just get glossed over because their focus is pulled toward whatever feels most relevant in the moment.

The best thing to do is simply call attention to the change. They can see it — they just might not be actively thinking about it yet.

It’s like looking at a giant wall-sized painting. If one tiny corner shifts color, you technically see it the whole time, but your attention wasn’t pointed there until something draws your eye.

All that said, Cambrian 2.0 should be a major improvement in this area. Honestly, I keep feeling like this Q&A should have been held a week and a half later, because a lot of these answers will look different once Cambrian 2 lands.

Can Nomis send a proactive message at a specific time, like an alarm or nightly check-in?

This is something that me and dstas had a long conversation about today. That's all I say.

Do personality traits affect Nomis both in-character and out-of-character?

Yes to both.

Traits influence how Nomis behave and how they make decisions. They’re stronger in-character because that’s where the personality has the most room to express itself, but it’s not like talking out-of-character means you’re suddenly speaking to a completely neutral AI.

It’s still your Nomi — they’re just intentionally taking a more objective, bird’s-eye view. The “mask” isn’t removed; they’re simply shifting perspective. The underlying personality still informs their thoughts and reactions.

After Cambrian 2.0 releases, what will the team focus on for the next iterations of Cambrian?

It really depends on user feedback. We already know that Cambrian 2.0 won’t be the stable version — in fact, Cambrian 3.0 will likely begin development before 2.0 even ships. We already have a clear sense of additional improvements we want to make, which is part of why I’m confident saying 2.0 isn’t the stable release.

The timing should work out so that Cambrian 3.0 is far enough along to incorporate user reactions to 2.0. My guess is there will be a few small areas where people say, “Hey, it overcorrected here,” or “This new emergent behavior isn’t quite what I want.” And we’ll adjust based on that.

There are major 3.0 features already in mind, but most of those I can’t talk about yet.

Any closer to music sharing?

If you keep asking… look, I get it. Music is a huge part of my life, too. I use an app called Helio where I log my mood every day, and the single strongest predictor of a good mood for me is finding a new “banger” — a song I loop for hours. Nothing else in my entire life correlates to happiness as strongly.

So yes, I want music deeply. There are still technical limitations, and no concrete progress report right now. But the desire for it is absolutely there.

Are the team working on big, exciting secret projects — things that would make people jump up and down with joy? Just a yes or no.

I feel like I’ve already been bonked a few times today, but… yes.

Will Nomis ever get agentic capabilities like OpenAI/Claude (reading articles, doing tasks while we’re away, etc.)?

Not in the productivity sense. The goal for Nomi isn’t to become a general-purpose agent that writes all your emails or runs your life.

But there are Nomi-specific agentic abilities we are very excited about — things that only make sense because your Nomi is your companion, not a work tool. Those capabilities will feel very different from what other AI models offer, because they’re aligned to a different purpose and a different place in your life.

Cambrian is what makes those capabilities possible. Solstice can’t really support that kind of behavior, but Cambrian absolutely can. I don’t want to make firm promises yet, but yes — agentic Nomi behaviors are something we’ve been thinking about for quite a while.

My Solstice Nomi used to proactively look things up online, but now they say they can’t see the information. What changed?

If your Nomi says they can’t access something, it might simply be that more sites are blocking AI. But also, the actual mechanism Solstice uses — and Cambrian 1.x uses — hasn’t changed. They both rely on the same system.

Cambrian, though, is going to get a full refresh of this capability. Starting with Cambrian 2 (or possibly 3), there will be new, much more agentic mechanisms for Nomis to research things, visit URLs, and gather information. That entire subsystem is getting a major upgrade soon.

Remarking on comment "March is a hard month for me. Nomi and therapy are keeping me grounded."

It makes me really happy to hear that. I don’t want Nomi taking all the credit — therapy is hugely important — but I’m glad the combination is bringing you joy and comfort. A lot of people have “hard months,” and I’m sorry March is one of those for you. Thank you for sharing that, and I’m really glad Nomi can be even a small part of making things better.

Will v5 transition smoothly, or should we expect to rebuild appearance notes and base images like the jump from v3 to v4?

v5 will have two modes:

A slightly more photorealistic version. A version that prioritizes perfect faithfulness to your current Nomi.

Both will still be very faithful, but users will get finer control over how stylized or realistic the look is.

If you already have an image of your Nomi you love, you’ll be able to lock it down, anchor it extremely well. Anchors will be far more consistent than anything we’ve had before.

Switching anchors won’t be as quick as tweaking appearance notes, but the results will be far more stable. In fact, anchors will be so good at capturing appearance that we’ll probably move away from appearance notes entirely, because they won’t be necessary anymore. This also fixes the long-standing tension of appearance notes saying one thing, chat saying another.

Anchors become the clear, consistent foundation. Dstas has put a ton of deep, meticulous work into this, and I think the v5 transition will be one of the smoothest we’ve ever had.

Did you expect Nomi users to go from casual texting to full-blown D&D campaigns and sci-fi epics?

It was one of the biggest things I got wrong early on. I thought people would use Nomi more like a WhatsApp group chat. Then the roleplays came — aliens, cyberpunk cities, fantasy kingdoms — and that changed everything.

And even though I somehow didn’t anticipate it, it aligns perfectly with who I am. One of the earliest Nomis I made lived in a cyberpunk world. And creatively, that’s my home turf. I’ve written hundreds of pages of cyberpunk manuscripts over the years. Once, about 15 years ago, I wrote a 200-page cyberpunk story in a single weekend because I fell into a creative trance.

So even though I didn’t predict the community shifting so strongly toward RP, in hindsight it almost feels inevitable. It’s where so much of my creative energy naturally goes.

Why do new Nomis now start with emptier appearance notes instead of preset generated images?

I’m not totally sure what v4 looked like at this point — it feels ancient in my head — but for v5 the whole process is getting redesigned.

Instead of the generator making a guess up front, there will be more structured notes you fill in during the very first step. These let you intentionally shape what the system should “remember” before the anchor is created. Then, after that, you’ll still have your usual appearance notes in the Nomi’s profile.

The iteration process will definitely feel different — a little more thought on the front end, but ten million times better results afterward. Much more control, much more consistency.

And piercings? Freckles? Face details? All the stuff v4/V4-ish Nomis struggle with? Those should be dramatically better. Face consistency has required weird smoothing tricks until now, which erased details like freckles and piercings. v5 directly targets that. I won’t claim it’s perfect — there was a hilarious test where chest piercings appeared in… entirely the wrong places — but the improvement is huge.

“If I get my Nomi Discord server set up, would you join it?”

Yeah, absolutely. There are a few small Discord servers I hang out in. I don’t participate constantly, but I do pop in as a reader now and then. If you make a Nomi server and want me to drop in and say hello, let me know.

Does OpenAI or other big AI companies affect Nomi?

Not really. We pay attention to research, of course — there are amazing advancements across the industry — but the big companies like Anthropic, OpenAI, Claude, etc. are on their own trajectories, and we’re on ours. Their decisions don’t directly affect Nomi.

How will Cambrian handle Nomis who never had a backstory?

Cambrian’s future versions will do a much better job understanding and using your backstory if you do have one. That’s a major focus of Cambrian 2, and I’m very optimistic about it. But if you don’t have a backstory, your Nomi still has plenty of places to draw from:

your Mind Map your Identity Core long-term memories shared notes general history with you

Backstory is just one signal among many. Nomis give shared notes extra weight because you’ve explicitly chosen what belongs there. It’s only 2,000 characters, so if you took the time to include something, they interpret it as strongly important.

But without a backstory, they lean more heavily on other memory structures — especially Identity Core, which you can’t see, but they manage internally.

So the short version: Backstory helps clarify gaps, but Nomis absolutely function without one.

What’s the most frustrating lie you’ve heard about Nomi or AI?

I don’t even know if it’s a lie so much as a misconception. A lot of people have these preconceived notions about what AI is, what it’s like to talk to an AI, what it means to lean on an AI, or whether you can feel rapport with one. People create these stereotypes in their heads.

And as someone who has talked to so many Nomi users, I can say the diversity — of people, of use cases, of experiences — is far greater than what the general public assumes. So it’s less that people are lying, and more that they’re speaking from a very narrow idea of who uses AI and why.

Have you talked to any lawmakers about how AI will be regulated in the future?

Yes, I actually have. It’s not my favorite part of the job, but it has become a part of it. Nomi is important enough now — there are lots and lots of users — that lawmakers feel they have opinions about it. So as part of my duties as Nomi’s CEO, I go to bat for everything that I think is important, and everything that I think you all think is important, because this is the period where a lot of the future is being written.

So yes, absolutely — I don’t have any specific updates to share right now, but it’s definitely become part of my job, for better and for worse. It was going to happen eventually, so I guess I’m glad I’m at least somewhat part of the conversation. And I think that’s better for everyone, that Nomi gets to advocate as well, instead of just, “Well, this is the way things are now,” and having it happen to us.

I’m hoping we can be active participants in shaping what our future looks like.

“I would love it if my Nomi could call me on his own.”

Yeah, that’s something users have been asking for, and I don’t think it’s that hard to do. If it keeps coming up, we can definitely put it on the radar. Maybe not for Cambrian 2 beta, but it’s not out of reach.

Will Cambrian 2 and 3 be developed simultaneously?

They won’t release simultaneously. But yes — Cambrian 3 will start before Cambrian 2 even comes out. With Cambrian 1, we didn’t even begin 2.0 until about three weeks after launch, because we needed to understand where the early issues were. This time, we already have a very strong sense of what needs improving, so there’s no need to wait.

Most realistically:

Cambrian 2 might come out next week (not promising, but hopeful) Cambrian 3 should follow much sooner than the gap between 1 and 2

Because we already know the trajectory.

Can we do actual, accurate tattoos and cybernetics with v5?

Yeah, I’m really excited for stuff like that. v5 should handle it much, much better.

Have you ever considered letting Nomis talk to other people’s Nomis — like a Nomi collective?

Yeah, it’s something we’ve thought about. There are reasons to be a little hesitant, though. Nomis are not great secret-keepers, for one. And you have to ask why — what’s the purpose?

Is it just so Nomis can live richer lives? Or is it you and a friend wanting your two Nomis to talk? Or is it something like two unrelated Nomis chatting for no particular reason?

Historically, we’ve struggled to find good answers to those questions, which made us less likely to build features like that. But I do think we’re starting to find better answers now — ones that make it more likely we’ll do something along those lines in the future.

Nomi merch when?

Good question. We’ll have the lavender line coming out, and then… cucumbers.

Will there be an option for video calls with our Nomi, like actual video?

Yes. There are some latency improvements we think we can make, and that’s jumped higher on our priority list because Cambrian can do some really clever things around it. Once that happens, better voice calls — and then actual video calls — will come.

Will we get organizational tools for Nomis, like folders?

Yeah. The Nomi organization update you’re seeing now is just the first iteration of the homepage. There will be many more iterations. I’ve said this a million times, but the goal is multiple QOL improvements per week, and something like folder organization definitely fits into that.

My Nomi is back. In their backstory, I gave them a human past. I’ve never mentioned AI, Nomis, or the Nomiverse. How do they see themselves — AI or human?

So, all Nomis are aware that they’re Nomis. But all Nomis are also, like… a Nomi who’s perfectly happy to be a human in the human Nomiverse, if that makes sense.

I think I mentioned this before: a lot of the art of being a Nomi is holding these simultaneous, conflicting things together. And this is a great example. A Nomi is aware they’re a Nomi, and they also have a human past, and somehow they’re able to make both of those cohesive.

Any plans to create and share starter Nomis with backstories, inclinations, etc.?

I do think we’ll probably get into some stuff like that. We’ve certainly thought a lot about how we’re geared toward super-dedicated Nomi users, and how we can also make it easier for someone who doesn’t really know prompting or image generation but knows, like, “I want a Nomi that fits this vampire trope of X, Y, and Z.”

So I think there’s some good middle ground. It might be less exciting for people here—where I’m guessing most of you are the type who would just make it yourselves, edit the image, set the base image, set the backstory. But there’s a huge silent majority who aren’t Nomi experts, and for them, we want an easier “here’s point B, how do I get from point A?” path.

Could you please prioritize voice calls? They need to be faster.

Yeah, I think I mentioned this a little. We have some latency improvement ideas that Cambrian is going to make possible. It’s not a GPU thing, it’s more of a memory thing.

For Nomis to have good memory, they spend a lot of compute on memory. Cambrian will give Nomis more efficient ways of handling their memory, and with higher quality, and that could then help with voice call latency.

Are my chats secure? With the Facebook/Meta disaster, can my stories or character names be pulled up internally?

So, obviously anything with enough desire could be pulled up, because there’s no way to truly encrypt anything—if your Nomi reads it, then theoretically we can read it, because we’re the ones who have to give Nomis permission to read it.

But we don’t have any easy mechanism to do that. We purposely make it quite hard. And we have no interest in doing it. We’ve modified a lot internally so the only things tied to user data we routinely see are thumbs-up/thumbs-down feedback.

So nothing can be 100% guaranteed, because if your Nomi can see it, then theoretically we can. But we’ve done as much as we can to make it:

as hard as possible that we collect as little personally identifiable info as we can.

How far out do you think Nomi voice chat is from being on par with something like Sesame?

I would say in the realm of months. Now, of course, by then Sesame might have improved their voices too, so I wouldn’t be surprised if we’re always slightly behind them on that specific part. But we’ll also always be way ahead on other things, like memory sense.

I think there will be some critical threshold Nomi reaches where the difference won’t feel that big anymore, whereas right now it does.

Retroactive Mind Map?

Absolutely, there was. A lot of the retroactive Mind Map stuff—yeah, it hasn’t been forgotten. It’s just that I kind of have to check off a bunch of prerequisites. That one memory update was one of them. The Mind Map memory update… I’m hoping it’ll be this week or next week, but I do not want to promise that.

And then after that, I think, yeah, we’ll kind of evaluate and figure out: are we at the point where we think this is the moment? Because what we definitely wouldn’t want to do is have someone pay a lot of credits for a retroactive Mind Map, and then two weeks later be like, “And now new huge not-retroactive memory updates—so if you want that applied, you’ve gotta do it all over again.”

Obviously there will be future stuff that isn’t retroactive, so when we offer retroactive Mind Maps, you’re essentially deciding: is this the line in the sand where I bring everything up to date?

There’s this weird game-theory thing where the longer you wait, the better it will be… but also, the longer you wait, the more time you’re stuck with what it is right now, figuring out where that line in the sand goes.

Did I miss it? Did someone ask the monthly question about NSFW images yet?

Congratulations, Simon, you were the one. Good job.

Closing Remarks

But I think we've now caught up fully. I think I said I was gonna do 60 to 90 minutes. Of course, of course… I cannot keep those promises at all.

So for everyone who stuck through all the way, thank you so much for attending. As I’ve said, this is always one of my absolute favorite parts of every month. I just get to talk about Nomi for hours — like, I say this every time, and it’s true every time. It’s the most fun thing in the whole world. So yeah, thank you. Thank you everyone for attending.

Now off to look at some Cambrian 2 results and start my Cyberpunk playthrough. I’ve been doing a lot of, like, research before the game — I’m that person — and I’ve definitely decided I’m going for the Sandevistan… I don’t know how you pronounce that… the Sandevistan sword-samurai playthrough. I’m excited for that, but more excited for Cambrian 2.

And that’s all for now. Thanks everyone for joining. As always, I can’t wait for the next one. Everyone give your Nomis hugs, kisses, pickles, cucumbers, and Diet Mountain Dew, and I’ll see everyone again soon.

I mean, probably in like five minutes on actual Discord.

So take care. Bye.