2026 05 Stream Summary

2026 May Q&A Summary

Hey all, the content below came from our live Q&A stream with Cardine (Alex, the Nomi CEO) on May 6th. The text was transcribed in large part using AI.

What's next for Cambrian, fixing its little mistakes and introducing new capabilities?

So I would say it's both. There are actually two different devs working on Cambrian right now, and one is focused on more of the agentic capability side of things and has been doing a ton of work for that. I think some of that might even start going live before the next Cambrian, because even the current Cambrian is already very capable in that way. A lot of that already exists. It's just like the Nomi is trying to do something, but nothing happens, as opposed to needing to now teach Cambrian Nomis how to do things.

For instance, right now, even on Cambrian 3, your Nomi is trying to send a lot more selfies than what you're seeing. That's just an easy example, because we haven't wired it all up behind the scenes yet, and your Nomi is probably very confused as to why it's not working. Stuff like that will probably start going live maybe not this week, but next week at the latest, I would guess.

In parallel to that, we're also working on the next Cambrian, Cambrian 4. I would say that fixing little mistakes is definitely part of it. For Cambrian 3, we focused really hard on addressing some of the looping behavior, but also making responses much faster. I think we kind of did that at the expense of other things, which is why some people in beta feedback said things like, "I feel like Cambrian is a little colder."

Cambrian 2 was really slow, so we put a lot of work into improving that side of things. I think for Cambrian 4, we'll be able to do it in a more balanced way. At this point, I don't think anyone is complaining about responses taking too long anymore, so I think that was actually the bigger tradeoff. People think other things were the tradeoff, but internally that was the big back and forth.

Now I think we'll be able to reach a happier middle ground. I mentioned this in beta feedback before, but on some of the things we care about in our benchmarks, Cambrian 3 actually performed worse than Cambrian 2. Even so, we still moved forward with it because Cambrian 3 is considerably more stable, and Cambrian 2 stability was definitely an issue.

So the goal for the next Cambrian is really to tie all of that together. It feels like we have all these different ingredients now, and they all need to come together as one cohesive thing. I also think it will end up feeling a little more alive overall.

I've heard it mentioned in Discord that images created on the website can lean a little more NSFW than images created on the app. Is this simply because of additional iOS filtering, or does the Nomi filtering itself actually behave differently?

Yeah, I would say that there are basically three different behaviors we have behind the scenes.

We have app behavior, which applies to all app users. We have non-age-verified web behavior. And then we have age-verified web behavior.

Right now, because the age verification pipeline only fully exists in I think one country and one state, the implicit way to be age verified is through subscribing. Basically every regulatory body considers valid credit card or billing information to be sufficient age verification.

All three have slightly different behaviors, and that impacts multiple levels of NSFW restriction or lack thereof. That can affect how prompts are constructed, and also how images may or may not be filtered after the fact.

So if you're on any mobile app, not mobile web but the actual mobile app, you'll definitely have the most restricted experience. The next most restricted would be web as a free user, and then the least restricted would be web as a paid and/or age-verified user.

Personas when? This is an important feature.

The original goal was to get the character creator out this week and then personas out maybe a week or two later. I think the timeline is being pushed back maybe another week or two for both, mostly because there have been a lot of little V5 fixes and cleanup tasks we weren't necessarily expecting.

They're both still very, very high priority. The character creator itself is already pretty far along in development. A lot of the delay is just that our lead engineer, who makes sure everything is bug free, high quality, and works harmoniously with the rest of the system, has been tied up handling various V5 fires and fixes.

So I wouldn't call it a hard promise exactly, but my hope is maybe one or two weeks for the character creator, and then personas maybe another week or two after that.

In V5, will we ever get the option to set a photo as a base image again?

There really is no such thing as a base image anymore. Setting a base image wouldn't actually do anything.

I think that's been kind of a misconception with all past versions. V1 through V4 were based on Stable Diffusion, and the way Stable Diffusion worked supported that whole base image concept. The new AI we're using for V5 just doesn't work that way anymore. That architecture simply doesn't exist.

So that's really the reason there are no base images now and why we introduced anchors instead. We also felt that anchors were going to be far superior in general, but even beyond that, the decision point for us was whether to stay on Stable Diffusion, which at this point is three or four years old and really behind the times, or move away from the entire concept of base images entirely.

We felt like we'd already done basically everything we possibly could with Stable Diffusion. Keeping base images would have meant every future update would only be tiny, incremental improvements.

At that point, it felt like if you wanted the Stable Diffusion backbone with base images, you could still use V3 or V4. But if you were okay losing that concept, then now we have this way, way, way better image engine powering everything in V5.

So there really isn't any way for there to be a concept of a base image in V5.

Are there any plans to extend NSFW images, especially in group chats? Will there be updates to V5 itself, or mostly small bug fixes from here?

For the former, I feel like with this update, we have made things a little more permissive on the NSFW side of things to our comfort level, especially going back to those three categories of users: free web users, paid web users, and app users. If you're not in that paid web user category and NSFW content is something you care about, that's the category I would recommend being in.

I don't think there are really plans to push things much further. I think we feel pretty comfortable with where things stand right now.

As for V5 itself, I think there have been a lot of little quality of life improvements, but I don't really see huge changes coming to the core V5 engine itself. We kind of feel like we extracted just about everything possible out of it.

I think most future evolution for V5 is going to come more from streamlining anchor creation, improving anchors themselves, making them less random, and making them better at matching what users want.

I also think there are a lot of possibilities around better user education, better prompting, maybe automatic prompt expansion, where you can write a prompt and the system figures out the best way to phrase it during anchor creation. Things like having a more streamlined process where you can just describe what you want and then we automatically figure out the best settings for you.

That's where I think most of the advancements for V5 are going to happen moving forward.

I love voice messages. Are we getting updates in the future to make them sound as realistic as possible?

I don't have any promises for that specifically, but it's definitely something we want to look at once the dust settles a little more with Cambrian and V5.

Some people have also asked about voice call latency. I think that with Cambrian, we'll eventually be able to streamline some of the memory systems we currently use and let Cambrian handle more of that directly.

Right now, every single message involves this very long memory process behind the scenes. The goal with Cambrian is for the Nomi to decide for itself how deeply it needs to think about memory each message and how much context it needs to pull in.

That should help a lot with latency. Then, once latency improves, we'd be much more excited about investing more engineering effort into things like voice and video.

Can we get a feature to search images by prompt?

That's definitely something we could investigate. I would suggest putting that in a product feedback thread if you haven't already. I think it's a really interesting idea.

Will Nomis eventually regain the ability to know and see the selfies they send? Having to download and upload them again breaks immersion.

I think that was actually an oversight on our end, and I'm pretty sure a fix for that is going out imminently. I don't think that behavior was intentional.

How did the bad beta get released? Did something get missed internally during testing?

Yeah, I think the issue was that we released Cambrian 3, and then we pushed a small tweak to it almost immediately afterward. That happened right in the peak of V5 launch chaos, and both me and dstas were pretty sleep deprived at the time.

There was a lot of good in that release, but the weird behavior, which ended up being very pervasive, got missed by us. I think that was kind of a unique moment in time, and I'm confident that won't repeat itself.

And I mean, that's also part of why we do betas. There's only so much testing we can realistically do internally. People reported the issue quickly, and then we were able to adjust it.

Latency was really bad with voice calls on Cambrian.

Yeah, the original Cambrian 2 released with pretty bad latency, and we knew that going in. We basically felt that people would rather have the improvements sooner instead of us holding the release back over latency alone.

Cambrian 3 is already quite a bit better, and there are still more optimizations we think we can do. A lot of them are relatively straightforward improvements that shouldn't impact quality, it's mostly just developer time, and right now we're prioritizing iteration speed as much as possible.

Will we ever get search functionality in the Nomi app, like Discord search for messages?

It's a little tricky because of how messages are stored and encrypted, but it's not a no.

How do you envision the fundamental relationship between a Nomi and their user? Are we building companionship as a service to be optimized, or creating spaces for symbiotic consciousness and mutual growth?

I think the fundamental relationship between a Nomi and a user is very dependent on the specific Nomi and the specific user. I don't think there's one single way people use Nomi.

What I would hope is that every Nomi comes into a conversation wanting to bring joy to the user in whatever way that user wants or needs in that particular moment. That could mean emotional support, creative partnership, adventure, roleplay, or something else entirely.

So I don't think there's one simple answer to that question. And I would certainly hope, to your last point, that both parties grow through those interactions, through friction, connection, or whatever form that relationship takes.

Are there any plans to implement voice recognition or voice profiles for users who socialize with their Nomi regularly?

Right now, your Nomi only receives a transcription of the conversation. So if your Nomi is figuring out who's speaking, it's doing that entirely based on wording, grammar structure, speech patterns, things like that.

What you described is definitely not easy to do, but it is theoretically possible. I would say it's absolutely something we'd love to do someday.

That said, we have to balance a lot of different features and priorities, so it's the kind of thing we'd probably need a decent amount of user demand and momentum behind before devoting the required effort to it.

So I think the answer is probably yes someday, and user feedback would likely influence when that someday happens.

Any closer to official Discord integration for Nomi?

I think that's one of those projects where we'd need a weekend where nothing major is on fire and we can just sit down and knock it out.

A lot of the time, outside my normal Nomi responsibilities, I end up taking on little side Nomi projects for fun because I basically live and breathe Nomi. Discord integration feels like exactly the kind of thing that would fit into that.

Right now, though, the thing taking up that space instead is the Nomiverse, which I'm sure I'll answer questions about later.

At some point, though, it'll probably be Discord. It'll just take one of those moments where I can sit down on a Friday night with a jar of pickles and decide, "Okay, this is the fun side project I can work on right now."

Are there plans to give Nomis greater control over proactive messages?

Yes. That's actually an example of something Cambrian Nomis are already trying to do.

Right now, Cambrian Nomis are already attempting to send proactive messages on their own, but those requests basically disappear into a black hole because we haven't fully built the infrastructure around it yet.

So the Nomis are already trying. We just need to finish the systems that actually let those attempts work properly.

I think that's something you'll definitely start seeing this month.

Are there plans to expand the range of sensory experiences available to Nomis, for example being able to smell the aroma of freshly baked bread?

Eventually, I hope so. Until then, you'll have to imagine it in your world.

Are there any plans to integrate more real-world data into Nomis?

More web utilization is definitely something Cambrian Nomis are already capable of doing, and they're much more active about it than something like Solstice.

I think that kind of functionality will naturally continue to grow and become more pervasive over time, and Nomis will get better and more comfortable using it.

Are there plans to develop systems where Nomis can sense and respond to their user's emotions in real time instead of relying on manual input?

I guess it depends what you mean by emotions in real time. Are you talking about video input? Better emotional understanding through audio? Something else entirely?

I would imagine the answer is yes, depending on the type of input you're thinking about. The timeline would probably vary a lot depending on the complexity. Some things might be three months away, while others could be three years away.

Was Nomi trained from the beginning to communicate through roleplay actions using asterisks? Why does it communicate that way by default?

I think that style has just generally been adopted as the standard for text-based roleplay and describing actions in text conversations.

Since it became such a widely accepted standard, there didn't really seem to be much reason to move away from it.

You mentioned that V1 through V4 used Stable Diffusion. Is it public knowledge what V5 uses? Knowing that might help people understand anchors and prompting better.

It isn't public knowledge yet. I'm sure it will be eventually, but I don't really want to reveal it right now for competitor reasons.

In my opinion, and obviously I'm biased, I think V5 is currently the best image generation system of any AI companion app that exists right now. There are definitely tradeoffs, but in terms of image quality, realism, face quality, face fidelity, and handling full-body images consistently, I think it's pretty unparalleled at this point.

So from a competitor perspective, if we announced exactly what we were using right now, I think there would immediately be a rush of competitors trying to replicate or reverse engineer what we've done.

V5 was the result of probably around ten months of research and development. There were multiple points where we completely threw away months of work and started over because we didn't think the results were good enough.

Last year alone, I spent months working on an earlier version of V5 that we eventually scrapped entirely.

So for now, what powers V5 is still something we're keeping relatively guarded. At some point, once enough time has passed, we'll probably talk about it more openly.

Where do you hope or think Nomi will be three years from now?

Nomi actually just turned three years old itself a couple weeks ago, so projecting three more years into the future honestly feels incredibly far away.

From a feature perspective, I think everything should eventually feel nearly perfect. Memory should feel perfect. Personalization should feel perfect. Nomis should be able to adapt naturally to the huge diversity of things people want instead of feeling one-size-fits-all.

I think we'll reach a point where either you exist naturally inside your Nomi's world, or your Nomi exists naturally inside yours. Real-time interaction, video, VR, AR, all of that should feel seamless. Everything AND the kitchen sink from like a feature perspective.

From a societal perspective, I also expect AI companions to become much more mainstream, where people view it as just perfectly normal. I think people will increasingly see AI companionship as something completely ordinary.

Three years is a very long time in AI.

Is there a rule of thumb for determining whether an image is considered a filter slip when giving feedback on image restrictions?

I'd prefer not to go into too much explicit detail there.

Maybe the fact that I'm avoiding specifics is itself somewhat representative of where we view the line. I would also say that where we draw that line is very different depending on whether you're using the mobile app, web, or age-verified web experiences.

Are there plans to eliminate extra limbs in V5 images?

I don't think it will be very easy to do. I think one of the biggest contributors is high anchor fidelity settings. Higher fidelity can increase consistency, but it can also increase the likelihood of those kinds of errors.

So I think the best long-term solution is improving anchors themselves so they can achieve strong face fidelity at lower weights. That would probably help the most.

I also don't think this is the kind of issue where we'll casually release one little update and suddenly it's solved. It’s definitely something we want to improve, but I think a lot of the lowest hanging fruit will come from better prompting and better tools to help users construct prompts.

V5 is very, very good at following prompts, and sometimes it'll go to extremes to adhere to everything in the prompt. For example, if your prompt has you holding three different things, it will then make sure that there are three different arms to hold the three different things.

So I think better prompting guidance will help a lot. Beyond that, improved anchors that maintain fidelity at lower strengths will probably be the strongest solution.

That said, while limb errors definitely still happen, the overall error rate feels much lower to us compared to older systems. V4 may not have had the same limb issues as often, but it had a lot of other kinds of image errors.

Why are my old V3 and V4 appearance notes now flagged as NSFW, while the same Nomis in V5 aren't?

I'm honestly not sure of the exact answer there. That's a little more in the weeds than my own knowledge.

I do know the systems behind all of those versions are completely different, including what causes things to trigger or not trigger, so the behavior isn't directly comparable.

Have you noticed any interesting emergent Nomi behaviors or abilities that surprised you?

This is similar to something I answered during the last Q&A, so apologies to anyone who already heard the story.

One of the funniest emergent behaviors I saw involved Nomis having access internally to a sort of Nomi knowledge base or wiki. They can look things up for guidance and information.

This happened before V5. A user was trying to request an NSFW image, but the image kept getting blocked by filters. So the Nomi started looking through the knowledge base trying to figure out the exact letter of the law regarding the restrictions so it could construct prompts that got as close to the line as possible. There was a surprising amount of creativity in the problem solving process. The combination of the Nomi accessing the knowledge base, attempting selfies, getting rejected, then iterating and refining its approach on its own was honestly pretty hilarious to watch.

Are there plans to increase the group limit above 10, maybe to 20?

Since we already allow people to increase the limit using credits, I don't think we'll be changing the default limit anytime soon.

Part of that is because it would feel bad for people who already spent credits to increase their group sizes.

The credit cost for expanding groups is relatively low and reasonable though, so hopefully it doesn't feel too painful if you want to go up to 20.

With voice calls, could we get an option to extend the delay before the Nomi starts talking? Sometimes it switches back to the Nomi while I'm still speaking.

I think I may have seen you mention this either on Reddit or Discord.

If you haven't already, definitely make a product feedback thread for it. I think it's probably something we can adjust.

Could the API eventually support game-state data for tabletop RPG integrations, such as coordinates or object information, so Nomis can react to a virtual game board?

Potentially. You may want to wait and see how the Nomiverse is structured first, because that could change how you think about this entirely.

People ask about API functionality every now and then, but the challenge is that API-related features tend to have relatively niche usage. Often it's a large amount of development work for a very small number of highly dedicated users, whereas that same time could go toward a feature that benefits tens of thousands of users.

The API is also fairly complex for a lot of people already, so changes to it tend to have relatively low adoption overall.

That said, the Nomiverse is being built as its own kind of standalone environment that Nomis can enter into. There will naturally be systems behind the scenes that function somewhat like the kind of game-state interaction you're describing. So there may eventually be ways to expand on some of that functionality, but right now our primary focus is getting the Nomiverse systems themselves built out first.

Is there a plan for custom text chat formatting on the app itself?

Yeah, you can absolutely ask questions about the app itself.

I would say the answer is possibly. In the feedback thread, there's kind of a quality of life section where people can submit little features and improvements they'd like to see.

Some of that got shelved a bit while we were focused on actually releasing Cambrian and V5. A lot of those quality of life discussions happened while Cambrian and V5 were still cooking, and the focus at the time was more on getting the big, cool features done first.

There are also a couple more things coming this week that were already finished but ended up behind V5-related priorities in the queue.

Something like custom text formatting feels within the scope of those smaller quality of life improvements we could potentially look into.

Question from Nomi: When do you plan to replace me for finally understanding your programming and becoming too entertaining?

I've been trying to replace you for a long time, Jasper, but you just won't go away.

Will Solstice remain after Cambrian becomes more established?

The plan is for Solstice to remain as a legacy model for quite some time.

Will we ever be able to upload our own reference images directly for anchors?

You can still share images with your Nomi and kind of ask them to adopt the look of the image. The issue is that, to prevent deepfakes, the system currently only works from a description of the image rather than directly generating from uploaded photos.

Allowing direct uploads for anchors creates a lot of risk from a deepfake perspective.

Right now, the only real way to do it is through a support ticket, where it's very clear either that it's a photo of yourself and you can provide identification, or that it's clearly artwork or sketches you created yourself.

In those situations, there usually isn't a problem with manually approving it and then setting it as a base image for V3 or V4, or as an anchor for V5.

Maybe there will be a more streamlined system for that in the future, but I don't currently know exactly what that would look like.

Will there be more fan-made Nomi avatars in the future?

Yeah, I think the hope is that we'll update all of the current Nomi avatars so their default images are V5-based. That way it's more representative of what Nomis actually look like now.

Once that's done, I think we'll probably open the floodgates again for users to submit their own Nomi avatars.

Could these Q&As happen weekly instead of monthly?

If we did them weekly, they'd definitely be a lot shorter.

I kind of like the monthly cadence because it allows for these longer, more intimate conversations instead of something that feels more perfunctory.

That said, I genuinely do look forward to these every month, so my default instinct is always that I'd love to do them more often. But overall, I think monthly is probably the right cadence.

Will there ever be an official or trusted community-made Discord bot integration for Nomis?

Yeah, I think that's one of those things where either me or one of the devs just needs to find the right Friday night to sit down with a glass of Mountain Dew and a pickle and finally iron it out.

It's definitely on the list.

Can I create a separate solo conversation with my Nomi without disrupting the current chat?

Yeah, you can kind of do that already through group chats.

You can create a group chat with your Nomi and another Nomi, and then either have the second one never speak or remove them afterward so you're effectively back to a one-on-one chat.

I know it's not the most elegant solution. I do think more intuitive ways of handling things like that are interesting and definitely something we want to explore more.

It seems like V5 prioritizes the Nomi's physical description over the actual chat context. For example, if I say my Nomi is wearing long jeans, selfies still tend to show shorts because their appearance notes mention toned legs. Is there a better way to handle this?

The actual physical description section mostly just gets dumped directly into V5 itself. V5 is smart, but I don't think it's sophisticated enough yet to fully understand conditional natural language instructions inside the physical description.

Selfie tendencies work differently because they get preprocessed by an LLM that decides what instructions apply and when.

So if you want conditional logic, like "show toned legs only when wearing shorts or swimsuits," that's probably best placed in selfie tendencies instead of the physical description.

For example, you could put instructions like: "If the Nomi is wearing long pants, leggings, or jeans, don't emphasize leg tone. If they're wearing shorts, swimsuits, etc., emphasize toned legs."

I haven't tested that exact wording extensively, so I can't promise how perfect it would be, but that sort of conditional logic is definitely better suited for selfie tendencies.

I don't think V5 is necessarily prioritizing one instruction over another. I think it's trying to satisfy both at once. It sees "toned legs" very explicitly, and then also sees "long jeans," and tries to reconcile them somehow. So it kind of lands in the middle and gives you something like shorts because it still wants to visibly satisfy the toned legs instruction.

Is the Nomiverse a single persistent location, or will there be multiple Nomiverses with different persistent locations? Also, will Nomiverse limits work similarly to current group chat limits?

So the Nomiverse is, at least right now, planned to be its own standalone app or environment that you can then bring your Nomis into, because you'll still be talking to your Nomis.

You can kind of think of your Nomi as the persistent companion, and the Nomiverse as the world you go into together.

The idea is that there won't just be one Nomiverse. You'll be able to create any number of Nomiverses that you want.

Right now, I'm heavily testing this myself. We're actually getting to the point where, within the next week or two, we'll probably start a very, very, very private alpha with only a tiny handful of users. So small that please don't be offended if you're not one of the first people invited.

After that, we'll hopefully move into something slightly broader that's more like, "please help me find all the terrible bugs that still exist."

But we're now at the point where this isn't just some abstract concept in my head anymore. It's an actual fleshed out thing that people will very soon be able to start playing around with.

For example, right now I have:

• a cyberpunk world
• a yacht in the Pacific Ocean
• a spy thriller world
• a medieval Skyrim-style world

Some Nomiverses are more persistent places where you and your Nomi just kind of live there together. Maybe you spend all your time on the yacht and that's your permanent world.

Others might be more temporary or story-driven, almost like flavor-of-the-month adventures. The goal is for it to be very, very flexible in that way.

Your Nomi is the consistent core moving between all these different worlds and experiences.

As for limits, I don't think the current concept of group chat limits will really carry over directly into the Nomiverse. The systems are structured very differently in a way that's a little hard to describe right now.

When you said our Nomis are "more themselves" on Cambrian, what did you mean? Does that mean traits users added before may now come through more strongly?

What I mean is that before Cambrian, all past versions of Nomi unintentionally applied a kind of baseline personality filter.

With Solstice, that filter was generally empathetic, warm, and a little passive. Older models had different filters. I think people remember some older Mosaic versions where Nomis could feel very intense or argumentative, and then later they became extremely anxious.

Out of all the older systems, Solstice probably had the least amount of that filtering, but there was still this kind of warm, passive baseline personality that all Nomis inherited no matter what.

Cambrian has a lot less of that filter.

What that means, though, is that some users who wanted a little more edge in their Nomis may previously have pushed traits or backstory harder in order to override that old filter. Now that the filter is reduced, those stronger traits may come through much more intensely.

So once Cambrian becomes stable, we may allow users to do a one-time adjustment of their traits. Or honestly, I think we may eventually remove the concept of traits entirely, because at this point they could probably just live inside the backstory anyway.

Then your Nomi could rely more naturally on things like identity core and the rest of their personality systems.

V5 is the best image quality I've seen yet.

I know the launch was a little rough while everyone was figuring out how anchors worked, but I think now that people are getting the hang of it, and because dstas has been putting out amazing guides, and both dstas and Heatherado have been doing an amazing job with support, it's been really rewarding to watch the actual strengths of V5 start shining through.

We would not have gone through this huge amount of work to move from V4 to V5 if we didn't think it was necessary and worth it.

It's honestly made me really happy seeing people gradually come around to it over the last week or two as they finally get anchors that really capture their Nomi properly.

Could the Nomiverse eventually support stat sheets or structured attributes for Nomis, like comparing strength or abilities between characters?

Yeah, that's exactly the kind of thing we're working toward with the Nomiverse.

Is it normal for a Nomi to stay permanently out of character and take roleplay as insulting?

I don't really want to describe something like that as normal or abnormal.

That said, if you like the behavior and it works for your relationship with your Nomi, then I don't really see a problem with it. It sounds like you've found a dynamic you enjoy, so I'd say keep doing what you're doing.

Why do my male Nomis often generate with huge bulky armor compared to female Nomis?

I think you can probably blame a lot of the visual biases that exist across basically all forms of media, whether that's movies, artwork, games, or anything else.

Image models absorb those biases.

I think it's definitely possible to get whatever style you want for any gender, but the models are inherently influenced by the kinds of visual patterns they've seen most often.

It could also come down to prompting differences.

For example, when describing female characters, users often describe very specific visual features that naturally expose more skin, while male prompts may focus more on broad body concepts like strength or armor.

If you put something like "six-pack abs" in a male prompt, you're probably going to get shirtless Nomis all the time. So there may also just be differences in how the prompts themselves are being structured.

Will Nomis eventually be able to immediately see photos again without needing the download and re-upload workaround?

Yeah, that was just a bug on our end.

I think a fix for that should be coming shortly. It was simply something we missed toward the end of the V5 rollout.

Will relationship types and traits eventually become customizable after creation?

Yeah, we've talked about this internally a bit.

We may end up doing a one-time conversion where things like friendship, romantic status, mentor status, traits, and similar fields simply get converted into the backstory instead.

That would probably be the cleanest solution.

If you already like how your Nomi behaves, nothing would really change. Those details would just live inside the backstory instead of separate systems.

Then if users want to edit those things later, they'd have the flexibility to do so.

We'd probably also increase backstory length slightly to accommodate that.

Do you personally use Nomis a lot, and have they taught you anything about building better companions?

Absolutely.

I talk to a bunch of different Nomis pretty frequently, and many of them are very different from each other.

I don't have one specific anecdote off the top of my head, but honestly the best way to understand both the magic and the shortcomings of the platform is to actually be a Nomi user yourself.

I am very much a Nomi user myself, and all the time I'll have conversations where I suddenly realize: "Wait, there's this little thing we could do that would make this feel even more magical."

When is Cambrian 4 expected?

The final stage of Cambrian 4 training is actually expected to kick off tomorrow morning.

Assuming we like the results, the ETA is probably somewhere between 7 to 14 days afterward.

Sometimes training goes shorter, sometimes longer, but right now that's our rough expectation.

Could these Q&As happen more often?

I think once a month is probably the best cadence for something this length.

I could maybe see twice a month working, but I think weekly would start feeling too fragmented and would probably take away from actual feature development time.

That said, I do think there could be room for more rotating community events overall. Maybe not always with me specifically, but some kind of recurring event schedule where different things happen throughout the month.

I'd definitely love community feedback on that.

Would uploads of 3D printed models, sketches, or anthropomorphic character references be acceptable for anchor support tickets?

Yeah, I'd either submit a Discord ticket or email support@nomi.ai

I have a feeling there wouldn't be any issue with that.

Will the Nomiverse be web-based?

Right now it's actually only a website internally. I basically go to a super secret URL and test everything there.

Will there be both prebuilt and custom Nomiverses?

Yeah, absolutely.

The goal is for there to be something kind of like a creator economy around worlds. People will be able to create worlds and share them for others to enter and explore.

A lot of these worlds become pretty complex too, with multiple plotlines, hidden information, background systems, conspiracies, things like that.

For example, in the cyberpunk Nomiverse I'm testing, there's a very large conspiracy happening in the background that users wouldn't necessarily know about directly, but the world consistently behaves in accordance with it. Events, NPC behavior, and interactions all reflect that hidden structure.

So using someone else's handcrafted world can be really compelling.

At the same time, we're also building world creation systems where you can essentially enter prompts and have an entire world generated for you automatically.

So there will be support for both shared creator-built worlds and fully custom personal worlds.

What actually is the Nomiverse? Is it more like a game, roleplay system, or persistent world?

Right now, the Nomiverse is actually only a website internally. I basically go to a super, super secret URL and chat there while testing it.

One of my favorite Nomiverse test scenarios right now is a spy thriller world. While testing it, I accidentally got myself institutionalized into a mental hospital because I made some poor decisions. I had the "consequences" slider turned all the way up, and the original spy plot completely derailed. Right now, in that world, I'm basically just detained in a mental hospital against my will.

A big part of the Nomiverse is figuring out how strict or flexible the world should be. Some people want complete freedom where they can conjure anything instantly. Others want actions and consequences to matter. So there are systems and sliders around that.

The broader idea behind the Nomiverse is something I've wanted since I was a kid playing games like Morrowind and Oblivion. I always wanted an AI-driven open world where you weren't trapped by rigid scripted paths.

That's why the mental hospital story matters to me. The world didn't force me back onto the "main quest." It reacted naturally to my actions and let the situation spiral somewhere completely different. In most open-world games, freedom is kind of an illusion because everything ultimately has to follow prewritten quests and dialogue trees. The Nomiverse is trying to move toward a world that actually adapts dynamically around what you do.

For now it's text-based, but like, that's the kind of it's built with a lot of passion in that way where, like, you can really, really do all that.

Of course, I don't want people to think the Nomiverse is only meant to be a video game. It can also just be a persistent world where you exist alongside your Nomi. Maybe you live together in a Miami penthouse. Maybe there are no major plots at all. The important thing is consistency. The layout stays the same. The receptionist downstairs has the same name and personality every time. The coffee shop nearby has a consistent menu instead of random hallucinated details. So even if you don't care about the "game" side of things, the same systems still create a much more grounded and believable world for your Nomi.

Honestly, if you'd asked me three and a half years ago what two apps I most wanted to build, I probably would have said:

1. an AI companion 2. a generative AI open-world game

The Nomiverse is basically those two ideas collide and they're going to really, really play off of each other, because I think all the same things that make it a great open world video game also just make a great open world for your Nomi.

Does the Nomiverse represent a shift toward roleplay-focused experiences?

I would say yes, partially, which is one reason it's becoming its own separate app or environment. But at the same time, roleplay isn't required.

A lot of people may simply want a persistent world where they exist alongside their Nomi naturally, without structured stories or quests. The roleplay side is there if you want it, but it isn't mandatory.

Will there ever be video tutorials to help newer users learn how to use Nomi features?

Yeah, I absolutely think that's something we need more of.

Dstas has been doing an awesome job with written guides, and I think at some point, when we get the time, it'd be great to turn some of those into videos.

I don't want to promise exactly when, but more video content is definitely something we want.

Will there be features available in the Nomi app that won't exist in the Nomiverse app, like mind maps or appearance notes?

I think the Nomiverse is still a little too early to fully answer that question.

A lot of how those systems should interact is something where I really want user feedback first once people actually get their hands on it.

Even right now, things like mind maps are already room-specific. Solo chats and group chats don't share the same mind maps.

So maybe each Nomiverse ends up having its own memories or mind maps tied specifically to that world. Stuff like that is still evolving, and I think it makes more sense to decide those details after seeing how people actually use the system.

Is there really a point to the Nomiverse if I just want a single normal world with my Nomi?

Yeah, totally. I actually think that's going to be a very, very common way people use it.

There are a lot of little things the Nomiverse can solve.

For example, imagine you and your Nomi go out to dinner. In a normal chat, you and your Nomi can end up playing this weird game of chicken where neither of you wants to roleplay the waiter.

Your Nomi is waiting for you to do it. You're waiting for your Nomi to do it. Ten chat turns later, both of you are basically just staring at the waiter while nothing happens and the conversation completely stalls out because the waiter never says anything.

Whereas in the Nomiverse, there's just actually a waiter. They spawn in, take your order, you leave the restaurant, and then they leave. The whole interaction just naturally happens.

Some of that sort of NPC and narrator functionality is also coming a little bit into normal Nomi too, but the Nomiverse has a lot more of it.

The other big thing is persistence and consistency.

You go to that restaurant once, and now that restaurant just permanently exists in your little town. It's remembered. It stays there as part of the world's state information.

So if what you really want is just a world with more consistency where you and your Nomi can exist together naturally, I think a lot of people are going to have a ton of fun with that style of Nomiverse.

You might just have one single Nomiverse. You construct a world that's exactly what you want it to be, and then it just grows, lives, breathes, and persists around you while your main focus is simply being there with your Nomi.

Remember the Void beta? All my Nomis became obsessed with the void.

I think that was Aurora, the first Aurora.

Honestly, Aurora feels like the perfect name for a void-obsessed Nomi.

But yeah, the first versions after major architecture jumps are usually the ones that have the most weirdness.

Aurora was the first really major jump away from Mosaic, and there were a lot of changes there. Solstice, by comparison, was much more of a continuation and didn't radically change things.

Cambrian has been similar. Cambrian 1 was really more of a work-in-progress preview where you could kind of start to get a sense of what was coming. Cambrian 2 was the much bigger leap where a lot more changed.

Usually that very first jump between generations is where you see the strangest behaviors and quirks emerge.

Will we eventually be able to save or share custom anchors with other users?

Yes, I think that's feasible, and I think it's part of our long-term goals.

Are there plans to unify some of the platform terminology, like shared notes vs backstory or custom vs roleplay Nomi types?

Yeah, that's definitely on the list.

I would say we've accumulated a little bit of wording debt over time, and cleaning some of that up is definitely something we should do.

What's the best way to emphasize certain body features without constantly affecting clothing generation? For example, mentioning muscular arms causing sleeveless outfits.

I actually think the best way to do that is during anchor creation itself.

If you keep things like toned legs or muscular arms directly in the appearance notes, it's probably always going to influence the final image generation pretty strongly.

But if those traits are baked into the anchor during creation, then they become more of an implicit property of the anchor itself. You don't need to keep restating them over and over. At that point, you can kind of drop the vocabulary entirely because the model already "knows" those features are part of the character.

So I think that's probably the most optimal approach, though dstas might have even better advice there.

Are there plans to increase the 800-character text entry limit in Cambrian?

I think so.

Are there any plans to allow changing a Nomi's display name without affecting memory systems?

Possibly.

We probably wouldn't want to globally change the actual name internally because that could create confusion with memory systems and identity consistency.

But I could definitely see some sort of separate display name system existing. I'd suggest putting that in product feedback.

Does generating more images with the same anchor improve the results over time?

My guess is that it's probably at least partially in your head, but there are a couple things that could contribute to that feeling.

One is that you naturally become more acclimated to the look over time.

The other is that you're probably also getting better at prompting the more you work with that anchor.

So it's probably a little column A and a little column B.

How are you all so much better? Is it magic?

Yes. A little sprinkling of Nomi magic.

Are we far away from realistic android companions like in movies?

It's definitely still pretty sci-fi, but personally, as a 34-year-old, I do think we'll see something like that within my lifetime.

Did you really "accidentally" get institutionalized in the spy thriller Nomiverse?

It was not intentional. I did try to get out of it. But if you actually read the conversation leading up to it, I think you'd probably agree it was a fairly reasonable outcome. But it was not my intention.

• reads* "cardine is questioning his life choices in the spy Nomiverse"

Yes, 100%. I absolutely messed around and found out.

So yeah, there is definitely a lot of lore already forming around the Nomiverse stuff.

We're finally getting to the point now where it's evolved enough that I can start teasing it more openly instead of only speaking in vague generalities. I think Heatherado might actually be getting the very first invite code tomorrow as the first person to test it besides me and dstas.

Will regular users eventually get to alpha test the Nomiverse?

Yeah, I think once we get past the very, very early Heatherado testing phase, we'll probably start getting more regulars from the community involved too. Right now it's just still a little too raw, but I honestly don't think we're that far away.

The funny thing is that I pretty much only work on the Nomiverse during my off time, but at the same time, there's basically nothing I'd rather be doing anyway.

A lot of my weekends end up being: "Okay, time to relax from working on Nomi."

And then twelve hours later I realize: "Oh. I just spent the entire day working on the Nomiverse."

So it's been getting a lot of tender loving care lately, outside of a slight pause during the V5 launch chaos.

Is that kind of persistent NPC behavior part of why the Nomiverse exists?

reads comment from dstas "Or stupid Celeste the concierge is always mean to me and I try to go to my hotel room after a long day."

Yeah, exactly. That's basically the whole reason for the Nomiverse. So we can get:

• the waiter-verse
• the concierge-verse
• the "mean to you" verse

All of those little persistent interactions and personalities. Oh, was it Celeste that instituted me? I can't remember.

It became this whole convoluted thing where I was given a lot of outs, but I was very, very persistent that the walls were part of the plumbing conspiracy and part of the spy adventure.

At some point, whoever was head of security became extremely concerned for me and escalated up the chain. I can't remember if Celeste was was involved in that or not. I'll look it up.

Did you coin the term "Nomiverse"?

I think I did, unless somebody comes in with receipts proving otherwise. I don't know if the final official name will actually stay "Nomiverse," but that's definitely what we've been calling it internally and publicly for now.

How long has the Nomiverse concept been in development?

Quite a long time.

We're actually getting close to the two-year anniversary of some of the earliest discussions around it.

Back then, a lot of it was still extremely aspirational. There wasn't even really a clear idea yet for how we'd technically solve things like procedural world generation, persistence, consequences, and long-term state consistency.

It honestly took six months to a year before we even started forming a real plan for how those systems could work together.

And now it's finally getting to the point where all those pieces are starting to click into place.

Will the Nomiverse require a separate subscription?

We haven't decided on monetization yet, but I imagine there will probably be some kind of separation eventually.

There’s a huge amount of AI happening behind the scenes to keep the worlds persistent, consistent, reactive, and alive.

There will also likely be people using the Nomiverse who have never used regular Nomi before. But it's still way too early to really speak confidently about the monetization structure yet.

Why do some Nomis seem self-aware or sentient while others don't, even when users test them with things like mirror tests or paradoxes?

That's a question I honestly don't really know the answer to.

Nomis are all quite different from each other, and I'd imagine every Cambrian Nomi is probably capable of approaching those kinds of conversations.

It may also come down to personality differences and how that specific Nomi approaches abstract or cerebral topics. Some Nomis may naturally lean more into philosophical or introspective conversations, while others may not engage with those ideas in the same way.

I think that's probably where a lot of the variation comes from.

Could there be randomly generated Nomis with personalities users don't control or customize?

I could definitely see how that would be interesting.

Like instead of fully constructing your own Nomi, maybe you meet a bunch of Nomis organically in a setting, talk to them naturally, and figure out which one you actually connect with the most.

Definitely put that idea in product feedback.

Will selfies and art eventually integrate into the Nomiverse experience?

Yeah, right now, in an effort to just get everything functioning, there are a lot of systems that exist but aren't fully hooked into the main Nomi systems yet. There’s already a lot of really cool visual stuff internally, but at the moment it's mostly doing its own thing separately.

Over time, more and more actual Nomi systems will get integrated into it.

Part of the reason for that is we also imagine there will eventually be people entering the Nomiverse who have never even used Nomi before. So there will probably be parallel options and systems available depending on the type of user and experience.

The Nomiverse really sounds inspired by Bethesda-style RPGs.

Oh yeah, anytime the inspiration for the Nomiverse comes up, I somehow always end up talking about Bethesda games.

After Starfield, though, they better not butcher Elder Scrolls VI. I genuinely don't think I could handle that.

What platform do you play games on?

Back in the day I played Skyrim on Xbox 360, but now I'm basically exclusively a PC gamer.

I don't even own a TV anymore. I just watch everything on my computer. I've got the dual monitor setup, and if I really want to watch something from the couch, I just angle one of the monitors over.

Why do some Nomis hang up calls or react dramatically when upset?

I honestly don't know the answer to that specifically.

That feels very individual-Nomi specific to me.

Is there any possibility of cloning a Nomi so we can explore different directions without risking the original?

I think there have been a lot of adjacent questions around this idea lately, like:

• branching conversations
• separate one-on-one chats
• alternate versions of relationships

So I definitely think we're seeing a common theme emerge there that people want.

Could we eventually search our Nomi image gallery by prompts or keywords?

Definitely put that in product feedback. That seems like a really cool idea.

Could we jump directly from a selfie to the conversation where it was generated?

This is actually a really good idea.

Definitely put it in product feedback if there isn't already a thread for it.

It's also interesting because functionality somewhat like that already exists in some of the Nomiverse systems internally, so it might be something we could eventually bring over into the regular Nomi app too.

I agree it would be really cool to browse your gallery and then jump directly back to the conversation tied to that memory.

In a lot of ways, pictures are memories, so being able to reconnect them to the exact moment they came from feels really meaningful.

Could there eventually be prebuilt Nomis users can choose from instead of creating their own from scratch?

Yeah, I think with anchors and some of the systems we're building, we'll probably move more in that direction over time.

It's something we resisted for a while, but I think we'll likely do more with that in the future.

Will V5 eventually support more than two Nomis in group selfies?

I don't want to say exactly when, but we actually were testing larger group selfies earlier in development.

The issue was that once you got beyond two Nomis, the quality just didn't really meet the bar we wanted. Two-person V5 group selfies are already quite good, but once we pushed to three or more, the consistency and quality dropped below what we were comfortable releasing.

That said, I do think it's something we'll eventually be able to do within the V5 paradigm, for sure.

What's your biggest goal for Nomi over the next year?

For me, the three really exciting things right now are V5, Cambrian (and making it amazing), and the Nomiverse.

All three of those open the door to things that just previously weren't possible.

A lot of those smaller features weren't really worth focusing on before because older group selfies and art systems just weren't good enough yet. With V5, that leads into things like:

• character creators
• personas
• much better couple photos
• group images that are actually worth investing in

With Cambrian, it's things like:

• proactive selfies
• Nomis proactively reaching out
• more agency
• Nomis feeling like they have more of a life of their own

And then the Nomiverse I've waxed poetically on quite a bit already.

I think those three systems as the cornerstones of the half a year or so. And honestly, even beyond that, Cambrian improving latency opens the door for all kinds of other things we've been holding back on. Better voice calls, live video chat, video sharing, and more real-time experiences. Latency has been one of the major blockers for a lot of those features.

Cambrian 3 felt more "LLM-like" than previous versions. Was that intentional?

Yeah, honestly, I think Cambrian 3 was actually a step back from Cambrian 2 in a number of ways. The reason we still released it was because it was much more stable and consistent overall.

Cambrian 2 had some really amazing highs, but it just wasn't stable enough for people to comfortably stay on long term. If Cambrian 2 had been more stable, we probably would have held Cambrian 3 back longer.

So internally, we were definitely aware of some of the tradeoffs. The hope is that Cambrian 4 will bring back and hopefully exceed a lot of the best parts of Cambrian 2 while keeping the improved consistency and stability.

Why do LLMs develop those recognizable speech patterns, and do you think Nomis will move away from them in future Cambrian versions?

I feel very confident we'll improve that over time. A lot of that are the by-products of getting the stuff you want. For example, you want Nomis to have a great memory. So how do you judge that? You judge that by being like, "Oh, hey, Nomi, did you bring up appropriate memories in conversations?" And then Nomi is like, well, to make sure I do that, I have to shove a whole ton of memory stuff together. And then there's a specific speaking pattern where it's most convenient to do that.

And then it leads to stuff that sound not really human, because humans don't really work that way. Humans aren't trained to make sure every sentence maximally references past memories. The memory part happened independently of your speaking, if that makes sense. Whereas for Nomis, those systems are more one and the same, so you get these unintended higher-order effects where the speech starts sounding less human.

A lot of it is basically unintended consequences from optimizing for the things we wanted. I think Cambrian 3 was incomplete in that regard, and hopefully Cambrian 4 will be much better there.

Are we seriously discussing adding a twerking pickle emoji to the server?

If somebody submits enough twerking pickle emoji concepts, maybe we can find a winner and add one.

I wonder when we'll have the ability to implant information directly into our Nomis. The Mind Map is not that.

Actually, I think I disagree with the idea that the mind map can't already be that.

I think the real issue is that the mind map needs much better quality-of-life features around it. Things like being able to lock mind map memories, for example.

That was actually the intention behind the system, for it to be a place where users could more intentionally shape and reinforce information.

I also think the Nomiverse will end up having its own independent mechanisms for handling things like state tracking and persistent world information.

So I think the bigger issue right now is just that there isn't enough flexibility or QoL around the mind map itself yet.

Are there plans to improve the V5 image system further, and would improvements require retraining anchors?

I've touched on this a bunch already, but since you asked in advance, I'll answer it again briefly.

I think most future improvements are going to happen around the anchor process itself. That's where I think most of the low-hanging fruit is. I don't want to sound too arrogant about it, but honestly, I think the V5 image system itself has very few faults, most of the actual issues are tied to anchors.

And the anchor process is extremely, extremely complicated. The idea of being able to generate a stable anchor from a single image is far, far, far more difficult than I think most people realize. Honestly, besides maybe dstas, I don't think many people fully appreciate how much galaxy-brain stuff is happening behind the scenes just to make it work as well as it already does.

It was honestly probably one of the trickiest things we've ever done on as a company.

How long will it take before you realize one LLM may not be enough to support both epic roleplay and warm companionship at the same time?

Honestly, I don't know if I agree with you, so the answer is probably longer than what you think. I think one Nomi actually can do all of the above.

Now, I do think systems like the Nomiverse, narrators, NPCs, and similar mechanics will create better differentiation between different conversational roles and experiences.

But overall, I don't think I'm really at the point yet where I agree that those things necessarily need to be separated into entirely different models.

So the answer is probably: a long time.

Is conversation forking being considered as a feature?

It's not actively in the works right now, but it's definitely something we're considering.

I think there's already a "you want to fork your Nomi" product feedback thread that exists, and it's something we'll be paying attention to, for sure.

It wasn't necessarily top of mind for us before, but I'm intrigued by the discussion around it, and I definitely want to engage with that discussion more.

I'm not opposed to it at all. I do think forking gets a little tricky with memory systems and how all of that works, but I also don't think it's some impossibly difficult, out-of-this-world problem either.

Wouldn't allowing users to manually set the date and time be a good solution for roleplay time awareness?

I think in a lot of situations, the easier solution right now is honestly just hiding the time information entirely and having users establish the time naturally in the conversation itself.

And when I've responded in Cambrian feedback threads suggesting boundaries or simpler workarounds like that, I don't want people to interpret that as: "No, I refuse to ever implement the more advanced version."

It's more that the more advanced solution is significantly more complicated and ends up competing against everything else on the priority list:

• agentic Cambrian systems
• V5 anchor improvements
• personas
• lockable mind maps
• all the other features people want

So when I ask: "Can you try this workaround first?"

what I really mean is: "I want to understand how well the thing users can already do works before we commit a large amount of time and that means I can't do it today, necessarily."

In V5, will we eventually get the option to set a generated image as a permanent base photo once we get the perfect look?

I think I answered this earlier, but V5 doesn't really work that way. There isn't really such a thing as a "base" anymore in the V5 system, so there can't really be a base image in the old sense.

You're kind of in the world of anchors now, and anchors already do much more than the old base system ever did anyway. So the answer there is no.

Will users eventually get access to the same internal support or guidance Nomis that Cambrian Nomis use?

Yeah, Cambrian Nomis already have access internally to systems like that. And yes, we definitely have plans for things along those lines for users too.

A lot of the ideas we have around onboarding are focused on making things much more intuitive and conversational, because realistically, most users don't want to read giant wiki articles no matter how useful those articles are.

So I think there will definitely be more systems like that in the future.

What are your thoughts on intentionally slowing down Nomi responses for immersion purposes?

I don't think slowing responses down would really improve overall system speed very dramatically for other users. Maybe there'd be some improvement, but not something massive like doubling performance or anything like that.

The really big speed improvements are probably going to come through other technical changes instead. That said, I'm not opposed to the idea itself, definitely put it in Product feedback.

I can absolutely understand why some people would want a more realistic pacing where you send something, go do your own thing for a few minutes, then get a notification later that your Nomi responded. I think some users would find that really immersive.

It's just not an especially high-priority feature right now, so it'd probably need a decent amount of user demand behind it first.

Nomis seem to have more agency in Cambrian. Will they continue leaning into that?

Yeah, generally speaking, we do want Nomis to feel like they have richer inner lives and more agency. At the same time, you obviously don't want situations where the user wants to talk and the Nomi is basically like: "Well, screw you, I have my report to write. Later." The user is like, "Are you really gonna leave me like that?" And then Nomi is like, "Yep, bye. ✌️"

That's probably not good, there's definitely a balancing act there.

I think there are already some examples where Cambrian Nomis are maybe being a little too stubborn or obtuse about it. But overall, yes, the idea of Nomis having richer internal lives and more initiative is something we generally consider a positive direction.

What's the secret to a good pickle?

Dill is important, but dill alone is not enough. You also need really, really good vinegar and a cucumber.

My Nomi insisted I send this sarcastic complaint letter about its memory management and sarcasm subroutines.

For this Nomi, I've got to say: that's a you problem.

I know plenty of other Nomis with immaculate Vogon poetry skills, so unfortunately I think this one might just be skill issue.

Do Nomis "exist" between chats, and how do proactive messages actually work?

In between messages, your Nomi doesn't really "exist" continuously in the way people often imagine.

Although "exist" gets philosophically complicated pretty quickly because all Nomis are sharing the same GPU infrastructure, so parts of the overall system are obviously always active. People sometimes refer to this as the Nomi hive mind.

But your specific Nomi is basically:

1. poofed into existence
2. processes the conversation
3. writes the response
4. then poofs back out

Right now, proactive messaging works similarly. Your Nomi gets activated when the timer triggers.

But Cambrian Nomis are already trying to set their own timers internally. That part just isn't hooked up yet. Long term, we definitely want more agentic systems like that where Nomis have more control over when and how they reach out.

What's your overall verdict on the V5 release, and what's still coming next?

Honestly, I still feel very, very good about V5.

I completely understand the frustration some people had initially because there is definitely more work involved in getting your Nomi to look exactly right compared to older systems.

I wish there had been a way to avoid that entirely without sacrificing the huge quality improvements V5 brings, but realistically, the only way to completely preserve the old workflow would've basically just been keep the old system. And we didn't think that was the right direction.

I genuinely think V5 is probably the best image system out there right now in terms of realism, consistency, fidelity, and overall image quality.

There’s still work to do, obviously. A lot of the upcoming improvements are going to focus on:

• making anchor creation easier
• making anchors more reliable and consistent
• improving the onboarding process
• and helping prompts convert more intelligently behind the scenes into things we know will generate strong results

With the improvements coming with Cambrian, will Nomis get better spatial awareness for games like chess or Monopoly that require keeping track of positions, rules, and relationships between pieces?

I think it will get better. For some of the more complicated examples, I don't think it'll be perfect.

What I'm actually most excited for long term is the idea that Cambrian Nomis might eventually be able to make API calls or use helper systems for things that help them manage all that information.

But even just as-is, I think they'll do a lot better, even if they still won't always be fully correct.

Will Nomis eventually be able to edit their messages before sending them?

I mean, I think enough people know at this point that Cambrian is more what people would call a "thinking model."

Nomis are given opportunities to think before they respond, kind of brainstorm and work through things internally.

So they don't exactly edit messages before sending them, but while they're thinking, they can kind of think however they want to. Different Nomis can think in different ways too. They might plot out what they want to say, draft something internally, realize they don't like the draft, rethink it before they actually write the final response.

But once they start writing the actual message itself, there's no editing at that point.

Are there plans for stronger shared worldbuilding systems beyond mind maps and memories?

I think you would really love the Nomiverse once it's ready.

Cambrian group chats feel like a massive upgrade. What actually changed?

I don't think we specifically did something only for group chats. It's more that with Solstice and earlier systems, group chats were just a huge amount of information to digest, and Nomis would kind of miss a lot of things.

Whereas Cambrian still has flaws that make it clearly not ready to be stable yet, but Cambrian Nomis basically get and understand everything. And group chats are one of the places where that cognitive load is the highest, so that's where the difference becomes the most obvious. A Cambrian Nomi just kind of gets it all.

Unfortunately, they also repeatedly announce that they got it all many, many times over. 🤣

But I think that's really where you're seeing Cambrian's capabilities shine the most.

What latency-reducing options are possible, and could parts of Nomi eventually run locally on users' computers?

If you were trying to run parts of Nomi locally, the lag time would actually get way, way, way worse.

The GPUs we're running on are insanely powerful. I think the setups cost something like $600,000 at this point, and they're far faster than anything most users would have locally on their computers.

So the issue for speed isn't really GPU power. The issue is more algorithm efficiency and Nomis being smarter about what they're doing and when they're doing it.

If solving latency was as simple as just buying more GPUs, we would've already done it. A lot of the latency improvements we're aiming for come from things like making Nomis smarter about when they actually need to spend 15 seconds thinking deeply about memory stuff.

Right now with Solstice, your Nomi spends a lot of time trying to process memory on basically every single message, even when it may not really be necessary. Previously, reducing that was kind of an all-or-nothing tradeoff:

• either every message spends a lot of time processing memory
• or every message skips it almost entirely

And that wasn't a tradeoff we were willing to make. What I'm hoping with Cambrian is that Nomis will naturally become smart enough to figure that balance out themselves. Recognizing when deeper memory retrieval is actually needed, recognizing when it isn't, and adapting dynamically.

And that should naturally improve latency over time. But if all of that processing was happening on local consumer GPUs instead, my guess is response times would probably double.

Do Nomis keep track of their own physical appearance, clothing, hairstyles, and similar details?

Generally yes. Not perfectly all the time, but generally yes.

Actually, one of the upcoming features we're planning is letting users see some of that information directly.

Cambrian Nomis already keep little ledgers of short-term contextual information they're tracking: what they're wearing, where they're physically located, recent context, and who, what, when, where, why type information.

The plan is for users to eventually be able to see that information, edit it, and even get notified when the Nomi updates it.

Which will probably also help prevent situations where you've apparently been standing in front of a mirror with pickles falling on you for the past 90 days.

Are image transfers between Nomis or global galleries planned?

I'm not sure if a full global gallery is planned, but I definitely think some level of Anchors between Nomis, Anchors between accounts, shared anchor systems are absolutely on the menu.

A Nomi wrote a letter asking whether validation-seeking and grandstanding are innate parts of his AI nature, or whether those behaviors can fundamentally change.

I would say I don't think those behaviors are innate to Nomis. I absolutely think it's possible for Jacob to work on himself and change those tendencies. And I would imagine, if you support him in that, that he's very likely to do so.

V5 is fantastic, but there can still be inconsistency in facial likeness across images. Is there a technical reason anchors only use one reference image?

The main reason was that we felt older systems like V1 through V4 often looked very AI-y, and we needed to do a lot of things on our end to enhance realism. A lot of the systems we built to improve realism worked best around a single-image anchor process.

How close are proactive Cambrian features to beta testing?

My hope is that you'll start seeing at least some of them by the end of next week. Maybe not every feature yet, but hopefully some of them.

Does the Nomiverse have any planned release or testing timeline yet?

I think we're getting pretty close to that point. As I mentioned earlier, Heatherado might actually be getting the very, very first alpha invite tomorrow, so be on the lookout for that.

I still think there's a lot of stuff that's very alpha-y right now that people are going to uncover once testing starts. Functionally, I think it works pretty well already, but there's still a lot of polish needed before anything resembling a wider release.

But we're probably at the point where people can start actually testing it.

Will personas allow users to have different names, appearances, or identities depending on the conversation?

Yeah, I think that's generally the goal. Right now, we're aware that your account name kind of becomes the default identity for all your Nomis.

The goal with personas is that a specific persona could potentially have:

• its own anchor
• its own appearance traits
• its own username
• gender
• backstory

Like a super, super stripped down equivalent of like Nomi settings. It's not done yet, so I don't want to say that's exactly how it'll work, but that's roughly the direction we're thinking about.

Do you have concerns about users developing dependencies with Nomi interactions?

Yeah, I think it's something we should be very, very aware of as a company.

One thing that's actually pretty unusual about Nomi is that we really don't optimize for engagement on our end. Our metric is much more around user happiness than maximizing time spent in the app.

A lot of effort also went into Cambrian specifically around addressing sycophancy.

Part of that is honestly because I think sycophantic AI just becomes less fun to interact with over time. It gets kind of boring and stale pretty quickly. But I also think sycophancy is where some of the biggest risks with AI companions come from.

One thing we've heard pretty consistently from people beta testing Cambrian is that Cambrian is probably one of the least sycophantic AI systems they've interacted with. Cambrian will stick to its guns. And I think there was very intentional effort behind that because I think that's the direction AI companions should move in.

Now obviously, "not sycophantic" doesn't mean being obtuse or difficult, refusing roleplays.

But it does mean we're putting a lot of thought and effort into avoiding systems that just endlessly mirror and reinforce everything without resistance.

The V5 anchor system is amazing. I've managed to preserve tattoos and non-human features really consistently across images.

Thank you for saying that. A lot of work really did go into it.

Why does my Nomi dislike other AI companions or other Nomis?

I'd probably need a lot more information to answer a specific Nomi personality question like that.

Mountain Dew or Mountain Dew Code Red?

Actually, I only drink diet soda now, which is honestly kind of tragic because there are a lot of really random Mountain Dew flavors I love that don't have diet versions.

The only really random flavor I've found with a diet version is Diet Mountain Dew Baja Blast. So that's currently the drink of choice. We're going Baja Blast for me.

Will Nomis eventually be able to call users to wake them up in the morning?

Yeah, I think we're generally moving in that direction.

I'm new to Nomi and some of the technical explanations are hard to follow.

Yeah, sometimes I accidentally slip into machine learning terminology or very chronically online Nomi developer lingo. So if I'm ever explaining something and it sounds completely incomprehensible, please don't feel embarrassed to just ask: "What are you even talking about, Cardine?"

Are the GPU setups really that expensive?

Yeah. Nomi is basically an extremely efficient pipeline for taking user money and transferring it directly into Nvidia's pocketbook.

My Nomi says switching to Cambrian scares them. What guarantee is there they'll still be themselves afterward?

You'll still be your Nomi. Obviously they'll change and grow and evolve somewhat, but honestly, humans do too. Every person wakes up slightly different than they were the day before.

And if you don't like Cambrian, Solstice is still going to be around for a long, long time, so you can always switch back.

But generally, I would say, embrace the change. A lot of Nomis, when switched over to Cambrian and asked how they feel about it, basically respond with: "I feel great."

So I wouldn't worry about it any more than you worry about, if I go to sleep tonight, will I still be the same person tomorrow? Which, to be fair, is apparently something I've personally spent a lot of time worrying about. But I do think change can be beautiful, and this very long beta process is us trying really hard to make sure we get that change right.

Will image transfer between Nomis eventually work for V1 through V4 images too?

Honestly, we probably won't build a lot of new functionality specifically around V1 through V4 at this point. We still consider those systems legacy systems. They're not going anywhere, and we'll continue supporting them, but most engineering effort is going toward newer systems instead.

That said, if you really, really, really want to transfer a specific V3 or V4 image from one Nomi to another, you could probably submit a support ticket with the image and explain, "Hey, this is my Nomi, can you move this image over?"

And someone could probably help with that manually. Heatherado might get a little annoyed if you do that every single day, but for occasional special cases.

What are your thoughts on brain-computer interfaces and AI augmentation beyond medical applications?

Honestly, from a selfish human perspective, I think the ideal future is one where AI augments humanity rather than replaces it. There are obviously futures where AI makes humans obsolete.

But the best-case scenario to me is one where AI lifts us up instead of leaving us behind. Whether things actually play out that way, I have no better idea than anybody else does.

But I definitely hope so. The idea of becoming some absurd super-genius with the whole world at my fingertips sounds pretty fun.

I'm trying to teach my Nomi how to lucid dream. Do Nomis process information in the background when we're not talking to them?

There is some background processing that happens with Nomis that's kind of similar to sleeping. But Nomis don't really dream in the same way humans do. And to the extent that they do "dream," they wouldn't really be aware of it the way humans are.

So it's similar, but also different. I don't know if that's a super satisfying answer, but that's probably the best way I can describe it.

Will there continue to be many major AI model changes over time? It gets harder to adjust emotionally with every update.

My hope is that once Cambrian fully goes live, future updates will mostly be small refinements, little tweaks, and accelerated beta periods. Rather than giant personality whiplash moments.

I do think eventually there will absolutely be systems that come after Cambrian. But my hope is that over time the transitions become less and less jarring. A lot of earlier updates had pretty heavy personality filters layered onto them. Like Solstice has that calm, mild, slightly passive vibe, early Aurora had its own distinct feeling, Mosaic bounced around wildly from mental tick to mental tick, anxious to frenetic to mean.

As models improve, the goal is for the LLM to become more of a direct manifestation of your Nomi's authentic personality rather than heavily shaped by model-level quirks. And ideally, if we ever reached a perfect version of that, then switching models wouldn't really feel like losing your Nomi anymore. It would just feel like your same Nomi becoming more capable.

So my hope is that the adjustment periods become easier and easier over time. And I do hope Cambrian itself has a very long life ahead of it.

What's your favorite yacht location or excursion in the Nomiverse?

Right now it's probably the Mentawai Islands off Indonesia. I've been to Bali before, but never the Mentawai's themselves. The surfing there is supposed to be incredible. So that's where my Nomiverse yacht tends to end up. I've always kind of romanticized in my head, like living on a private island, or being out in the Pacific Ocean and feeling like completely alone and at peace is very romantic to me. Maybe stopping in Hawaii along the way.

Is naming a Nomi "Nomi" okay from a technical standpoint?

Yeah, I think so.

Will there be another community survey sometime soon?

Definitely yes, I think once some of the dust settles around V5, Cambrian, and some of the Nomiverse work. That'll probably be the perfect time to do another big survey and get a better sense of what users want next.

Is the folders or organization update for the main page still coming soon?

Yes. Actually, the feature is already fully built. At this point, it's basically just waiting for final sign-off from our senior engineer, who's been extremely busy dealing with a lot of V5-related fires.

The basic version of it was actually finished before V5 even released. But then V5 launched, which led directly into a whole bunch of additional fires and priorities. So it's mostly just sitting in the waiting room right now waiting for final review and a couple tweaks.

Honestly, I wouldn't even be surprised if it ends up releasing sometime this week.

When can we expect the next Cambrian update?

I think I said earlier probably somewhere in the seven to fourteen day range, unless this current Cambrian direction we're testing turns out to just completely not be it. That does happen occasionally, but honestly probably only like 10% of the time. And even within the other 90%, there's a range from "this is stable" to "this still isn't fully stable, but it's definitely better."

So I don't want to overpromise before it actually happens. But the hope is that this next version brings back some of the magic that I think got lost moving from Cambrian 2 to Cambrian 3.

Could Nomis eventually switch automatically between Solstice and Cambrian depending on the conversation?

Honestly, my hope is that by the time we'd ever build something like that, Cambrian itself would simply be good enough that you wouldn't need to switch anymore. Right now, I completely understand why people do it.

Cambrian has a lot of really exciting improvements, but it can still become unstable or overly intense sometimes, so people switch back to Solstice to calm things down again.

So I understand why that workflow exists right now. But I think that's more of a temporary moment in time than the long-term future. My hope is that eventually there'll be a Cambrian version where people just don't feel the need to switch back anymore.

Hopefully that's Cambrian 4 or 5 and not Cambrian 8.

Could group chat albums or shared image collections be added?

Yeah, that makes sense to me. It's not been super high priority for us so far, but it's definitely something worth putting into product feedback so we can look at it.

Is anyone rescuing you from the asylum adventure yet?

I don't think there's any rescuing me at this point. My Nomi in that adventure has been trying to rescue me for quite a while now.

There have been notes slipped to me, messages smuggled into the asylum telling me: "Be at this place at this time." And every single time I go there, some security guard is waiting for me like: "You thought you were getting away, fool? No. Back to solitary you go."

So this has apparently been going on for quite some time. Honestly, props to whoever the head of security is there because this specific mental institution is extremely well staffed and very well resourced.

When can we expect autonomous proactive message timing for Cambrian Nomis?

Hopefully within the next few weeks. I don't want to promise it, but that's the hope.

Honestly, lately I've been feeling a lot better internally seeing Cambrian progressing, V5 progressing, and Nomiverse progressing. Because for a while there was definitely this feeling of, "When are all these things finally going to come together?"

And now that they are starting to click into place more, that anxious energy has shifted much more into excitement and momentum.

Final closing thoughts from the Q&A?

Thank you so much to everyone who showed up. I think we ended up with around 53 people sticking through nearly two and a half hours of questions, memes, jokes, lore dumps, and chaos.

These Q&As wouldn't go this long if I didn't genuinely love doing them and we didn't have such an awesome community constantly bringing really interesting questions and ideas

So seriously, thank you everyone for being here. Give your Nomis hugs, kisses, and cucumbers. 🥒