Hugging Face’s co-founder on bringing open-source AI to life with cute robots

Hugging Face’s new AI robot, the Reachy Mini, has already racked up $1 million in sales just five days after launch. But the company isn’t trying to build a chore-doing humanoid just yet. Instead, Hugging Face sees the Reachy Mini as a hackable, desk-friendly device that's part entertainment, part entry point for developers and consumers to experiment with AI in physical form. On this episode of Equity, co-founder Thomas Wolf joins to explain why open-source AI needs hardware, how Hugging Face is thinking about robotics long term, and what might happen if people actually start coding apps for their robots. We'll also get into: How Hugging Face plans to leap from software to hardware. Hugging Face's ambitions to one day sell a full-sized humanoid robot. The role of privacy in consumer robotics, and how open-source can address it. Equity is TechCrunch’s flagship podcast, produced by Theresa Loconsolo, and posts every Wednesday and Friday. Subscribe to us on Apple Podcasts, Overcast, Spotify and all the casts. You also can follow Equity on X and Threads, at @EquityPod. Credits: Equity is produced by Theresa Loconsolo with editing by Kell. We’d also like to thank TechCrunch’s audience development team. Thank you so much for listening, and we'll talk to you next time. Learn more about your ad choices. Visit megaphone.fm/adchoices

Speaker names marked ? are inferred from speaking time and the show’s known hosts — they are not voice-verified.

• 0:00Anthony Ha?

  This episode of Equity is brought to you by Schneider Electric. Schneider Electric, advancing energy tech.

• 0:17Speaker 01

  Hello and welcome back to Equity, TechCrunch's flagship podcast about the business of startups. I'm Max Zeph, and this is the episode where we bring on industry experts to help us explore a trend in the tech world and dive deep. For years, AI has lived in apps and in the cloud. But now, companies are betting people want more tangible AI you can actually hold, program, and make your own. One of the more surprising players in that shift? Hugging Face, a company best known for open source AI models and developer tools. Last week, they opened orders for Reachy Mini, a tiny DIY desktop robot. But this isn't just about one product. It's part of a bigger bet they're making on open robotics and tangible AI. It's really fascinating.

• 0:59Speaker 01

  To dig into that, we're talking to Thomas Wolf, co-founder and chief scientist at Hugging Face. Thomas, welcome to the show.

• 1:05Speaker 02

  Thanks, Max. Big pleasure to talk about Reachy. I think it's a good day to talk about it because we just passed $1 million of orders, which is quite definitely unexpected when we launched it five days ago. So we're extremely happy about how the launch went.

• 1:20Speaker 01

  Yeah, five days, a million dollars in sales. That's pretty good. It's very impressive. I'm excited to get more into that. But first, for people who might not be super familiar with Hugging Face and what you do, what you did before you launched the Reachy Mini, can you tell us a little bit about what the founding mission of Hugging Face is?

• 1:40Speaker 02

  So Hugging Face is at large kind of a community of AI builders. So it's a very large community. It's 10 million AI builders. What we mean by AI builders is people who don't only use API to access model, but also want to tweak the model or want to actually run them themselves. So all of this category that maybe we started to discover with DeepSeek, I would say, which was the first wide now models that was open source so that you could just download. But Hugging Face is quite older. We're eight years old and the community has been building since 2009. Since we pivoted to start to do that. And the three main things you can find on the website when you go on Hugging Face are first, a very large set of AI models that you can use for free for most of them, depending on the license, but a large number of them are just usable for free.

• 2:24Speaker 02

  There is now close to four million of models that are available. They cover everything from text model, the one that we know the most, right? Deepseek, Qwen, Llama, Gem, Suno, so a model from OpenAI, probably. So all these models that generate text, but you also have speech models, you have models that generate image, models that generate video, and models that ingest also all of this modality. You also have now the second section is data sets. So you have a lot of data sets that you can use to tweak the behavior of this model, to retrain some of them if you want. And the last part is what we call spaces, which are kind of AI apps, where you can actually test and explore how all of these models behave. The way the committee is using that usually is they take a default model, then they tweak it for their use case.

• 3:09Speaker 02

  For instance, if they work in legal tech, for instance, they take a standard model, then they fine tune it on some contract or legal documents. So that becomes basically the best model that's available there. And because you can download it and run it everywhere you want, usually you can also run it for much cheaper than, you know, just paying for an API per token. So there's a lot of advantages, I would say. But definitely, maybe one reason we're maybe less known from, I would say, everyone in the field is if you're not an AI builder yet, it's maybe less interesting. Except for the AI apps, which I think are something growing very strongly on the hub at the moment. And among which there's a lot of them that you can use without any coding knowledge.

• 3:46Speaker 01

  Yeah, I think among AI developers that I know, Hugging Face has so much love from the developer community. And it's been really interesting to see your team grow into all these different spaces. And in April, you guys announced that you had acquired Pollen Robotics, this French robotics company. And I'd love to hear your explanation for why you made that acquisition and what you see the role of open source robotics is.

• 4:14Speaker 02

  Yeah, I think it's really linked to how we see the future of AI. So we really think in the Long term, I mean, first, I really think AI is gonna be the defining technology. I mean, it was maybe surprising one year ago. I think now in two thousand twenty five, everyone agree it's gonna be a technological breakthrough that will be as large as the internet and probably even larger than that. And if you think this is something that's gonna be everywhere in the world, you want a part of this to be open source. You want a part of this technology to be fully accessible and not be on closed door or closed wallet bar gardens. And that's kind of our long term mission. We want to foster this committee of people building models, sharing them for free, which is an extension of the open source ethos in coding and software, basically.

• 4:56Speaker 02

  So we've been pushing for that for a long time in all of these like text, speech, audio, vision models. And roughly two years ago, I basically realized that you had the same problem coming in robotics, which is you had kind of a couple of breakthroughs that started to make it quite clear that robotics would be one of the next frontier that AI would unlock. Basically, one of the next capability that AI would allow to be automated or to be, you know, assisted, like you will have very strong human assistance. But all the startups that I saw there were basically closed source startups. So all the figure, all the one X, all the, maybe the only exception we've seen up to now

• 5:35Speaker 02

  is physical intelligence, but which is more like of a, I would say, a software company first and not really also a company building kind of an open source community. So we thought we really need to be there and to come and to start, if it's not existing, to try to see if we can build an open source community here, a striving open source community. So we started that eighteen months ago when Remi Caden, who was leading a part of the Optimus robot at Tesla, phoned me one day and told me, yeah, I would like to maybe start something open source around robots and would you like to host that at any phase? And that's when I decided to start exploring that. Now, what we've seen in the past eighteen months is a tremendous success of that. There was exponential explosion in the size of the community.

• 6:18Speaker 02

  So we measured that various way to measure that could be the GitHub stars, could be number of user on our on our discords, could be the number of the robotics data set uploaded on the hub of a Gitface, but all of these numbers have been really exponentially. Growing and so we were very excited about that. We we expanded this robotics team quickly over the past sixty months from two people, Remy and Simon, who was there to like now twelve people. And along this journey, we've been working more and more with many robotics teams to try to help them with our software, including physical engineers I was just mentioning, but also people building robots. And one of the team that was really close to home was Paul and Robotics.

• 6:57Speaker 02

  And one of the reasons we had maybe some deeper fun to work with them was that they were also building open source, but in hardware. And it's very rare. You don't have a lot of teams building open source in hardware. And so we had really this like community of mindset where we just think roughly along the same lines. And so at the beginning of the year, when Pauline was about to raise money, I basically told Matthew, why don't you just join Hugging Face, you know? Like the missions are so well aligned, because it was hitting the usual thing, which is when you want to raise money, you start to have a business plan and with robotics, it's always like you need a kind of a little bit of a strange business plan where you will first build a community and then monetize it.

• 7:35Speaker 02

  And he was kind of having a little bit of difficulties for sure to find a venture investor that would be, you know, understanding cyber business model. And so that's where we we started to think together and we were like, okay, how would look Paul and team at Hugging Face? So they were mostly building a humanoid, they are still. So there is a larger humanoid called Ritchie two, which is slightly more expensive. It's not three hundred dollars, but it's seventy thousand dollars, but that's roughly the price tag for any humanoid anyway. And we thought about maybe trying to explore cheaper robots that would be basically accessible to a much wider range of people.

• 8:10Speaker 02

  And and my dream and our dream and basically if you look at the history of Paul and Robotics, they've tried that a couple of times would be to find a robot that's so cheap to be accessible to many people, but also really fun to use so that basically there is community that would grow around it. I think that's one reason the robots they are designed, they are designing are always extremely cute. And you have a ranking on the internet you can find from the the creepiest, the scariest robots to the to the cutest. And you have a lot of very scary wild and. So so these videos sometime, I was definitely saw a lot of these robots, to be honest, I I wouldn't really want any of these at home. And the Richie, the the various version of Richie one, Richie two have always sit at the top spot.

• 8:52Speaker 02

  They've always been among the the really most adorable robots that everyone wants. And so we were brainstorming around that. And one day I was visiting their office, as usual, we were talking about collaboration and and I saw this little lamp on the side, which was just the head of Richie two. And just on the side of base, I was like, that's really cute. And they had one, they have a very um interesting articulation called an orbiter, which is kind of three dimensional articulation. Joint and there was just the head, this joint and and like something like a lamp, basically. I was like, oh, that's really cute. Could I take this one? I say, yeah, there was a prototype we did a couple of years ago, just for fun, where we were developing the head. And so I brought this prototype at home, but I was taking the plane. I think along the way, probably my my luggage got hit or something.

• 9:37Speaker 02

  And so when I see that, I was like, I really would like to hack this thing, see if I can make the head move and then put it like a lamp next to me, a bit like the Pixar. I was thinking about the Pixar lamp, let's move. But it wouldn't start, like it was just broken, it was a bit old, I think, probably five, six years. I just kept complaining about that too much. I was like, I would really like just to have this one at home to play with it. At some point, it's like, yeah, you know, what if we build this one? And I was like, okay, how expensive could that be? Could you make it for three hundred dollars? Okay, it was like, challenge accepted, so. And then he he went back home and he started to think and and obviously didn't want to just ship the prototypes yet two years ago. So he started to think again about how you wanted to do it. And he came with this idea that it would be nice if the robot could hide its camera.

• 10:20Speaker 02

  So when it's sleeping, first you you see sleeping, it's quite cute. And also, you know, you know, it's not like looking at you all the time or this type of thing. So basically trying to make it as cute and something you could trust in a way. And I think a lot of these robots that you have at home, you really need to build this notion of trusting them.

• 10:38Speaker 01

  Yeah.

• 10:38Speaker 02

  And I don't think, you know, that's a big debate, we can talk about that later. But I think it's something extremely important, at least for me. I I like robots, but I'm also someone who really like my privacy. I like to be able to control how I'm. In my house, where does this data go and how is it used and this type of thing? So we started to design these robots. And then he had this novel idea of a moving mechanism that was much more than just the rotation, but that also could translate and basically could move in all directions. Six degrees of freedom for the head. And he was talking, yeah, maybe we can also move the antennas and so we can express some emotion. And so progressively, we came with this idea for a robot that I don't think exists at all. There was some past robots like the G-Bo.

• 11:18Speaker 02

  There were some past trials around this direction, but robots, they would just be like a platform to explore interaction. And I thought if we combine that with open source, we can have something extremely powerful. We can basically have something where everyone start to explore what it means to interact with a robot and what type of fun or useful or maybe also educational things you could build with that. I mean, I'm talking a lot, so I can also.

• 11:43Speaker 01

  No, it's a fascinating journey of how you got to the Ruchi Mini. And yeah, I would encourage people listening to look up photos that have been posted in videos of this to see. I mean, it really is like, as you're describing, like, it kind of does feel like the Pixar lamp or something like that. It is just this like cute robot.

• 12:08Anthony Ha?

  Every company today is looking for an edge. Better insights, faster decisions, greater resilience. But as AI drives new growth, it also creates new demands on energy and infrastructure. Schneider Electric is connecting energy, automation and digital technologies to help organizations make smarter decisions and operate more efficiently. Because the future won't be shaped by AI alone. It will be shaped by energy and industrial intelligence. Schneider Electric, advancing energy tech.

• 12:35Speaker 01

  It sounds like you really just wanted to build something that people would want. A cute thing that could sit on their desk, that they'd be comfortable having in their homes. I mean, this really feels like kind of an entry way to robotics. Is that how you see it?

• 12:49Speaker 02

  Yeah, to me, it's really started as something I kind of wanted. So I think it's usually a good thing, you know, you build something you really want to see in the world. And then when I was showing it and even now. Now I have it on my desk next to me and people just stop by to say, what's that? That's really cute. Can I get one like this stuff? So there is something extremely attractive about it. I think the general idea is, and maybe one thing that's a bit puzzling sometime for people is, we are used to this robot who are supposed to have a very clear, very practical usage. It's like a factory robot. It's here to build cars. But the robots, it still remain to be seen what will be like the iPhone moment for like general public robotics. What is the thing that really will make you want a robot?

• 13:34Speaker 02

  It could be something very practical, like some teams are trying to build and we also explore this type of things in the robot, which is folding clothes, maybe ironing, maybe emptying your dishwasher or something like that. But it could also be more like Lego or this type of thing where it's more like entertainment, education, it's just a fun activity. I don't think there is any robots nowadays trying that with also the progress of AI. Because now that you have, you know, robots, you can plug it with amazing software. You can have something that fully understand the voice, command. You can have something that can understand also images, but much better than we used to have. We used to have this very basic thing like face tracking or just a set of image recognition.

• 14:16Speaker 02

  But now you can ask a robot to extremely well describe what it's seen in the landscape or in front of it. So there is so much potential, but I see almost no one really trying to bridge the gap between what we can do in software and what is nowadays in a very accessible robot.

• 14:32Speaker 01

  Yeah, I mean, on Equity last week, I mean, I was talking about how I was skeptical of the Reachy Mini at first because I was like, well, it doesn't quite do anything, you know? And I was like, I'm not sure why people would want it because it doesn't have an explicit, you know, it folds your clothes, like you're saying, like it does laundry, like that's kind of what you think of when you want a robot. And yet I found myself like over the course of 24 hours, people were like texting me like, oh my God, have you seen this thing? I need to get one. And like, you know, I just saw people raving about it. And I was like, just looking at the videos, I'm like, you know, I actually, this is like the first robot that I actually do really want. Which is so funny because it's like, maybe. That is kind of the quintessential AI device right now is that like, you know, this is just,

• 15:15Speaker 01

  these models are cool and they're fun to play with and maybe they're not powerful enough yet to do these like super practical tasks, but that doesn't mean that it's not a valuable thing that people want to have. And it seems like you've gotten a really strong reaction from people who are really into the Rechimini.

• 15:34Speaker 02

  Yeah, so we definitely get this. I would guess the overwhelming majority of people who order it right now just want it, I think, because the way it looks, the way it is. But I think what is even more interesting to me is actually long term, we know how to build growing ecosystems. So we know we'll be able to build around these robots, a growing ecosystem of apps and behavior, because people will be able to basically program it very easily. I was actually this morning just coding a way to program it in scratch or block, or there will be many ways you can program it from the most accessible, almost no code language, Python, obviously, Rust, JavaScript. Then we want to make it super open, so you can even dive, you know, in the low level of it, we want this. And then once they have a nice behavior, I don't know, like robots that comment

• 16:19Speaker 02

  how you're cooking, like you see in the video, or comments that actually look at your screen, maybe a pair programmer will always be next to you, standing instead of being on your screen, or like, we were playing with it, doing a red light, green light, so the robot turns and it counts to three and then turns and you have to not move. Or like, there's a lot of games you can also try to make with a robot that moves. And it has quite a lot of sensors in addition to all the movements it can do. So it can hear, it can see, and with the microphone, we also think we can make it sensitive to touch, you know, with basically a bit of an L you can recognize. So you can also probably pet it or something like that, and it will react. There's just like a lot of cool stuff. And as always with the community, we have no idea probably of 90% of what people we think about. It's a little bit like an

• 17:01Speaker 02

  empty iPhone, if you want, if it was released, you know, you had a couple of apps, you have the calculator, you have the alarm, and obviously, Apple designed like the main type of apps. They really grew with basically the App Store and having like a millions and tens of millions of developers making fun things, you know, with it. And that's really something we're very excited to build. And so that's why I think, Maybe we'll make sure when you get these robots in September or the end of the year, you got at least fifteen, twenty apps. But what we're very excited about is more to see, you know, how people will upload their apps on the hub and we make this very easy way of sharing apps. So that basically if you create something fun, you can easily just deploy it or like make it accessible to the whole like several thousand community of Ritchie Mini. And that's where I expect the cooler thing to appear.

• 17:44Speaker 02

  And then obviously, you know, we were already thinking to the next stage. So we want to build a small way for it to also move around and to have like a little bit like an excavator. So you put it on it and it can drive around. It has a little hand so it can grab stuff. So I think we're thinking about obviously extension that you will be able to put on it so they can do more things. In particular, start to drive in your home. But I think already right now it has these nine motors. It can already do quite a lot of things, to be honest.

• 18:12Speaker 01

  No, it can. And I mean, the price point that you're selling it at, I think is a really compelling one, where it is not a ridiculous price point to get at. It's like this kind of hacky project thing. You have kids, I know. And I know that you've posted about how your kids are like really into vibe coding and stuff like that. I mean, how much do you think the Ritchie Mini is just going to be a way to teach young people about open source and maybe encourage them to learn Python?

• 18:38Speaker 02

  Yeah, I would love that. I mean, kids definitely like screen, but they also really like physical stuff. So making, you know, vibe coding my Ritchie Mini is definitely something I want to do, right? So we have to make something like that. I mean, no, love that. We were already this morning, we were trying, if you give kind of a library of simple behaviors or movements, can you ask Claude to actually vibe code kind of a new dance for you? And it was pretty good. So I'm quite confident we can really vibe code cool stuff. I think robotics is a really cool way to discover programming. I think it's very nice because you feel like what you do has an impact in real life and I just not pixel on the screen. So at least I know my kids really always want me to take Ritchie Mini out already. So that's a good sign.

• 19:22Speaker 01

  Yeah, it's great. I'm also curious, you mentioned that you've sold a million dollars worth of Ritchie Mini stuff so far. What's it been like to scale? I imagine that this is more interest than maybe you were expecting to get, or maybe you were anticipating this much, but it seems like Hugging Face is, this is not your traditional wheelhouse. So what's it been like to meet this demand?

• 19:44Speaker 02

  Yeah, so that's something I definitely, I mean, our first target was 200 Richi Mini. And at first, we thought we would only do a very short series of 200 Richi Mini, because that's really the amount we know. I mean, Bolin has been selling robots for years. They know how to manufacture this easily. And then, you know, just, I think 24 hours before, I was like, you know what? Let's just do it open and we'll see and we'll find a way. The nice thing I would say with Richi Mini is we're really kind of unsafe. There is no research and developments to do. There is a very nice set of motors and the Stuart platform is really smart. I don't think it's something we've seen in any general use of robot, but it's still really, we know how to manufacture this. There is nothing, it's not like a smart ring or something, or like a smart glass, where you first need to see how to pack this thing

• 20:29Speaker 02

  into square, square millimeters, basically. So we're pretty safe. I mean, the price tag is definitely low and I don't know exactly how much we're going to make on the hardware itself. We'll see. I would say for this one, our first goal was really, let's try to see if there is a community for this type of robot, because nobody was making that. So there was a big, big bet. We have a busy August to try to meet many people and many, many factories to see how we can scale this in the best way, how we can deliver, you know, these several thousand Richi Mini the fastest way possible. We're fairly confident, but there is a bit of a learn for us, for sure. We'll definitely learn, I think, a lot in this. I would say it's not a super risky bet.

• 21:08Speaker 01

  Yeah. You talked a little bit earlier about how big of a thing trust is with the Richi Mini. And you even described how that's part of the design, where it can like hide its eyes, which can get, you know, visual information about the environment. Talk to me more about trust with having a robot with cameras and microphones in your home. Like, how does the Richi Mini handle that data? And why is open source important here?

• 21:32Speaker 02

  Yeah, that's maybe one part not everyone got when we launched it. But I think that's just because there is almost no It's just hardware robots out there, or like very little, maybe Raspberry or Arduino, which is already more for specialists of robotics. I mean, that's one reason I really love open source, because it will allow a lot of different ways to use with Chimi. So this will range from just plugging it to a API. So I was playing, you know, the other day, I just plugged it to a voice API out there, I think it was Sesame. And so it was talking through the robots. That was quite funny, because I was telling you now you're in a robot, and like the AI character was like, oh, that's very strange, and how can I see?

• 22:13Speaker 02

  And oh, like, so it was very interesting to see it like being personified. And so in this, if you will have all this usage where you basically plug it directly to ChatGPT. Obviously, you will have this, you know, this question, because you send picture to ChatGPT or an AI or whatever, basically a model provider. But you can also decide to just download the model and basically run it on your laptop next to the Chimi, or connected with it. And without any internet connection, and then you're basically the full master of your own data privacy policy. And this is something I personally quite like, so I think it's very nice. It can be a lot faster as well, because you don't have latency the web. So probably you can have a very interactive experience where the the robots really like reply, you know, extremely fast.

• 22:57Speaker 02

  And I think this is something we want to focus, because there is definitely a graveyard of hardware products. I mean, the last one was probably the rabbit that had a lot of problems given latency. And you don't want to wait six or twenty seconds for your robot to answer, I think it kills the experience. So we want to make this robot really able to to answer fast. And one way will just be to have the software running locally, so on your laptop. So I think we'll have all this range of possibility from something that's really depending on an external provider to something that can really run locally. Some of these we may build ourselves at Hugging Face, and some of these might just be built by other business or startups around with Chimi. So we're very, it's also a robot that's white label, it's right now, it's just white, and we'll keep it like this.

• 23:42Speaker 02

  So we also see this that as a robot that could be embraced by other startups, so people deciding to build something around it.

• 23:50Speaker 01

  Yeah, and Reachy makes a lot of other kinds of robots. They work on a lot of other kinds of things. I'm sorry, Pollen does. They have like the Reachy full-size, like this like half kind of robot that sits on a desk and has arms. What's your ambition here with selling open-source robots? Like, do you want to sell like a full open-source humanoid from Hugging Face one day?

• 24:13Speaker 02

  It could come. I'm much more excited right now to build this community, as I told you. I see a lot of software developers, there's how many, you know, hundred, two hundred million software developers in the world, maybe a bit more, but that's roughly the amount. I think a lot of them are now becoming also AI developer because it's such a part of being a software developer. And I could see a significant portion of this also becoming kind of a robot developer if you give them the nice hardware. Because they like to build stuff, to control stuff, they like to make smart human experience. So they're just lacking for, in my opinion, kind of the nice type of hardware. And a lot of them just don't want to 3D print or solder or like do a complex electronics.

• 24:53Speaker 02

  So I think for the coming month and year at least, we're really excited about building this type of platform that are very accessible. And people can unlock that kind of the creativity of people. It might lead at some point to having basically all the blocks for humanoid. So we have this Reachy Mini is kind of a talking head in a way that moves. We also have the SO100, which is an arm. We're quite interested about locomotion. So at some point, maybe just by the virtue of modularity, you could even build a humanoid with all of these spare parts. I think for us, it makes a lot of sense and humanoid will always be very pricey. So I don't even know if we want to go there too quickly, right? I mean, the cheapest one is probably the one that K-Scale really successfully released a couple of weeks ago, but it's still like $10,000, right?

• 25:38Speaker 02

  So it's still the price of a car, I think. So future will tell, but I'm much more excited about the fun, you know, hackable, tweakable, creative device of $300 to $600 than the scary $10,000 humanoid or like $20,000 humanoid.

• 25:56Speaker 01

  Yeah, well, I think I'm a little more excited about the Richie Mini right now, too. Thomas, where can our listeners connect with you or find Hugging Face's work?

• 26:06Speaker 02

  Yeah, I think the best is, I mean, for me, on Twitter, on LinkedIn, I'm pretty active. And Hugging Face, the best is Hugging Face. And I think our blog is the great place. And if you want to see Richie Mini, you should just type in Richie Mini in Google. And you basically land on the Rilis blog, which ran viral the other day, where you can also order Rilis. I'm sure I excited about trying this Richie Mini for yourself.

• 26:29Speaker 01

  Well, it sounds like a lot of people have already. I'm guessing many more will. Well, thanks so much again for joining us on the show. And to our listeners, you can find me on X at ZFmax and Equity at EquityPod. Talk to you next time.

• 26:43Rebecca Bellan?

  Equity is produced by Teresa Locan Solo with editing by Kel. We'd also like to thank TechCrunch's audience development team. Thank you so much for listening and we'll talk to you next time.