SEAR

The robotics data race won’t be won by whoever collects the most footage It’ll be won by whoever captures the hardest, most useful real-world behavior That’s why Sear is going deep on kitchens Building toward that with @LiftoWorld

Glad to be in this lineup ❤️ We’re building the data layer for robots that need to learn from real kitchens @joinsear 🤝 @LiftoWorld

Sear is pretty simple We give you smart glasses + gloves, you cook like you normally would, and the session becomes real-world training data for home robots If it passes review, you get paid

Anyone can record a cooking session The harder part is capturing the right moments clearly enough that the movement, context and action are actually useful for training That's what we care about

Dinner is doing more than feeding people 👀 We're turning everyday cooking into real-world training data for home robots Thanks to @LiftoWorld for featuring Sear and breaking down what we're building ❤️

Big ideas need the right backing We're excited to announce our fundraising partnership with @LiftoWorld Together we're opening new doors to help bring the future of cooking and home robotics closer to reality More to come ❤️

A finished dish tells you the outcome The useful part is everything that happened before it when someone waited, adjusted, checked, corrected, or stopped eventually That’s the kind of decision-making our home robots need to learn

A recipe can tell a robot what tool to use It still has to learn how to hold it, move it, adjust it and know when the job is done That’s the part real human data teaches ♥️

Collecting footage is easy Knowing what’s actually worth training on is the hard part Every Sear session has to earn its way into the dataset through quality, visibility and useful movement More footage isn’t the goal, Better data is ❤️

Most robot training data has to be created on purpose But Sear changes that since dinner is already happening so why not train the robots on the OG data

Sear is simple: we send you smart glasses and gloves, you cook the meal you were already making, and if the session passes review, you get paid That cooking data helps home robots learn how people actually cook

Cooking is never just one move after another Every step changes what comes next and that’s exactly what robots need to learn

To a human, this is just cooking To a robot, it’s a sequence of movements, decisions and corrections it still has to learn Real kitchens are full of training data

Give five people the same recipe and you’ll probably get five different ways of making it For home robots, that variation isn’t noise It’s the real world

AI learned what objects look like from the internet Robots have a harder problem They need to learn how much pressure a tomato can take, how it moves when gripped, when it starts to slip, and how to adjust without crushing it Seeing the world was one challenge Learning how to physically interact with it is the next one

Humans really wrote “season to taste” and expected robots to figure it out

“Cook until ready” makes perfect sense to a human To a robot, it raises about 20 more questions??? How soft? How much resistance? How long? What changed? When do you stop? A lot of cooking knowledge was never written down It lives in human behavior 🧬

Robots don’t just need more data. They need data they can trust. Sear captures real cooking actions through cameras + motion-tracking gloves, then puts every session through independent human verification. Capture → Verify → Consensus → Trusted Data Because for Physical AI, seeing what happened isn’t enough. You need proof it actually happened. Sear is building Proof-of-Cook.

This might look like someone simply making breakfast But for a home robot, every reach, grip, change of angle, handoff and tiny correction is something it has to learn The finished dish isn’t the valuable part. The thousands of movements that got you there are. That’s why real-world cooking data matters

🍳🤖 5000 HOMES ARE NOW COOKING WITH SEAR Real kitchens. Real meals. First-person video + hand movement captured together. At this scale, our network can model toward ~91,000 verified hours of cooking data per year. And we’re still early. The goal isn’t more footage It’s better cooking data Sear is going deep on the kitchen