Dream Labs raises $90M and opens a public API for one robot brain

Two things at once. First, the money: $90 million in seed funding, led by B Capital and Kindred Ventures. A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and Adrian Aoun joined, among others.

Second, a product: Dream API. Dream Labs calls it the first publicly available API for a robot foundation model. An API is a doorway that lets developers plug a service into their own software.

Here, the doorway takes a robot's camera feed and a plain-English instruction and hands back motor commands. That works even for robots the model has never seen before, the company says.

By the numbers: $90 million raised, more than 20 hardware partners, and supported robots that cost about $2,000 to $10,000 each.

Dream Labs was founded in 2026 by former leads of NVIDIA's GR00T robotics research. Its press release says the founders created DreamZero and World Action Models while at NVIDIA. The company is based in San Francisco.

Joel Jang is co-founder and CEO. The Robot Report notes that B Capital general partner Daisy Cai said the team "did some of the defining work in robot learning at NVIDIA." She added: "Now anyone with an API key can build on it."

Dream-1 is the company's flagship model. Most robots are taught with data from the robot itself, often through a person guiding it. Dream-1 takes a different road.

It is trained on video of people doing everyday tasks. From that, it predicts what will happen next and works out what action the robot should take. Because it learns from video and not from one machine, Dream Labs says its skills carry over to other hardware.

Out of the box, the company says Dream-1 can control any robot with grippers, "zero-shot." That means it tries the job without extra examples first. For a two-armed robot with parallel grippers, it needs no additional data.

In real life Picture a $2,000 robot arm that arrives with no skills. You connect it, type "fold the towel," and the arm tries. No months of programming first.

Jang used a car comparison. "When you get into a rental car, you don't relearn how to drive," he said. "You find the controls, and your past driving experience does the rest." In his words, a new robot can do useful work "in minutes, not months."

Dream Labs starts from a plain observation. Factories run a record 5 million industrial robots, and nearly every one is set up for a single task. What one learns usually cannot move to another machine.

Meanwhile, low-cost robots now sell for a few thousand dollars but ship with no built-in intelligence. The company says teaching one robot one task can take specialists months, and a new task or new hardware often means starting from zero.

That is the gap Dream Labs wants to fill. The Robot Report describes the idea as a model that lets robots transfer what they learn to one another.

The big picture: hardware has become cheap faster than software has become smart. Dream Labs is betting that the missing piece is a general brain, sold as a service.

Three products launched together, according to the company:

• Dream API: developers sign up, connect a robot with grippers and the robot carries out the task. The models run in data centers around the world, so robots respond quickly wherever they are.
• Fine-tuning as a service: for unsupported hardware or new skills, developers submit a small batch of demonstration data and get a tailored model within hours.
• Robot Playground: anyone with a browser can type commands to a robot in Dream Labs' San Francisco office and watch it respond live and unedited.

Jang pointed to that last one as a matter of honesty. "Instead of edited demos, we're giving people a working model they can test themselves, mistakes included," he said.

The honest answer is tabletop work. Dream-1 currently handles picking and placing objects and folding laundry. Support for wheeled mobile robots and dexterous hands is planned, which would open the door to bigger jobs.

The range of robots is also narrow on purpose. Dream Labs supports machines that cost between about $2,000 and $10,000. For its more than 20 hardware partners, it goes beyond the baseline and refines the model for each robot so it performs at its best.

Ryan Chan, co-founder and CEO of MakerMods, is one of those voices. His company sells $2,000 bimanual robot arms. He said his team's robot "was up and running in a few minutes and attempting tasks we never programmed."

That is one customer's view, shared in the announcement, so treat it as a first impression and not a benchmark.

Mostly into computing power. Dream Labs says it will use the funding primarily for compute as it expands the tasks and robots Dream-1 can handle and scales to support more users.

That fits the product. A model that runs in data centers and answers a robot's camera in real time needs a lot of chips running all the time.

The Robot Report adds one more detail about timing. The story ran on October 22, a few days after the announcement, as part of a week full of robotics news, including NVIDIA's Isaac updates and a new research humanoid from Agile Robots. We covered that one in our Agile ONE SN story.

Plenty of companies are building brains for robots, and most tie them to their own bodies. A humanoid maker trains its model on its own robot, and the skills stay home.

Dream Labs takes the opposite side. It wants to be the shared layer that many makers can use, in the way a phone maker uses someone else's operating system.

We have looked at a similar idea from another angle, with Runway's world model for robots. Both bets rest on the same belief: video of the real world can teach machines how to act.

Whether one model can really serve many bodies is still the open question. The company's own list of tasks today is short, and independent testing will matter more than any launch announcement.

• Developers: you can try robot control through an API today, with a browser playground to test it before you buy hardware.
• Hobbyists and makers: low-cost arms in the $2,000 range may soon come with a ready-made brain, so the hard part shifts from code to ideas.
• Everyone else: nothing changes at home tomorrow. This is early tabletop work, and the company says wheeled robots and dexterous hands are still to come.
• Watchers of the robot race: a shared brain is a different bet from one robot, one model. It is worth following who signs on.