Tool Suggestions
Dream API Turns a Camera Feed and a Plain-English Instruction Into Robot Commands
On October 19, San Francisco startup Dream Labs announced $90 million in seed funding and, alongside it, Dream API — which the company calls the first publicly available API for a robot foundation model. The pitch is simple: feed it a robot's camera view and a plain-English instruction, and it returns motor commands — even for robots it has never seen before. Factories worldwide run a record 5 million industrial robots, yet nearly every one is programmed for a single task, and what it learns cannot transfer to another machine. Low-cost robots now sell for a few thousand dollars but ship with no built-in intelligence. Teaching one robot one task can take specialists months, and a new task or new hardware often means starting from zero. Dream Labs is betting that the missing piece is not better hardware, but a brain any robot can use.
How Dream-1 works: video, not robot demonstrations
The flagship model, Dream-1, works across robots from different manufacturers because it is trained on video of people performing everyday tasks, not on demonstrations from any single machine. It predicts what will happen next in a scene and determines the robot's resulting action — and because it learns from video rather than from one robot's sensors and joints, its capabilities transfer across hardware. Out of the box, Dream-1 can control any robot with grippers, zero-shot, with no additional training data required. The company was founded in 2026 by former leads of NVIDIA's GR00T robotics research, who previously created DreamZero and World Action Models, and it launched with more than 20 hardware partners on board. Supported robots today cost about $2,000 to $10,000. Alongside the API, Dream Labs launched Fine-tuning as a Service: developers and robot makers with unsupported hardware or new needs can submit a small batch of demonstration data and receive a tailored model within hours. The company says it will use the funding mainly for compute, to expand the tasks and robots Dream-1 handles and to support more users.
What it can do today — and how to test it without owning a robot
The honest scope is narrower than the pitch. Dream-1 currently handles tabletop tasks such as picking and placing objects and folding laundry. Support for wheeled mobile robots and more dexterous hands is planned, not shipped. That is the right order for a general robot brain — tabletop manipulation first, mobility later — but it means the product today is a tabletop tool, not a general-purpose robot worker. The most interesting part of the launch is how the company invites scepticism: the Robot Playground lets anyone with a browser type commands to a robot in Dream Labs' San Francisco office and watch it respond live and unedited. CEO Joel Jang framed it deliberately: "The hardware is ready. What’s been missing is a brain any robot can use. When you get into a rental car, you don’t relearn how to drive. You find the controls and your past driving experience does the rest. Dream-1 approaches a new robot the same way, so a new robot can do useful work in minutes, not months. And instead of edited demos, we’re giving people a working model they can test themselves, mistakes included." That last clause is the one to hold them to. A live, unedited robot that makes mistakes in public is a far stronger claim than a demo reel — and the Playground is where the claim can be checked.
Who should try it, and what to watch
Dream API is most useful right now for two groups: developers building on low-cost robots who would otherwise spend months teaching one task, and robot makers whose hardware ships without intelligence. If you are in either group, the Playground is the fastest evaluation: try the tabletop tasks, watch where it fails, and judge whether the failure modes are acceptable for your use. The things to watch over the next months are whether the partner count grows past 20, whether wheeled robots and dexterous hands actually ship, and whether the fine-tuning service delivers tailored models in hours as promised. The funding — $90 million led by B Capital along with Kindred Ventures, with participation from A.Capital, Unusual Ventures, SV Angel, Wing VC, Scribble VC and Adrian Aoun — buys compute, not proof. The proof is a robot you have never programmed doing useful work in minutes. Dream Labs has given everyone a browser window to check that claim. Use it before you believe it.