SAN FRANCISCO — A robot arm lifts a flask beneath a live camera while a stranger at a browser decides what it should do next. Elsewhere in the same online experiment, other machines draw with paintbrushes or fence with toy swords. The movements are entertaining, but the audience is also producing something Enigma considers more valuable: data about how people naturally try to control robots.
Enigma emerged from stealth on 27 July 2026 with a $71 million seed round led by Index Ventures and Ribbit Capital, with participation from Conviction Partners and individuals associated with several major AI companies. The start-up was founded less than a year ago by Jonathan Jacobi and Gal Niv and says the money will support research hiring, computing infrastructure and further real-world deployments.
Its launch project puts more than 100 proprietary robots online from facilities in California and Israel. Visitors can interact with the machines through live browser sessions. The company says the test will compare text, speech, video demonstrations and more direct controls such as tapping, dragging and dropping. According to Enigma, those interactions will help it study which interfaces feel intuitive and provide training signals for its robotics models.
A network for studying robot control
Enigma describes its technology as a three-part stack: foundation models for robotics, a hardware-agnostic abstraction layer and a user interface. The ambition is to make the same intelligence usable across different bodies, from a warehouse arm to a humanoid, without rebuilding the software for every platform. That remains a company claim. Enigma has not published comparative benchmarks showing cross-hardware transfer, detailed task-success rates, latency figures or enough information to judge how its systems recover when either the model or network connection fails.
The distinction matters because an internet-controlled robot is not necessarily an autonomous one. A machine may execute a motion locally, follow a human command, receive teleoperated demonstrations or rely on an AI policy to choose actions. These are different technical arrangements with different safety, staffing and connectivity requirements. Enigma’s public experiment appears designed to explore the boundary between them rather than present a finished commercial product.
That approach fits a wider shift in embodied AI. Robot developers are increasingly treating human demonstrations and operational data as core infrastructure. Anton Robots’ review of Apptronik Apollo 2, for example, examines a development platform built around teleoperation, simulation, fleet tools and data collection, while the Unitree G1 review considers a commercially accessible humanoid used for research, education and embodied-AI experimentation. Enigma is concentrating less on selling a particular robot body and more on the interface and learning system that might sit across many of them.
What robot buyers should watch
For companies evaluating robots, the launch is a reminder that the machine itself is only one part of the purchase. Remote operation also depends on network reliability, cybersecurity, software access, data ownership, update policies and a clear procedure for recovering from failures. Integration can easily cost more than the arm alone, as Anton Robots’ 2026 robotic arm price guide explains when accounting for tooling, vision, safety equipment and engineering.
Enigma says it is already working with partners in entertainment, retail and health, but it has not named customers, published prices or announced a generally available product. Nor has it shown that data collected from casual online users will reliably improve performance on commercially useful tasks. The network should therefore be understood as a research platform and a public demonstration, not a robot service that businesses can order today.
The experiment is nevertheless useful because it directs attention towards a problem that robotics marketing often leaves outside the frame. A robot may be physically capable of moving an object, yet still demand an expert operator, a lengthy prompt or a custom integration before it becomes useful. Enigma has raised an unusually large seed round to attack that usability gap. The harder test will come after the live streams: whether the resulting models transfer between machines, withstand ordinary environments and make robots simpler to deploy rather than merely simpler to watch.