Here is a thing that is true about robots, and that nobody selling you a robot really wants to dwell on: robots fail constantly, and when they fail, they don't tell you why. A self-driving car swerves. A warehouse arm drops a box. A drone loses its footing on a gust. Somewhere inside the machine there is a perfectly complete record of exactly what happened - camera feeds, lidar point clouds, GPS traces, the readings from a dozen sensors, all time-stamped to the millisecond. The record exists. The problem is that the record is enormous, it is in a weird format, and no human being can watch it in real time. This is the problem Foxglove sells a solution to.
Foxglove is a San Francisco company, founded in 2021, that builds what it calls a data and observability platform for robotics. In plain terms: it is the software you use to record everything your robot did, ship that recording off the robot and into the cloud, search across mountains of it, and then play it back frame by frame in 3D so you can figure out what went wrong. If that sounds unglamorous, that is sort of the point. The glamorous part of robotics - the foundation models, the dexterous hands, the videos of humanoids doing backflips - gets all the attention. The unglamorous part is data engineering, and it turns out the unglamorous part is where most of the actual work lives.
The founders had already built this once
The two people who started Foxglove, Adrian MacNeil and Roman Shtylman, did not arrive at this problem theoretically. They worked together at Cruise, the autonomous-vehicle company, where they built more or less exactly this kind of internal tooling for a fleet of self-driving cars. MacNeil had also been the first director of engineering at Coinbase, which is a useful thing to have on your resume when you want to convince investors you can build infrastructure that does not fall over. So the Foxglove origin story is not "we had a brilliant idea." It is closer to "we had already suffered through this problem inside one company, we watched every other robotics company suffer through the identical problem, and we decided nobody should have to build it a fourth or fifth or fiftieth time."
"We used to reinvent everything. Infrastructure requires lots of money and time. When you don't need that, it dramatically shortcuts getting products to market."
That last observation is the whole business. Every robotics team, left to its own devices, ends up rebuilding the same plumbing: a way to log sensor data, a way to move it, a way to look at it. It is expensive, it is slow, and it is not the thing that makes your robot special. Foxglove's pitch is that you should stop doing that and rent the plumbing instead.
The clever move: give away the format
The most interesting thing Foxglove has done is not its paid product. It is a file format called MCAP, which the company built and then released as open source. MCAP is a container for the kind of messy, mixed data a robot produces - video and point clouds and JSON and Protobuf messages, all in one file, each with its schema embedded so the file is still readable years later. It has reader and writer libraries in six languages, which means teams can plug it into whatever stack they already run without adopting anyone's worldview.
Then the interesting thing happened: MCAP became the default bag recording format in ROS 2, the dominant open-source robotics framework, starting with its Iron release. When you run ros2 bag record today, you are, by default, writing a file format that Foxglove created. There is an old lesson in developer tools here, and Foxglove executed it cleanly - the fastest way to own the standard is to give it away, and then sell the platform that reads it best.
"Robots are eating the world that software could not."
Foxglove · Series B announcement
Why now, and why $40 million
In November 2025, Foxglove raised a $40 million Series B led by Bessemer Venture Partners, with its existing backers Eclipse and Amplify Partners along for the ride. That brings the total raised past $62 million since the seed round in 2021. The framing everyone uses for this is "Physical AI" - the idea that foundation models, cheap sensors, and edge compute have converged to the point where machines can finally sense, think, and act in the real world. It is a real trend and also a convenient one for Foxglove, because every new robot in the world is a new machine producing data that somebody has to store, search, and visualize.
The customer list does the arguing here. NVIDIA, Amazon, Anduril, Wayve, Waabi, Saronic, Dexterity, Bedrock Robotics, The Bot Company - these are companies with almost nothing in common except that each of them builds machines that generate more data than any human can watch. Wayve, the autonomous-driving company, says Foxglove cut its troubleshooting from days to minutes. That is not a story about a smarter algorithm. It is a story about being able to see data you already had.
Whether Foxglove becomes the durable infrastructure layer for robotics or just an early winner in a crowded field is genuinely unsettled - it competes with ROS's built-in tools, with newer entrants like Rerun and Sift, and with the eternal temptation of every engineering team to just build it themselves. But the bet underneath the company is simple and hard to argue with: if robots are going to be everywhere, then the boring work of remembering what they did is going to be a very large business.
