Sabi wants a person wearing a baseball cap to give an AI assistant a command without typing or speaking. On October 9, 2026, the company announced a $50 million seed round led by Khosla Ventures. It said it planned to show early versions of the interaction at CES 2027. The funding announcement gives a date to watch. It does not give a prospective wearer a measure of how often the cap gets a command right.
Sabi says its design uses sensors in the cap and a custom EEG chip. EEG records electrical activity in the brain. Recording activity is one step; identifying the particular instruction a wearer meant to give is a separate task. Sabi describes several intended uses, including turning imagined speech into text and using a trained mental command to prompt an assistant. A demo of a trained shortcut would be useful evidence for that shortcut, but would tell us little about free-form dictation.
Time the task, including the mistakes
Imagine a hypothetical demo for an office worker who can comfortably type but would rather not speak aloud in a shared room. The task is to open a draft message on a laptop without sending it. One command is enough. Compare it with the keyboard shortcut the worker already uses to open a draft on the same laptop.
Start the first attempt before the wearer puts on the cap. If they need to adjust the fit or practice the command, record that time separately from the time the command takes once they are ready. A regular wearer may pay the setup cost only once per session. A person reaching for the cap to complete one brief task pays it all at once.
Show a miss. If the command does nothing, how does the wearer know whether to try again? If it opens the wrong thing, show how the wearer closes it and returns to the task. Include a period when the person is reading quietly and has been asked not to give commands. Report how many unintended actions occurred and how long that quiet period lasted. A clean run with one practiced wearer is a useful demonstration, but its results should not be generalized to people who haven't trained with the device.
The keyboard comparison applies to this hypothetical office worker. Someone who cannot use that shortcut would need a comparison with an input method they can use, and may care more about completing the task unaided than about shaving seconds from it. The example is a proposed test, not a result or a claim about whom Sabi is targeting.
Keep the send button separate. Even if opening a draft works reliably in this hypothetical test, a command that sends a message would need a way for the wearer to see the recipient and cancel before anything leaves the computer.
When Sabi shows the cap, look for an uninterrupted attempt from setup through confirmation. If the demo shows only successful commands, the next useful evidence is a count of all attempts under the same conditions, including the ones the wearer had to repeat.