Brim

Brim is a concept for a compact sports visor that would carry Muse, Meta's personal AI agent, on your head instead of your face. It would be built on Meta's open-source Muse Gadget SDK.

It's for people who want an agent hands-free outdoors without wearing smart glasses. Nothing is built yet, so Brim's own figures are design targets, and the rest come from cited sources.

The visor would do four jobs. A haptic strap around the head would tap directions and alerts. The brim would shade the eyes and carry a wide-angle camera at its front edge.

An open-ear earpiece would read Muse's replies aloud. Much later, a display would project down from the brim. All four would run from one thumb-sized ESP32 board. (Muse here is Meta's agent, not the EEG headband that shares its name.)

My bet is that an agent's first wearable doesn't need a screen.

A visor already sits on the forehead all day in the sun, with a band around the head and a brim out front, and it never touches your eyes. I'd give each of those parts a job before adding anything new.

Build Brim on the Muse Gadget SDK

Brim would be a Muse gadget (the SDK's name for hardware you build yourself), so the agent, the account and the phone app already exist.

Meta released the ESP32 and Linux device SDKs under Apache 2.0, and at Connect 2026 it said Muse is coming to its AI glasses. Glasses get official hardware. A visor would have to be built.

The SDK already covers three things Brim needs. A gadget pairs with the Muse app over Bluetooth LE, with a press on the device to confirm. It sends push-to-talk voice notes, which Muse transcribes and answers.

And it can advertise its own commands. Muse could call a new haptic.cue command the way it calls a tool, or the SDK's camera.capture, which today runs on the SenseCAP Watcher.

The catch? Muse answers in text. The SDK's README suggests sending each reply to a text-to-speech service of your choice, and it names the spot in its chat code where the speaker is already wired up. Brim's earpiece would hang off that hook.

The controller would be a Seeed XIAO ESP32-S3 Sense: 21 by 17.8 mm, with 8 MB of PSRAM, a detachable camera and a microphone.

The SDK already runs a plain XIAO ESP32-S3 experimentally. The camera-equipped Sense would need a port.

              Muse (your agent's cloud VM)
                          ▲
                          │ Wi-Fi, encrypted Noise session
                          ▼
       ┌────────── XIAO ESP32-S3 Sense ──────────┐
       │  push-to-talk mic · commands Muse calls │
       └────┬─────────────┬──────────────┬───────┘
            │             │              │
       brim edge      head strap      right ear
       wide-angle     8 coin motors   I2S amp and
       camera, light  haptic.cue      open-ear speaker
       camera.capture                 (spoken replies)

I'd order the outputs by what they cost you, and I'm calling that order "the attention ladder". Feel first, hear second, see last. A tap costs a moment. A voice costs a sentence. A display costs your eyes.

Keep the visor compact: camera and UV in one brim

I'd keep Brim's brim short, about 6 cm, even though a longer one protects more skin.

Obvious version: a wide tennis visor, because more brim means more shade. It would fail on the head.

The camera would sit at the brim's front edge, so every extra centimetre adds torque that tips the visor forward and presses the band into the forehead. A short brim keeps that weight close. The battery would ride at the back of the strap as a counterweight.

The cost is real. Diffey and Cheeseman measured sun hats on model heads in a 1992 study and found a brim has to be wider than 7.5 cm to give reasonable protection to the nose and cheeks.¹

At 6 cm, Brim would shade the eyes and forehead, not the whole face, and a visor has no crown to cover the scalp. Sunscreen stays.

My test: "Could I run for an hour without the band creeping up my forehead?"

The camera would look where your face points, through a wide-angle lens. I'd wire its indicator light to the camera's power rail rather than switch it in firmware, so no code path could run the camera with the light off.

Feel it first: a haptic strap, one motor at a time

I'd fire one motor at a time and keep directions on the sides of the head.

The strap was inspired by Walnut, a neural-interface startup founded in Stanford, California, which is building an EEG headband. I'm borrowing the place, not the sensing. Brim's strap would only output.

Obvious version: buzz several motors at once to draw a shape, like an arrow sweeping left.

A 2012 IEEE study by Dobrzynski and colleagues put 12 coin motors around the heads of 22 people. They counted one active motor correctly 94% of the time, two motors 40% of the time, and five just 5%. Shapes don't survive the skin.

The same study placed a buzz least precisely at the very front and back of the head.

So the forehead motor would only ever mean "Muse has something", where precision doesn't matter. "Behind you" would alternate between the rear corners instead.

Eight motors would sit in the band, 45° apart, which covers four cues.

Hair matters too. Myles and colleagues found in a 2015 study that dense hair may slightly dampen vibration at the head's least sensitive spots.

The forehead motor would sit on skin and most of the others over hair, so I'd make each motor's strength adjustable.

My test: "Can I tell left from right with my eyes closed, while walking?"

That test isn't hypothetical for everyone. Dobrzynski's study opens by saying head-worn vibration can extend the perception of visually impaired people, and taps would leave the ears free for anyone who navigates by the sound of traffic.

Deaf and hard-of-hearing wearers could get the same cues without the earpiece. I'd want visually impaired and deaf testers shaping the cues before anyone relies on them.

Hear it second: one open ear, on a wire

I'd put the earpiece just outside the right ear, on a short arm from the strap, and leave the left ear free.

Why a wire? Because the chip can't stream to earbuds. Espressif's documentation says the ESP32-S3 supports Bluetooth LE only, with no Classic Bluetooth or LE Audio.

A wire along the strap to a small I2S amplifier also means one battery to charge, not two.

Open-ear matters outdoors, where traffic and a running partner should still reach you. The strap would play the Muse double tap before Muse speaks, so a voice never starts cold. You'd feel a reply coming before you hear it.

See it last: a holographic display under the brim

The top rung is a display projected down from the brim, and it's the part I'm least sure can ship.

A hologram can't float in open air, because you only see light that scatters off something or is steered into your eye.

The honest version is a holographic film on a small clear combiner that flips down under the brim, steering the projector's light into the eye and letting the world through.

North's Focals glasses did this inside a lens, with a laser projector, at about 300 by 300 pixels (The Verge's figure, cited in Karl Guttag's analysis).

Glasses have one big advantage. They sit on the nose and ears, locked to your eyes, which is how Meta's Ray-Ban Display holds a 600 by 600 display in the right lens.

A visor rides on the forehead. A frown or a shifting strap could slide the image out of the eye box (the small zone where it's visible).

Two things could make it worth trying anyway. The brim would already shade the combiner from overhead sun, which should help contrast. And the optics would stay off your face, so it could work over prescription glasses or sunglasses.

What Brim can't do, and what I haven't tested

• It won't replace sunscreen. A 6 cm brim would shade the eyes, not the nose and cheeks, and there's no crown.
• It won't read your mind. The strap would only tap. Walnut inspired where it sits, not what it senses.
• It won't replace a cane or a guide dog. The strap would point the way, but nothing on Brim would detect obstacles.
• It won't talk to your earbuds. The ESP32-S3 has no Classic Bluetooth, so the earpiece would be wired.
• It wouldn't film quietly, but the light can fail. The light would share the camera's power, so firmware couldn't switch one on without the other. A dead LED would still hide a capture.
• It won't work without a connection. SDK gadgets reach Muse over Wi-Fi, so on a run Brim would lean on your phone's hotspot.
• It can't fix the SDK's pairing limits. The README warns that community pairing can't stop an active man-in-the-middle attack.
• It has no battery figure. The SDK ships a power-profiling tool, and I'd measure before I quote a number.

The next step is a strap-only prototype: eight motors on a XIAO, the four cues, and a week of wearing it before a camera goes anywhere near it.

Most moments should stop at the bottom of the attention ladder: a tap. For another hardware concept, see Dexy, a cyberdeck for engineers who supervise coding agents.