How AI glasses work — and where the AI actually is.

The four parts in the frame

Take any camera-and-audio pair apart — ours, the big-brand ones, the $99 ones on a marketplace — and you find the same four components. The differences are in quality, weight and whose app is on the other end.

1. A camera in the front corner. On OXIVUE frames it is an 8 MP sensor (a Sony IMX219 on the M02C and M14) set flush in the corner of the frame, visible from the outside. It takes photos and short 1080p clips from your eyeline.
2. Open-ear speakers in the temples. Two directional drivers above the ears, nothing in the ear canal. This is the part that makes the product a pair of glasses rather than a headset with lenses.
3. Microphones. For calls, for the voice assistant, and for translation — they are what hears the other person. The M14 adds a bone-conduction mic and dual-mic noise cancelling for wind.
4. A radio and a battery. Bluetooth 5.4 to your phone and a cell of roughly 300 mAh. That battery is the reason everything intelligent happens somewhere else.

What is not in the frame: a display, a processor big enough to run a language model, GPS, a heart-rate sensor. On the M14 there is onboard memory (4 GB) for capture; the others rely on the app.

Where the AI runs — not on your face

A frame that weighs 41 to 52 grams cannot carry the chip or the battery that a language or vision model needs. So the intelligence lives in two places outside the glasses:

• The companion app on your phone handles pairing, settings, the file transfer and the wake-word. On OXIVUE frames that is the free HeyCyan app; it needs no platform account and no subscription.
• Cloud modelsdo the understanding: speech recognition, translation, and on the M02C the “Hey Cyan” visual assistant that describes what the camera sees. The app sends audio or an image up, gets text or speech back, and plays it through the speakers.

This is why every honest spec sheet in the category says “requires a smartphone” and why the assistant goes quiet in a tunnel. It is also why the big-brand pair asks you to sign into its platform: the app is the product. The trade-off you are actually choosing between brands is whose app, whose account, and whether there is a subscription — not which frame has the better AI chip, because none of them have one. Our no-account guide goes through that trade.

How a sentence in Spanish becomes a sentence in your ear

The pipeline is five steps, and the delay you feel is steps two to four:

1. The microphones in the frame pick up the other person speaking.
2. The audio goes over Bluetooth to the app on your phone.
3. The app sends it to a translation model over data or Wi-Fi.
4. The translated speech comes back to the app.
5. It plays through the open-ear speakers — you hear English, they keep speaking Spanish.

The other direction works the same way, so a conversation is two pipelines running in turn. Three consequences follow, and they are true of every product that translates this way: it needs the phone with you, it needs a connection, and there is a pause after each sentence. The M08 family covers 40+ languages and the M14 139 through the same app. The translation guide is specific about what the pause feels like and where it stumbles, and the translation page shows the frames.

How capture works — and why nothing records on its own

Capture is a deliberate touch on the temple. There is no always-on recording, no motion trigger and no scheduled capture; if you do not touch the frame, the camera is idle. A photo is an 8 MP JPG from your eyeline; a clip is 1080p at 30 fps, with electronic stabilisation on the M14. On frames with onboard memory the file lands in the frame and syncs to the app later over Bluetooth or, on the M14, Wi-Fi; on the others the app is where it lands.

The camera is visible in the frame corner from the outside, and the etiquette and law around a wearable camera are real — the recording-laws guide is the honest read before you wear one into a gym, a courtroom or a conversation.

Audio without earbuds

The speakers sit above the ear and beam sound down into it. Because the canal stays open you hear the room, the road and the person you are talking to; because the drivers are directional, the leakage at normal volume is modest. Turn it up in a quiet room and the person next to you will hear something. There is no music stored in the glasses and no standalone streaming — they are a Bluetooth 5.4 headset, and the audio comes from your phone.

Battery and charging

Roughly 300 mAh, which is why the split between listening and calls is the number to read. The M08 family is listening-biased — about 7 hours of audio, 4 of calls — and the M14 is call-biased at 4 and 7. Capture alone uses less than streaming music. Charging is about 2 hours; the M02C uses a magnetic pogo connector. Every frame charges from a cable, and none of them charge from a case.

What AI glasses cannot do

• Nothing in the lens. No captions, notifications, maps or metrics. That is AR, a different and heavier category.
• Nothing without a phone. Assistant, translation and streaming all need the app; only capture on the onboard-memory frames works with the phone elsewhere.
• Nothing offline that needs a model. Translation and the visual assistant need a connection.
• Not waterproof. IPX4 and IP65 are splash ratings; see the water-sports guide.
• No prescription lenses in our line — the prescription guide is candid about the options.

Bottom line

AI glasses are a hands-free camera, an open-ear headset and a microphone for an assistant that lives on your phone. Judge a pair on the four parts you can touch — camera, speakers, microphones, battery — and on whose app you are agreeing to, because the intelligence is the same shape everywhere. The OXIVUE M02C at $159 is the most complete version of that in our line: 41 g, IP65, an 8 MP Sony sensor with stabilisation, the visual assistant, and a free app with no account. The M08 at $149 is the same idea with fewer extras. Every spec side by side on the comparison page.

Frequently asked questions