Voice-to-text has moved beyond simple transcription, but it still depends on the microphones built into laptops and phones — and that, according to WIRED reporter Julian Chokkattu, is the gap a new London startup called Relay is attacking. Relay is building the Relay Q, a dedicated portable microphone for high-fidelity voice-to-text anywhere, paired with a macOS app that can transcribe speech, correct misspoken sentences, and automate actions inside other applications. The software debuts now; the Relay Q hardware follows in early 2027, WIRED reported.
Why a keyboard is the bottleneck
Relay's founding story is rooted in the shortcomings of existing dictation tools. Raymond Zhu, Relay's co-founder and formerly a hardware product manager at Nothing, told WIRED that he and co-founder Cookie Xu — Nothing's former head of AI and services — used Wispr Flow's transcription software to "type" most of their messages at work. The software helped them think through ideas, he said, but the office created a physical problem.
"In the office, we never really could get the right volume for input," Zhu told WIRED. "If I whisper too quietly, the Mac wouldn't be able to pick it up; too loud and my coworker next to me can hear." He added: "It calls for dedicated hardware."
That experience shaped the product roadmap. Relay is starting with desktop instead of mobile because Zhu believes work conversations are the most complex, WIRED reported — chats with friends and family are simple on phones, but in front of a computer people share ideas, brainstorm, and use varied tone and formality.
The Relay Q and the founding team
Relay is based in London, founded by Xu and Zhu, both former Nothing employees. The company is debuting its macOS app now, with plans to bring it to Windows, iOS, and Android. The Relay Q microphone is not expected until early 2027.
| Attribute | Detail |
|---|---|
| Company | Relay (London) |
| Founders | Cookie Xu, Raymond Zhu (former Nothing employees) |
| Software | macOS app, debuting now |
| Hardware | Relay Q portable microphone, early 2027 |
| Voice engine | Google's Gemini models |
| Launch platforms | macOS first; Windows, iOS, Android planned |
| Data handling | Local storage on device; Relay servers transit data to Google, not retained |
How the Relay app works
Chokkattu tested a beta version of the macOS app with and without the dedicated microphone. During setup, users choose a hotkey — the default is Fn — then press and hold it while the cursor sits in any text field to activate dictation. The transcription engine is powered by Google's Gemini models.
Relay can also perform contextual actions it calls Skills. One Skill, for example, asks Relay to message a colleague in Slack; Relay launches the app, finds the recipient, and drafts the text. Chokkattu reported trouble with that Slack skill in testing, but the app successfully crafted a calendar event in Google Calendar. Preset Skills exist for WhatsApp and Gmail, and users can make their own and connect other apps.
Privacy trade-offs and the early verdict
Skills require sweeping system permissions. Relay needs microphone access, and users can grant it permission to record a snapshot of the screen whenever it is activated — that screen capture is how Skills identify what to automate. The company says data is stored locally on the user's device, where users can manually edit or delete it. When data is sent for processing, it passes through Relay's servers in transit to Google's servers, but Relay does not retain it. Chokkattu noted that granting AI models a view of the operating system adds inherent privacy and security trade-offs, similar to Microsoft's Recall feature.
Hands-on results were mixed. Relay's transcription was not as impressive as Google's Rambler feature on the Pixel 11 series, Chokkattu wrote; it does not capture every word perfectly and sometimes misplaces punctuation. It does understand when the speaker misspeaks and restates a sentence, cutting the parts that do not matter. Emoji insertion was inconsistent: when Chokkattu asked for a heart emoji, Relay did it flawlessly, but when he tried to add the same request at the end of a sentence, it typed out "heart emoji."
The competitive field
Relay enters a market that WIRED describes as software-focused. Wispr Flow can join meetings to summarize discussions, and Google is debuting smarter on-device dictation on its latest Pixel smartphone. Relay's differentiation is dedicated hardware: a portable microphone for high-fidelity voice-to-text anywhere, built to solve the office-volume problem Zhu described.
WIRED's testing suggests the software is functional but not yet best-in-class at transcription, and the Skills automation carries real privacy costs. The company plans to bring the app to Windows, iOS, and Android, with the Relay Q arriving in early 2027. Zhu's office-dictation problem — demonstrated to WIRED by pulling an earbud from his ear and speaking into it — is the gap the hardware is designed to close.
Sources:
