A new London-based startup is challenging the dominance of traditional typing by introducing a dedicated voice-to-text platform designed to reduce reliance on computer keyboards. Founded by former Nothing employees Cookie Xu, who led AI and services, and Raymond Zhu, a hardware product manager, Relay is debuting its macOS desktop software today. Alongside the software launch, the company announced that its companion hardware device, dubbed the Relay Q, is slated for release in early 2027.
The Office Noise Dilemma That Inspired Relay
During their tenure at Nothing, Cookie Xu and Raymond Zhu relied heavily on transcription tools like Wispr Flow to draft work messages and brainstorm. While dictation accelerated their thought process, operating in an open-plan office presented constant acoustic challenges regarding input volume.
Achieving the right vocal volume in a shared workspace proved exceedingly difficult. As Raymond Zhu noted, "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." To overcome this limitation, he resorted to holding an earbud directly in front of his mouth, highlighting the clear necessity for dedicated directional hardware.
Relay chose macOS as its initial deployment target because desktop workflows involve substantially higher communication complexity than mobile devices. While mobile messaging often consists of brief personal exchanges, computer-based work demands structured ideation, varied tone, and contextual adaptability across professional tools.
macOS Integration, Automation Skills, and Gemini AI
The Relay software operates independently on macOS without requiring specialized hardware. During setup, users configure a global hotkey, with the Fn key set as default. Holding the hotkey while the cursor resides in any text field activates Relay's voice dictation engine, which leverages Google's Gemini models for processing.
Beyond standard dictation, Relay introduces automated contextual actions termed Skills. For instance, instructing Relay to send a message via Slack allows the app to open the workspace, locate the designated recipient, and draft the text automatically. While tests showed reliable calendar event creation in Google Calendar, complex app integrations like Slack encountered minor beta glitches. Relay includes built-in Skills for popular software such as Gmail and WhatsApp, while allowing users to construct custom integrations.
System Permissions and Privacy Architecture
Executing contextual Skills requires granting Relay extensive system-level permissions. In addition to microphone access, the software requests permission to capture a visual snapshot of the screen upon activation. This enables the AI to interpret active applications and navigate software interfaces.
Regarding data handling, Relay stores user data locally on the host device, providing options for manual editing or deletion. Data transmitted for processing passes through Relay's infrastructure en route to Google's servers without being stored permanently on Relay servers. Nevertheless, granting AI systems screen-monitoring privileges introduces inherent security and privacy trade-offs comparable to features like Microsoft's Recall.
Accuracy, Contextual Awareness, and Customization
Relay's transcription capabilities currently do not reach the precision levels observed in features like Google's Rambler on the Pixel 11 series. Initial testing revealed occasional punctuation errors and word misinterpretations. However, the system demonstrates solid capabilities in filtering false starts, self-corrections, and verbal filler.
The system also handles compound instructions effectively. Dictating a message regarding dinner plans while simultaneously asking Relay to insert a Google Maps link resulted in proper execution. When tested on hardware configurations like a Mac Mini paired with an Apple Studio Display, a slight latency occurred between hotkey activation and speech recognition, an issue Relay expects dedicated hardware to resolve.
Users can tailor the app to match personal communication habits through custom system prompts. Rules can be established to omit trailing periods in Slack or automatically alter communication tone depending on the recipient. Relay detects frequent contacts to adjust formality levels or translate messages into foreign languages for international colleagues.
Relay Q Hardware: Design and Wearable Functionality
The upcoming Relay Q hardware consists of a compact cylindrical device featuring an end-mounted microphone array shaped like the letter Q. Users can press and hold the top button to trigger dictation on desktop or mobile environments, or detach the button module to clip it directly to clothing.
Designed specifically for low-volume dictation, the device allows users to whisper quietly while maintaining high transcription accuracy. When reflecting on the design goal, Raymond Zhu posed a central operational question: "How long can you go with your workday without touching your keyboard?"
Market Landscape, Smart Ring Competitors, and Pricing
Dedicated voice dictation hardware has seen growing interest across the hardware industry. Competitors like Sandbar recently introduced the Stream Ring, while Pebble unveiled the Index 01, both focusing on wearable ring form factors. Relay evaluated a ring design but ultimately pursued a cylindrical microphone wand to differentiate itself in an increasingly crowded smart ring market.
The Relay Q will ship with a desktop charging dock alongside an optional MagSafe attachment compatible with iPhone and Qi2 devices. The hardware will retail for $150, which includes a one-year subscription to the software service. Standalone software access will cost $15 per month following the included period, though hardware purchase remains optional for software users.
