The companion app is available for Mac. The hardware companion runs on ESP32-S3 boards with AMOLED displays (2.16" 480×480 or 1.8" 368×448), which ship from a third-party store.

Does Zeph require a cloud connection?

No. Zeph is local-first and encrypted — audio and state stay on-device by default with no cloud round-trip required. You can also bring your own model or use the cloud if you prefer.

Can I build apps for Zeph?

Yes. A free SDK lets you build sandboxed WASM apps — a glance, a tool, or a whole app — and one SDK is portable across every Zeph device, with sample apps, docs, and community support included.

Zeph — Voice-first hardware companion for your AI agents

Zeph is a voice-first physical AI interface: a small desk hardware companion paired with a macOS app that lets you push-to-talk dictate into any app system-wide, watch your AI agents' status (rate limits, CI, tasks) on an always-on screen, and approve agent actions with a press. It's local-first and encrypted, works with the AI tools you already use (Claude, Cursor, Gemini, GitHub Copilot, and more), and supports sandboxed WASM apps built on a single SDK that runs across all Zeph devices.

Key Features

System-wide push-to-talk dictation

Press to talk and your voice is typed wherever your cursor is, in any app and any field — no hotkeys or laptop mic needed.

Always-on agent glances

The device screen shows the signals you care about at a glance: agent rate limits, CI status, and what's next on your calendar.

Agent replies and approvals

Your agents push what needs your attention — an approval, a result, a heads-up — to the screen via MCP, and you respond right on the device without reaching for your Mac.

Works with your existing AI tools

Compatible with Claude, Claude Code, Cursor, Google Gemini, GitHub Copilot, Perplexity, Obsidian, Notion, Raycast, Warp, Zed, Ollama, and Linear; bring your own model or use the cloud.

Private and local-first

Audio and state stay on-device by default — encrypted, with no cloud round-trip required.

Open WASM app ecosystem

Build sandboxed, portable WASM apps with one SDK that runs on every Zeph device — ship a glance, a tool, or a whole app.

Hands-free notes

Capture distraction-free notes and thoughts by voice the moment they land — no typing, no phone in hand.

How It Works

Press to talk

Hold the physical companion and speak; your dictation or prompt reaches your agents or types into any app via the macOS companion app.

Glance to read

The always-on AMOLED face displays live status from your agents — rate limits, CI, standup times — and pushes items that need you via MCP.

Answer on the device

Approve a deploy, dismiss, or respond with one press, without leaving your editor or reaching for your laptop.

Pros & Cons

Pros

No ongoing subscription — buy the device once and it stays yours
Local-first and encrypted: audio and state stay on-device by default with no cloud round-trip required
Works with a wide range of tools users already have (Claude, Cursor, Gemini, Copilot, Notion, Ollama, and more), plus bring-your-own-model support
Open SDK for sandboxed, portable WASM apps — apps run on every Zeph board
Free developer tier with full SDK, sample apps, docs, and community support
Physical push-to-talk removes hotkey friction and laptop-mic distance issues

Who It's For

Best for

Developers running AI coding agents (Claude Code, Cursor, Copilot) who want voice prompts, rate-limit visibility, and one-press approvals
Voice-first workers who want hands-free dictation into any app without a phone or hotkey
Privacy-conscious users who want audio and state kept on-device rather than in the cloud
Builders and makers interested in developing sandboxed WASM apps for AI hardware
Owners of existing AI toolchains who want a dedicated always-on desk interface for their agents

Not ideal for

Windows or Linux users — the companion app is currently Mac-only

Use Cases

Dictating into any app system-wide without touching the keyboard

Asking your AI agents questions out loud and watching runs and rate limits on the device screen

Approving deploys and agent actions with a single press while coding

Capturing distraction-free voice notes and ideas the moment they land

Monitoring CI status, rate limits, and next calendar events at a glance

Building and publishing sandboxed WASM apps for a dedicated AI hardware device

Pricing

Hardware-first pricing: the Zeph One device plus macOS app is a $149 one-time purchase with no subscriptions to talk to your own agents. Building for the platform is free forever. Team/fleet pricing is not published — you contact the company to scope it.

Zeph One

$149 one-time

The hardware companion plus the macOS app — push-to-talk voice for the agents you already run. Includes the Zeph device, macOS companion app, push-to-talk dictation, on-device local-first operation, and Bluetooth pairing. Currently in early

Developer

Free

Full WASM app SDK, local-first runtime, ecosystem publishing, sample apps and docs, and community support — free forever.

Teams (Fleet)

Contact for pricing

Coming soon. Includes bulk provisioning, a fleet dashboard, priority support, SSO and policy controls, and guided onboarding. Pricing scoped per team via contact.

FAQ

What is Zeph?

Zeph is a voice-first physical AI interface: a small hardware companion that sits on your desk, paired with a macOS app. You press it to talk to your AI agents or dictate into any app, and the always-on screen shows agent status like rate limits and CI.

How much does Zeph cost?

Zeph One (the device plus macOS companion app) costs $149 as a one-time purchase — there are no subscriptions. The developer SDK is free forever. Team/fleet pricing is scoped through contact with the company.

Is Zeph open source?

The page doesn't claim the product is open source, but it offers a free, open WASM SDK for building sandboxed apps that run on every Zeph device, and maintains a GitHub presence.

Where does Zeph run?

Zeph