ZZeph
Library

Everything you add to Zeph

Model and speech providers, coding agents that drive it over MCP, apps and skills. Added in the companion, running on the device or your Mac.

20 in the library6 run on your Macyour key or oursBrowse the full library

Featured

Keep it all local.

No key. No account. Nothing leaves the box. The model and the microphone both stay on your machine.

OllamaLM StudioApple IntelligenceWhisperMLX WhisperParakeet
A ZephOS transcript screen: a waveform above the words “Running ten minutes late, start without me.”

Bring your own

Providers

The models and speech engines Zeph can speak to. Run them on your Mac, or point the device at a key you already have.

Runs on your Mac

6

Apple Intelligence

On-device language model via Apple Foundation Models. No key, no download.

Runs on your MacFreeLLM

LM Studio

Local language models + embeddings via LM Studio's OpenAI-compatible server.

Runs on your MacFreeLLMEmbeddings

MLX Whisper

On-device speech-to-text via Apple MLX (Metal). No key, runs offline.

Runs on your MacFreeSpeech-to-text

Ollama

Local language models + embeddings via a running Ollama daemon.

Runs on your MacFreeLLMEmbeddings

Parakeet

On-device speech-to-text via NVIDIA Parakeet (ONNX). No key, runs offline.

Runs on your MacFreeSpeech-to-text

Whisper

On-device speech-to-text via whisper.cpp. No key, runs offline.

Runs on your MacFreeSpeech-to-text

Your key · cloud

7

Anthropic

Claude models for agent and summarization

Your keyLLM

Deepgram

Streaming speech-to-text + Aura text-to-speech

MeteredSpeech-to-textText-to-speech

Google Gemini

Gemini models for chat, agents, and tools

Your keyLLM

Groq

Fast inference on open models

Your keyLLM

OpenAI

GPT models, Whisper transcription, and speech

Your keyLLMSpeech-to-textText-to-speech

OpenRouter

One key, 400+ chat models

Your keyLLM

Vercel AI Gateway

One key, hundreds of models, one bill

Your keyLLMSpeech-to-textEmbeddings

Coding agents

Drive Zeph over MCP

The coding agent you already use can control the device directly. Claude Code and Claude Desktop connect with guided setup; any other MCP client works today with the base zeph-device skill.

Works out of the box

Claude Code

Anthropic's coding agent, in your terminal.

Claude Desktop

Claude on your desktop, MCP-ready.

Supported

Codex

OpenAI's coding agent.

Cursor

The AI code editor.

Hermes Agent

Point it at Zeph's MCP server and the base skill drives the device.

OpenClaw

Connect it to Zeph's MCP server and the base skill drives the device.

Any other agent

Any MCP client works with the base zeph-device skill; connect it in the companion's MCP page.

On the glass

Apps run on the device

An agent talks to Zeph from your computer; an app runs on the device itself: a glanceable face for one AI tool you use, answered with a button or your voice. These are on the way.

A Perplexity face: 271 Pro searches left today, the last answer “Fed held rates at 4.25%” with 6 sources, and a hold-to-ask voice cue.

Perplexity

Ask by voice, hear the sourced answer, and keep an eye on your searches left.

A Notes face: a recording waveform at 04:12, the live transcript “…ship the calm hero first, then wire the footer to Kit”, and a summary-plus-action-items chip.

Notes

Record a thought or a meeting; it transcribes live and files an AI summary with the action items.

A Deep Research face: a progress ring at 12:04 researching, 58 sources and counting, reading an arXiv paper, with a read-rate sparkline.

Deep Research

Watch a research agent work through dozens of sources, and buzz you when the report lands.

A Claude Max face: a 5-hour cap at 40 percent and a weekly cap at 68 percent, weekly caps you first, burning about 14 percent a day.

Claude Max

Your 5-hour and weekly caps at a glance: which one stops you first.

An ElevenLabs face: a playback waveform speaking “Ship it Monday” in your voice, with 12.4k characters left this month.

ElevenLabs

Speak a line and hear it back in your own cloned voice, your character budget in view.

A Morning Brief face: “Your day, 3 things spoken aloud” over a 9:00 standup, three unread from Sam, and today’s focus.

Morning Brief

Your calendar, inbox and one focus for the day, gathered and read aloud.

Collection

Talk to your agent.

Hold the button, say the thing, let go. Some of these engines never leave your machine; the rest want a key you probably already have.

WhisperMLX WhisperParakeetDeepgramOpenAIVercel AI Gateway
A ZephOS push-to-talk screen: a pulsing dot above the words “Listening — release to send”.

Collection

Bring your own model.

You are already paying for one. Point the device at that key. It will never ask you for another, and it will never ask you which one to use twice.

AnthropicOpenAIGoogle GeminiGroqOpenRouterVercel AI Gateway
A ZephOS agent result screen: Approved, deploying build #2391, sent to agent.

Put an agent on your desk.