Dictate anywhere.
Your AI talks back.

Polished text, searchable meeting notes, and voice control of your Mac. Talk to Codex, Claude, or local AI. Speech recognition stays on your Mac.

A productivity tool.
Not a pretend human.

Your AI will cheer for your worst ideas. Apologize for things it didn't do. Pretend to care. We've all learned to ignore it. We're building something better, a tool that helps you get stuff done, stays honest about what it is, and keeps the quirks in on purpose.

Your voice. Your character. Your choice.

Speak naturally.
Get text that looks written.

on-device
BEFOREwhat the mic heard
so um import the num pie library and create a data frame from the csv
AFTERwhat Super Voice Mode types
import numpy as np
import pandas as pd

df = pd.read_csv("file.csv")
← drag or wait

One key. Any app.

Hold Right Option anywhere on your Mac. Text lands at your cursor. Slack, Gmail, Cursor, Xcode, Terminal, browsers, Obsidian. If you can type there, you can dictate there.

Slack Gmail Xcode Cursor Obsidian Terminal Browsers
⌥
hold Right Option
→
speak
→
cursor
paste

Dictation

Hold Right Option, speak, release. Clean text appears wherever your cursor is.

AI Writing

Ask what you want written. With highlighted text, Right Option can talk to AI instead of replacing. Free.

Live Transcription

Double-tap the dictation key for longer hands-free sessions with live subtitles. Pro.

Calls and thoughts that become memory.

Capture the room, the call, or a solo thinking session. Super Voice Mode saves notes, transcripts, decisions, and ideas to a local markdown vault your assistant can search later.

Explore Meetings + Notes → No bot joins your call

When there is no tool,
use the interface.

Speak the outcome and Super Voice Mode chooses the best route: a structured tool when one exists, the app interface when it does not, and visual control only as a last resort.

Explore Computer Use →

Structured tools first

Use APIs, MCP servers, and native agent tools whenever they fit the task.

Understand the interface

Work through accessible buttons, menus, forms, and fields in real apps.

See when needed

Fall back to the visible screen only when the interface has no better route.

Your agents. Their native tools.
Voice becomes the control layer.

For bigger jobs, Super Voice Mode connects the workflows above to the agents you already use. Speak the outcome, follow the work, and continue the same session without rebuilding context.

Works with
Codex Claude OpenCode Ollama Hermes
Ready
Ready
You
⌥⇧
hold ⌥⇧ and speak…
Assistant
·

A canvas for the conversation.
Beyond chat.

Voice forgets. Chat is slow. Slate is the missing middle: a shared document that takes shape while you talk. Press ⌘⇧S, speak, watch it write.

Explore Slate → In active development
Slate document taking shape beside a voice conversation

Select. Speak. Transform.

live on any text field
draft.md
## Notes · draft
 
A few things from this morning's sync:
 
we should probably think about maybe adding a feature that lets users export their data in some kind of format
 
Follow-ups: schedule review, loop in design.
⇧⌥
IDLE→Make this better
Rewrite selection
or just speak a command
press⇧⌥anywhere

Meet Iris.
An assistant with a name.

shipping today
🌟
Iris
your AI

"Copy that, one moment."

direct dry humor volunteers context calls you "Boss"

Every assistant out there is a faceless chatbot. Iris is your AI: a voice, a name, opinions, and a personality that shows up the same way every time you hit the hotkey.

Iris is the anchor. Stack a persona on top to shift the voice without losing the character.

See Iris with personas ↓

Iris, filtered.
Same assistant, different voice.

30+ bundled, mix and match
Character
shipping
+
Persona
youGive me a status update on the deploy.
Iris · Drill Sergeant

Switch characters up top. Swap personas to change how they sound.

Zero bytes to the cloud.
Not a feature. The foundation.

1,247
dictation sessions today
0
bytes sent to any server
100%
processed on your machine

Speech-to-text runs on your machine. Correction runs locally by default, with optional Claude Code processing. When you use the assistant, your AI service handles the request. We never see it. Audio never leaves the machine.

Clear boundaries. No guessing.

local-first by design
Dictation audio
Stays on your Mac

Captured, transcribed, and discarded locally.

AI correction
Local by default

Use local MLX or choose Claude Code to process text with your account.

Assistant requests
Your chosen backend

Sent only when you ask Claude, Codex, OpenCode, Ollama, Hermes, or a local model to act.

Super Voice Mode servers
Never receive content

No account, no telemetry, no hosted transcript store.

Hold & speak

Right Option anywhere on your Mac. Text lands at your cursor.

On-device transcription

State-of-the-art speech-to-text. Runs privately on your machine.

Meetings + notes

Capture calls, spoken thoughts, transcripts, and action items in a local markdown vault.

Bring your own AI

Route to any advanced model. Cloud or local. Your keys, your data.

Character voices

Spoken assistant responses, on-device. Hands-free.

30+ personas

Stack up to two. Some useful, most hilarious.

Hold a key. Ship faster.

No account. No telemetry. Requires macOS 15 or later.