PricingRegisterSign InDownload for macOS
PRIVACY-FIRST · ON-DEVICE

Your machine. Your model.

Hold one key and dictate into any app. Your voice stays on your Mac, your model never changes unless you choose, and text lands in under 200ms.

Free download for macOS 14+. Apple Silicon recommended.

Windows / Linux coming soon.

PRIVACY
100%on-device
MODEL
pinnedby you
LATENCY
<200ms
  • · Your voice never leaves your Mac
  • · You choose the model. Nothing swaps silently
  • · Sub-200ms in any app
RECPUSH-TO-TALK
--:--:--Z
127MS · LIVE
END-TO-END
INPUT · MIC16 kHz
0s1s2s3s4s
DICTATION · ANY APP● LIVE
Holdspeak · release →
NO CLOUD · YOUR MODELSUB-200MS
VX-MONITOR / 01

Latency

You'll feel it.

Local inference means no cloud round-trips and no silent model swaps. Speak and see text appear in under 200ms. Quiet, smooth, predictable.

The Science Behind Instant Response

Why Speed Feels Different

40+ years of research reveals a critical threshold: Doherty & Thadani (1982) showed responses under 400ms create “addictive” user experiences. Because transcription runs on your Mac, there's no network wait and no server deciding which model you get. CodeVox operates at under 200ms, keeping you in the flow.

0ms

CodeVox Latency

0ms

Flow Threshold

The Research

Decades of human-computer interaction research establish clear perceptual thresholds.

1968|Miller

Response Time Foundations

Established that human-computer interactions function like conversations. Delays break the conversational rhythm just as awkward pauses do between people.

1982|Doherty & Thadani

The 400ms Productivity Cliff

IBM research discovered that sub-400ms responses create "addictive" experiences. Productivity increases exponentially as response time decreases.

1993|Nielsen

Three Fundamental Thresholds

100ms feels instantaneous, 1s maintains flow, 10s loses attention. These thresholds remain foundational to UX design.

2015|Google RAIL

Modern Validation

Response under 100ms, animations at 60fps, idle in 50ms chunks, load under 1s. The RAIL model validates classic research.

The Magic Numbers

Each threshold represents a perceptual boundary in human cognition.

100ms
200ms
400ms
1000ms
2000ms
100ms
Instantaneous

User feels they directly caused the outcome

200ms
Good Response

CodeVox operates here

CodeVox Zone
400ms
Flow Maintained

Maximum for "addictive" experience

1000ms
Flow Strained

User notices delay, thought continues

2000ms
Flow Broken

Concentration breaks, productivity drops

Flow State Explained

What is Cognitive Flow?

Complete absorption in a task. Intense concentration, loss of self-consciousness, distorted time perception, and peak performance.

How Latency Destroys Flow

When latency exceeds perceptual thresholds, the brain shifts from “doing” to “waiting”, and the flow state collapses.

The Direct Manipulation Illusion

Sub-100ms responses create the illusion of directly controlling output. Your voice becomes an extension of your thoughts.

“When a computer and its users interact at a pace that ensures neither has to wait on the other, productivity soars, the cost of work done on the computer tumbles, employees get more satisfaction from their work, and quality tends to improve.”
Doherty & ThadaniIBM Systems Journal, 1982

How We Compare

See where different voice-to-text solutions fall on the latency spectrum.

CodeVox
<200ms400ms threshold
Dragon
300-1000ms400ms threshold
Cloud Dictation
500-2000ms400ms threshold
Whisper (Local)
1000-5000ms400ms threshold

The white line marks the 400ms Doherty Threshold, the maximum latency for sustained flow state.

Setup

Get started in minutes.

01

Download

Grab the DMG — free, no account needed. Drag CodeVox into Applications and launch it.

02

Allow mic

Grant microphone access once. Audio is processed on-device. We don't store it.

03

Start talking

Hold the hotkey, speak, release — the text lands in whatever app you were in.

Start talking.

Free download. macOS 14+. Apple Silicon recommended.

Windows / Linux coming soon.