Self-hosted · Private · Enterprise-grade

Dictate anywhere.
Never leave your perimeter.

Turn speech into polished writing in every app. Local Mode for individual. Self-hosted cleanup LLM for Enterprises.
Zero third-party APIs. Ever.

Download for Mac

FREE · NO ACCOUNT · 21 MB · OFFLINE-READY

Slack — #product

REC

Without CursorTalk~40 WPM

Type a message…

With CursorTalk~150 WPM

Type a message…

WORKS IN EVERY APP

Gmail

Slack

Notion

VS Code

Cursor

Teams

Figma

Outlook

Google Docs

Apple Notes

iMessage

Grammarly

Terminal

Obsidian

Discord

Linear

Jira

Zoom

Salesforce

ChatGPT

X (Twitter)

Gmail

Slack

Notion

VS Code

Cursor

Teams

Figma

Outlook

Google Docs

Apple Notes

iMessage

Grammarly

Terminal

Obsidian

Discord

Linear

Jira

Zoom

Salesforce

ChatGPT

X (Twitter)

FEATURES

Everything you need. Nothing you shouldn't trust.

Enterprise-grade dictation that keeps privacy as an architecture decision - not a policy checkbox.

3× FASTER INPUT

Speak at 150 WPM.

Your mouth is 3× faster than your fingers. CursorTalk bridges the gap with real-time ASR and LLM cleanup.

150

GLOBAL HOTKEY

One shortcut. Every app.

Customizable. Works in native apps, browsers, terminals - anywhere text goes.

⌘⇧Space

OFFLINE

Works on a plane.

Whisper runs locally. Cleanup syncs when back online.

HOW IT WORKS

Hotkey to polished text in under 2s.

Press hotkey & speak

LOCAL

Hit ⌘⇧Space anywhere. Mic captures audio entirely on your machine.

Whisper transcribes locally

~0.9s · LOCAL

distil-whisper-large-v3 runs via whisper.cpp. Audio stays in RAM - never written to disk, never uploaded.

~500 bytes sent to your VPC

TLS 1.3

Only plain text. No audio, no voice biometrics. Sent over HTTPS to your org's own AWS endpoint.

Cleaned text pasted at cursor

~285ms · YOUR AWS

Fine-tuned Qwen 2.5 14B removes filler, fixes grammar, adds punctuation. Inserted directly where your cursor is.

PRIVACY BY ARCHITECTURE

Your voice stays yours. By design.

We engineered the pipeline so raw audio physically cannot leave the device. Not a policy. An architecture constraint.

Audio never uploaded

Whisper runs on-device. Raw waveform never touches a network socket.

Zero third-party APIs

No OpenAI. No Anthropic. No Cerebras. Your VPC only.

Zero persistence

Processed in RAM and discarded. Nothing logged or trained on.

DATA COMPARISON

Cloud dictation~320 KB audio

→ OpenAI → Anthropic → Cerebras → Baseten

CursorTalk~500 bytes text

→ Your VPC only

640×

less data on the wire

DOWNLOAD

Stop typing. Start talking.

Free for individuals. No account required. Works offline after first launch.

macOS

Universal · Apple Silicon + Intel

Version1.0.0

Size21 MB

RequiresmacOS 12+

30-second installWorks offlineCode-signedAuto-updates

PRICING

Free to start. Sensible at scale.

No per-word billing. No API pass-through. You pay AWS; we hand you the rest.

Individual

FOREVER FREE

$0/mo

Download free

✓ Unlimited local transcription✓ 500 cleanups / month✓ All apps · all languages✓ Personal dictionary

Frequently asked questions

No. Whisper runs locally via whisper.cpp. Only the plain-text transcript (~500 bytes) is sent over HTTPS and in the Enterprise plan, that server lives in your own AWS VPC.

Wispr sends raw audio to OpenAI, Anthropic, Cerebras, and Baseten. We send zero audio to zero third parties. Whisper runs on-device; the cleanup LLM runs on infrastructure you own.

Mac with Apple Silicon (M1+) or Intel with 8GB+ RAM. macOS 12 (Monterey) or later.

~1.3-2.0s end-to-end after you stop speaking. Cloud tools are faster (~0.7s) because they skip on-device ASR - but they send your audio to 4+ third parties.

Yes - the Enterprise plan includes a CloudFormation template, model weights, Docker image, and IaC. Deploy in ~10 minutes. You own it with zero vendor dependency.

Never. Transcripts are processed in RAM and discarded. Nothing logged, stored, or used for training under any plan.

Dictate anywhere.Never leave your perimeter.