Free local voice dictation app

Talk easily.
Local voice dictation that does the typing.

Press a key, speak, and clean text lands wherever your cursor is. A fast local transcriber with optional AI polish, and it all runs on your machine.

100% on-device No account Windows & Linux
Works everywhere you write

Dictate into any app, editor or terminal. It lands at your cursor.

A global hotkey fires in whatever has focus. No plugins, no integrations, no copy and paste.

VS CodeCursorTerminalVimZedWarpSlackDiscordGmailNotionLinearGitHubObsidianFigmaDocsChatGPTClaudeRaycast
Typing
~45 wpm
Talking
~200 wpm
about 3x faster than the keyboard
The transform

Say it messy. Get it clean.

Tokezly doesn't just transcribe. It writes down what you meant: fillers and repeats gone, self-corrections applied, grammar fixed. Want it raw instead? Flip the switch off.

Literal · AI off

Verbatim

“um so the the websocket reconnect it doesn't back off it just hammers the endpoint”
you get
um so the the websocket reconnect it doesn't back off it just hammers the endpoint
Default · AI on

Clean

“um so the the websocket reconnect it doesn't back off it just hammers the endpoint”
you get
The WebSocket reconnect doesn't back off. It just hammers the endpoint.
On command · /

Prompt

“…it just hammers the endpoint” /commit
you get
fix(ws): add backoff to reconnect loop
What “Clean” does

It edits while you talk.

When AI rewrite is on, every dictation gets tidied before it lands. No proofreading pass, no cleanup.

  • Drops filler words

    “um”, “uh”, “you know” and the rest never make it to the page.

  • Cuts repetition

    Repeated words and false starts collapse into one clean line.

  • Fixes self-corrections

    Change your mind mid-sentence and it keeps only what you meant.

  • Formats lists & code

    Spoken steps become bullets; commits, diffs and JSON come out structured.

The actual app

It lives where you work.

No mock-ups, no marketing render. This is the real Tokezly interface, running live. Press a key in any app and it appears, listens, and lands clean text at your cursor.

Real-time

Streams as you speak.

Words land at your cursor the moment you say them, with live latency and words-per-minute right in the popup. No window to manage, just a capsule that appears and gets out of the way.

Pastes straight into any editor, terminal or chat.
Voice-activity detection trims the silence live.
Tens of milliseconds, fully on-device.
Swappable engines

Pick the model for the job.

Parakeet for speed, Whisper for accuracy, Moonshine and Canary when you want tiny. Every transcription engine is local. Switch per task from one panel.

Local-first ASR: audio never leaves the device.
Real accuracy & speed meters per model.
Download only the weights you actually use.
AI rewrite · optional

Cleanup that never leaves your machine.

The rewrite layer is one toggle. Off, Tokezly is a plain transcriber. On, Gemma 2 2B polishes every dictation on-device. Cloud models are strictly opt-in, gated behind an explicit switch.

Fully local by default: Gemma 2 2B, on-device.
Temperature, length, streaming, preview. All yours.
Cloud refinement stays off until you turn it on.
Zero-click flow

Three keys to clean text.

1

Press the key

A global shortcut summons a tiny capsule, wherever you are, in any app. No window switch, no focus steal.

2

Just talk

Speak naturally. A live waveform shows it's listening; voice-activity detection trims the silence as you go.

3

It's typed

The text streams in, gets cleaned (if AI is on), and pastes at your cursor. A quick toast, then it's gone.

Private & offline by default

Your voice stays yours.

Tokezly is built local-first: the models live on your disk and the work happens on your hardware. Nothing is uploaded, nothing is logged to a server, no account required.

Audio & text never leave the device
No telemetry, no cloud, no sign-in
Cloud models are strictly opt-in
Engines

Local models, your pick.

Parakeet V3ASR · fast · default
WhisperASR · accuracy
Gemma 2 2Brewrite · on-device
Cloudopt-in · API key / SSO
Get Tokezly

Talk easily. Start now.

Free to download. Models pull on first run. No account, ever.

v0.1.15 · Windows & Linux (x64) · models downloaded on first launch