Privacy-first · Stored on your device · Bring your own API key

Turn speech into text
you can use right away.

Hold a shortcut and speak. Let go, and it transcribes, removes filler words, smooths the sentences, and drops the polished result right at your cursor. Your recordings, data and API credentials all stay in your own hands.

macOS 12+ Signed · Notarized One-time purchase · No subscription
Typefree · Cleaning up a dictation
Listening…

So So I looked at the the flow, and I think like the overall icons and layout, and the text too, I think we should optimize it — consider centering you know, consider… arranging it nicer, otherwise the the background feels empty, and then the text is just too cramped, and it sits a bit too high. The layout just doesn't feel great I guess.

Cleaned-up output
I looked at the flow, and I think the overall icons, layout and text all need some optimization. Main things to consider: 1. Center it so the arrangement looks nicer; 2. Right now the background feels empty while the text is too cramped; 3. The position is a bit too high. Overall, the layout doesn't feel great.
Cloud recognition AI clean-up Auto paragraphs
Why Typefree

As capable as subscription tools,
but your data and credentials stay yours.

Honest & transparent

Recording, recognition, clean-up, paste-back — we tell you exactly where every step goes. No overclaiming, nothing hidden.

Local-first data

Recordings, history and API credentials are stored on your own device by default, never passing through any server of ours.

Bring your own cloud (BYOK)

Use your own Volcengine / Qwen / Zhipu accounts. Switch recognition and clean-up models freely, never locked to one provider — and we never bill you.

One stable codebase

Every update is thoroughly tested before release — dependable enough to live in your daily workflow.

Real results

Not a raw transcript —
the sentence you actually meant to send.

These four are real output from the real product pipeline — casual spoken takes on the left, Typefree's automatic clean-up on the right: fillers removed, slips fixed, sentences smoothed, points split out, numbers and terms normalized.

Self-correction · Auto-fix
Said casually

I can come Friday afternoon, no wait, not Friday, it's Thursday afternoon, anytime after 3 — can you set it up?

Cleaned up

Thursday afternoon, anytime after 3 works for me — can you set it up?

Journal · Quick notes
Said casually

So I've actually realized lately that working from home isn't that efficient for me, because, um, the environment's just too familiar and relaxed, and I keep getting, like, distracted while working—fiddling with this, fiddling with that, going to the bathroom, getting some water, just easily distracted, so my efficiency's pretty low. Because I noticed when I'm at a cafe outside, I'm way more focused, way more efficient, get more done too. So I'm planning to, that is, just go work at a cafe from now on, should be better. And my upstairs neighbor happens to be renovating lately, so it's kinda noisy at home.

Cleaned up

I've realized lately that working from home isn't very efficient for me. The environment is too familiar and relaxed, so I keep getting distracted—fiddling with one thing, then another, getting up for the bathroom or a glass of water—which drags down my productivity. At a cafe, by contrast, I'm far more focused and get a lot more done. So I'm planning to work from a cafe from now on. It should help—especially since my upstairs neighbor has been renovating, so home's been pretty noisy.

Coding · Debug instructions
Said casually

Can you fix this login page — uh no, don't overhaul it, first look at why the button keeps spinning. Then if the frontend state isn't resetting, fix the frontend; if the API isn't responding, add the error message. Also don't touch the signup page, and run the tests at the end.

Cleaned up

Can you fix this login page. Don't overhaul it — first look at why the button keeps spinning. 1. If the frontend state isn't resetting, fix the frontend; 2. If the API isn't responding, add the error message. Also, don't touch the signup page, and run the tests at the end.

Numbers · Currency formatting
Said casually

I worked out the quote — the basic plan is around 3,600 a year, the pro plan about 9,800 a year, and buying three years should be about 20% cheaper.

Cleaned up

I worked out the quote: 1. Basic plan: about ¥3,600 a year; 2. Pro plan: about ¥9,800 a year. Buying three years should be about 20% cheaper.

Speech clean-up Auto paragraphs Bullet points Slip correction Number formatting Term correction Punctuation fixes
How it works

Four steps, from a sentence to text on screen.

The whole thing usually takes a second or two, and every step is visible and traceable.

1

Record locally

Hold your chosen shortcut to capture audio; release to stop and processing begins immediately.

2

Cloud recognition

Calls the ASR provider you configured to turn speech into text, injecting your hot words.

3

AI clean-up

Longer passages go to the LLM you configured to remove fillers, smooth sentences and tidy formatting.

4

Term fix · Paste

A final pass applies your glossary, then the polished text is pasted straight at your cursor.

Typefree · Today
5,042
Chars today
38
Sessions today
1.9s
Avg. latency
216
Total entries
RecentAll entries →
Um let's not ship this version yet, you know let me double-check the data first.
Let's not ship this version yet — let me double-check the data first.
Write me a prompt, like to turn meeting notes into a to-do list.
Write me a prompt: turn meeting notes into a to-do list.
Thursday 10am sync with claude on the cursor plan.
Thursday 10:00 AM — sync with Claude on the Cursor plan.
Volcengine · Configured Qwen · Configured History kept locally
Security

Where your words go,
spelled out clearly.

Your recordings, history and API keys live on your own device; requests go straight to the providers you configured, never through us.

Local recording

Audio is captured on your device and only enters processing after you let go.

Straight to your provider

Requests go directly to the ASR and LLM you configured — never through a server of ours.

Credentials stay local

API keys, history and original audio are kept on your own device first.

Local paste-back

After a glossary pass, the result is pasted straight back at your cursor.

Advanced

Plenty of room for those who like to tweak.

Beginners can run on defaults; power users can swap models, tune the glossary and review history to fit it precisely into their workflow.

Switch models freely

Recognition uses Volcengine BigASR; for clean-up, switch anytime between Qwen, Zhipu GLM and Doubao, choosing by cost and quality.

Qwen Zhipu GLM Doubao No polish · Raw

Hot words & glossary

Add easily-misheard product names, people and English tool names to your glossary, with optional weights. Both recognition and clean-up will prefer them.

Cursorweight 10
Typefreeweight 8
klawd → Claudeterm fix

Searchable, replayable history

Every entry is stored locally with the raw transcript, the cleaned result, timing and usage. Swap a model or add new hot words and re-process the same take.

Keep original audio

Optionally keep each original recording (macOS) to replay and verify. Set retention yourself — 1 day, 1 week, 1 month or forever.

Pricing

Three ways to use it — pick what fits.

New users get a free 7-day trial, on us; after that, use your own key for free forever, or buy once for unlimited input — never a subscription.

New users
Free· 7-day trial

Download and go — zero setup, for new users.

  • 8,000 characters total over 7 days
  • Recognition + AI clean-up, on us
  • No API key needed — just press and speak
  • After 7 days, switch to either option below
Download free for Mac
LifetimeYour own key
$29$6958% OFFone-time · forever

Unlock unlimited daily input. Pay once, use forever — not a subscription.

  • Unlimited daily input, no character cap
  • Still uses your own key, nothing relayed
  • One license key activates 1 Mac
  • Full refund within 30 days if unsatisfied
Buy Lifetime · $29
Free forever
Free· no time limit

Add your own API key and use it free, long-term.

  • 1,500 characters of input per day
  • Full recognition + AI clean-up + term correction
  • Costs run on your own key — we don't relay or bill
  • Switch recognition / clean-up providers freely
Download free for Mac

Pay by card or WeChat Pay — securely processed by Paddle.
Don't see WeChat Pay? Set "Country" to China on the payment form.
After paying, don't worry if the page doesn't change or returns to the homepage — that's normal; your code is emailed within ~5 minutes (check spam). WeChat confirmation can take 1–10 minutes; your license code is emailed automatically, so there's no need to pay twice.

Get started

Three steps — no complexity dumped on first-time users.

Beginners just install and grant permissions; power users can go further with providers, hot words and glossary.

STEP 01

Download & install

Download the signed, notarized DMG and drag it into Applications. New users get a free 7-day trial with zero setup; after that, 1,500 characters a day for free or buy for unlimited.

STEP 02

Configure provider credentials

Enter your recognition and clean-up API keys — you obtain them yourself from Volcengine and Qwen; Typefree never relays or bills. First time? Follow the step-by-step guide and you're done in 5 minutes.

STEP 03

Grant access, then speak

Grant microphone and accessibility permissions, hold the shortcut in any field, speak and let go — the text lands at your cursor.

FAQ

Answering the things people most often misread.

How is this different from the system dictation?

System dictation is closer to word-for-word transcription — it types out every “um” and “you know.” Typefree cleans the language after recognition — removing fillers, smoothing sentences, normalizing numbers and punctuation — so the output is ready to send or use as a prompt.

Is it fully offline?

No. The main path calls the cloud recognition and LLM services you configured. The emphasis is local-first and never routing through a server of ours — not “works without internet.”

Which providers and models are supported?

Recognition uses Volcengine BigASR; clean-up supports Qwen, Zhipu GLM and Doubao, switchable anytime — or you can choose “no polish, output raw text.”

How does pricing work? Can I try it free first?

Yes. New users get a free 7-day trial — zero setup, on us, 8,000 characters total over 7 days (up to 5,000/day); after that, add your own key and keep 1,500 characters a day for free, forever; for unlimited daily input, buy once for $29 — forever, not a subscription. Full refund within 30 days if you're not satisfied.

Can people who aren't great with APIs use it?

Right now it suits users willing to set up provider credentials. The guide has sign-up links and instructions (a 5-minute walkthrough); if you just want to see results, check the demo on this page.

Are my recordings and data safe?

Recordings, history and API credentials are stored on your own device by default (on macOS, ~/.config/voicepolish/). Requests only go to the providers you configured, never through us. You set how long history is kept.

Make voice input a daily tool you can trust.

For people who care about where their data goes, want to buy once, and don't want to be locked into a subscription. Start free, upgrade when you need more.