Lispr is a push-to-talk voice dictation and translation app for Mac and Windows. Hold a key, speak, release. Your words land wherever your cursor is. Speak in ~99 languages and switch mid-sentence. Hold the translation key instead and the translation lands, in any of 32 languages. Median latency 346 ms. The mic is off until you hold the key, and your audio is not retained on our servers. Free plan: 60 dictation minutes a week, no account. Pro is $79 a year for unlimited, or $7.99 a month.
Free
Launch Team / Built With
EneyProactive AI for Mac. Free test, 20,000 credits.
Promoted
superwhisper
Wispr Flow
Hey Product Hunt 👋
I'm Konstantin, co-founder of Codebridge, a software development company. Lispr's first user was me.
Why I built it
My workday is Claude Code sessions, client emails, Teams threads, and spec reviews: thousands of words typed across a dozen apps. Then I noticed that when I dictated instead of typing, I got several times more done. The effect was strongest with AI tools. When you talk to Claude or Cursor, you give whole paragraphs of context you'd never bother to type, and the answers get far better. Typing made me ration what I told the AI.
I wanted one tool that types wherever my cursor is: chat, email, code editor, browser. I tried what was on the market and kept hitting the same walls: multi-gigabyte model downloads, accounts, subscriptions, or latency that sent me back to the keyboard. We're a dev company, so we built our own.
The multilingual part is personal too. We're a Ukrainian company. Ukrainian inside the team, English with clients, and many of our people live abroad and run daily life in a third language. So translation got its own keys: you set two, and holding one along with the dictation key changes what happens when you let go. Release with just the dictation key and you get the transcript; release with a translation key held and the translation lands instead. When you drift between languages mid-sentence (we all do), Lispr follows. No setup, no mode switch.
What Lispr is
A free voice dictation and translation app for Mac and Windows. Hold the key, speak, release. Your words land in whatever app your cursor is in. Hold a translation key too, and on release the translation lands instead of the transcript.
Where it earns its keep:
Draft Slack messages and emails without touching the keyboard
Prompt Claude, ChatGPT, and Cursor by voice, with far richer context than you'd type
Write in Notion, Docs, anywhere text goes
Speak in ~99 languages, switch mid-sentence
Teach it your vocabulary, so client names and jargon come out spelled right
Dictate in one language, release, and it lands in another of 32, via two configurable per-language keys. No other Mac dictation tool has this.
Speed and footprint
Median latency is 346 ms from key-release to text on screen, measured server-side on live traffic. The whole app is a 3.67 MB download, with no model file and no GPU requirement. It runs on macOS 11 and later, including Intel Macs, and on Windows.
Privacy, the specifics
Your microphone is off until you hold the key.
Audio streams to a hosted Whisper large-v3-turbo model for transcription. Our servers don't store it, and no transcript content is logged anywhere. The inference provider holds audio up to 30 days only for abuse review, then deletes it.
Nothing trains on your voice, transcripts, or translations unless you opt in, and the opt-in is double-gated.
No account. Download, grant mic permission, start talking.
Is there a catch?
No. Lispr is free and the free tier stays. Codebridge is a profitable consulting company, and Lispr's architecture pays per call, so infrastructure costs scale with usage, not with always-on GPU capacity. It costs us very little to keep free. If we ever add a paid tier, it will be for heavy or team-scale use, never for everyday dictation.
For the PH community
We're reading and answering every comment today. What gets named in this thread will shape what we build next: iOS and Android are already on the list, and the requests here move up the queue.
What I'd love from you
Download it and tell me where it trips: lispr.ai
Which languages do you work in? We built this for multilingual days, and I'm curious how multilingual this community is.
Thanks to our early users in 29+ countries for finding the rough edges, and to @myroslav_budzanivskyi, our CTO, who took Lispr from first commit to a notarized public release in a single day, then shipped 67 releases in the three weeks after.
Konstantin
the no-account, stateless-relay answer to the audit question above was refreshingly honest, more teams would just say "we don't log anything" and leave it there. that raises a question about the vocabulary feature though - if it learns client names and jargon from my dictations over time, that's a profile of sorts even without an account. is that vocabulary list stored purely on-device, or does it live server-side somewhere tied to an install id, since "no persistent identity" and "the app remembers your jargon across sessions" seem like they need to be reconciled somehow
@galdayan Fair question, and thank you for reading the audit answer that closely. The vocabulary list is stored only on your device. There is no server-side copy and no install id it's tied to. So "remembers your jargon across sessions" means a local file on your Mac, not a profile on our side. That's how it reconciles with "no persistent identity": the persistence lives on your machine, nothing on our side persists between requests.
@konstantin_karpushin1 that's a clean answer, local file beats a policy promise every time. follow-up: if someone gets a new Mac or reinstalls, does that vocabulary just start over, or is there some manual export/import so people don't lose months of learned jargon
@galdayan Honest answer: today it starts over. The vocabulary is a local file and we haven't built export/import yet. You just moved it up the roadmap, a transfer option is going into the next release.
On how the list builds, it works both ways. Auto-grow: about once a day Lispr reviews your recent dictations, spots the words it got wrong (brands, surnames, client jargon) and adds the correct spellings itself. Say "Lispr" and the first transcription might come out "Lispro"; the next pass figures out what you meant and fixes the entry. Manual: you can add terms in advance and edit or delete anything the auto-grow learned. Both live in the same on-device list.
@konstantin_karpushin1 good to hear the transfer option is coming. on the auto-grow side: does the daily review apply corrections automatically once it's confident, or does it queue them for you to approve first? a one-off misheard word could get "corrected" into something wrong if it's fully automatic and you don't happen to catch it that day
Congrats for you! Wondering how it handles contexts like coding tools versus marketing or social writing, since those are pretty different voice-to-text use cases.
@crystalmei Thank you! I'm on the marketing side and use it all day, so here's the honest user answer.
Lispr doesn't change its behavior per app. It types what you said wherever your cursor is, the same way in VS Code and in a LinkedIn post. The difference is in how you use it. In coding tools I don't dictate syntax; I dictate prompts to AI assistants, commit messages, and comments, and that covers most of what a keyboard did for me there. Our CTO dictates code comments through it daily.
For marketing and social writing it's the main way I draft now: long text with full context comes out faster than I could type it.
Two things help across both contexts. It transcribes what you say without rewriting it, so your wording stays yours in both worlds. And there's a vocabulary feature, so product names and client names come out right instead of being "corrected" into dictionary words. It even learns new terms from your dictations.
If you try it in a specific tool and something feels off, tell me which one. The team ships fixes fast.
@thys_beesman In transcribing the lectures (online or offline) to avoid writing down the lecture notes manually.
@thys_beesman Three surprised us.
The biggest: talking to AI tools, vibe coding above all. When your reader is a model, spelling and paragraph breaks stop mattering; the model understands context. So people speak two minutes of context and constraints instead of typing three sentences, and the output gets better. Half our team dictates prompts now, including in the terminal. Lispr's part in this is speed and fidelity: it types your words as you said them, anywhere, fast. The app never rewrites you. The freedom to be loose comes from who's listening.
Second: thinking out loud. Jeremy Caplan of Wonder Tools listed Lispr last week as a free alternative to Wispr, under what he calls "bionic dictation": talking through an idea before the editing impulse kicks in. We built a typing tool and people use it as a thinking tool.
Third is on this page: Katrya's review above, dictated with Lispr 😄
Nice one. I added voice input to my own app recently and the hard part wasn't transcription at all, it was getting the mic to behave the same on every device. Does hold to talk work in any text field system wide, or is it per app? Congrats on the launch, hope it goes well today!
@henry_s_jung Thanks Henry! System-wide: anywhere you can put a cursor, you hold the key and talk. Per field, not per app, so it works in random Electron apps, browser inputs, terminals.
And agreed on transcription being the easy part. For us the hard bits were speed and consistency: we start streaming audio to the model before you finish speaking, run a quality check before the text lands in the field, and route recognition to the nearest region (Asia to Asian servers, US to American) so latency stays flat wherever you are. Mic behavior across devices is its own separate pain, you have my sympathy there.
Congrats on shipping voice input yourself, and thanks for the kind words!
As an indie dev who basically lives in AI coding sessions, the "typing made me ration what I told the AI" point hits hard — I under-explain to Claude/Cursor constantly just to save keystrokes. Sub-350ms and it lands wherever the cursor is could genuinely fix that. Curious how it handles code/technical terms and jargon vs everyday speech?
@lennoxbeflying You just described the exact moment Lispr clicked for us: typed prompts shrink because typing is work, spoken ones carry the full context, and the agent stops guessing. On jargon: nobody dictates syntax, the win is the paragraph around the code, what you tried, what broke, which constraints matter. That's everyday speech, and it handles it out of the box. For your own terms there's a Vocabulary feature: teach it your project names, APIs, and brands, it applies the right spelling while transcribing, and once a day it scans recent dictations to pick up new terms on its own. Works cross-language too: say a term in your native language, the correct Latin spelling gets typed.
I was curious about the custom vocabulary feature. Does Lispr learn technical terms over time, or do you need to add them manually?
@vaishnavi_makode Both! There's an Auto-grow option: once a day Lispr scans your recent dictations for new words and learns the right spellings on its own. You can also add terms manually, and everything in the list is editable. The whole vocabulary stays on your device, no server copy.