Wispr Flow
Voice-to-text that actually works — and the secret weapon I didn’t know I needed.
What This Is
Wispr Flow is not a code editor. It's not a builder. It doesn't host anything. It does one thing: you hit a hotkey, you talk, and clean text appears in whatever app you're in. Cursor chat. Claude Code prompt. Slack. Email. Notion. SQL editor. Doesn't matter. Wherever the cursor is, that's where it writes.
I'll be honest — I rolled my eyes at this category for years. Voice typing was always the demo that worked great in the keynote and fell apart the second you tried it in a real meeting. Filler words everywhere. Punctuation a guess. Technical terms butchered. I'd type faster than I'd fix the transcript.
Wispr Flow is the first one I actually use every day. The AI cleanup layer is what makes it work. It strips the "ums" and the "you knows," it punctuates correctly, it knows that when you say "camel case get user profile" you mean getUserProfile, and it produces output that reads like you wrote it carefully — not like someone left a voice memo.
That's the whole pitch. And once you've got it, you can't go back. Investors apparently agree — Wispr closed a $260M Series B this year at a valuation around $2 billion. Voice input stopped being a niche while nobody was looking.
Why It Belongs In A Vibe Coding Guide
Because vibe coding is writing. You're describing what you want and an AI builds it. The bottleneck isn't the AI — it's how fast and how clearly you can get the description out of your head.
Typing is a tax. A small one, but you pay it on every prompt, every clarification, every "no, I meant the other one." Stack that over a day and it's an hour. Stack it over a year and it's the difference between shipping the thing and not.
Wispr Flow removes the tax. The first week I used it I noticed my prompts getting longer — because I'd describe things the way I'd describe them to a coworker, not the way I'd type them at 70wpm. And longer, more specific prompts produce dramatically better AI output. Compounding effect. Real one.
Who Should Use This
Anyone prompting AI tools more than an hour a day. Which, if you're reading this guide, is probably you.
Founders who think out loud. If you'd rather leave a voice memo than write an email, you already know.
Anyone with wrist or shoulder issues. Not marketed as accessibility tech but functionally it is. Your hands will thank you in ten years.
Heavy writers who happen to code. Docs, dev advocates, technical bloggers. Wispr Flow eats prose for breakfast.
What Actually Matters
The AI cleanup is the product. Raw transcription is a commodity. The thing Wispr Flow does that nobody else does as well: it makes your spoken English read like written English. That gap is the whole game.
It works everywhere. Hotkey. Talk. Text appears. In Cursor, in a terminal, in a GitHub issue, in a Discord DM. It doesn't care which app. No "open the dictation window then paste the result" garbage. It just types where you are.
Voice commands for editing. "New paragraph." "Delete that." "All caps." "Period." The learning curve is twenty minutes. Then you stop reaching for the keyboard.
It knows tech-speak. "Snake case get user by ID" → get_user_by_id. Edge cases trip it, but for common patterns it gets it right and you stop spelling things phonetically like an animal.
Privacy mode exists. Audio processing happens in the cloud. Wispr Flow has options to limit what gets sent. Read the privacy docs before you talk credentials at it. (You shouldn't be talking credentials at any voice tool, but I had to say it.)
Cost
Free tier: 2,000 words a week on desktop, unlimited on Android. Real evaluation amount. Light users may stay here forever.
Pro: $15/month, or $12/month billed annually. Where regular users land. Removes limits, full voice commands, latest models.
Teams ($10–12/user/month) and Enterprise plans for company-wide rollouts — SSO, SOC 2, the compliance alphabet.
If you prompt for a living, you'll pay this back in saved time on day one.
What It's Good At
Web apps: Four stars. Excellent for browser-based tools, AI chat sidebars, GitHub, anything text-field-shaped.
Mobile: Four stars. This used to be the weak spot — desktop only. Not anymore. There's an iOS keyboard and an Android app now, and Android dictation is unlimited even on the free tier. Voice-prompting from your phone is a real workflow.
Local apps, terminals, scripts: Five stars. This is where it sings. Pair it with Claude Code in a terminal — you'll never want to type a prompt again.
Hosting
N/A. It's an app that turns voice into text. Nothing to host.
The Story
I was on the phone with my wife while building a prototype for a client demo the next morning. I had headphones on. I was pacing the kitchen. The codebase needed a whole new feature — auth, a dashboard route, a couple of API endpoints, a deployment to Vercel.
Old me would have hung up, sat down, and typed for an hour.
What I actually did: stayed on the call, walked over to the laptop, hit the hotkey, and described the feature in two breaths while my wife was telling me about her day. "Add an admin route protected by Clerk auth, dashboard at /admin showing user signups for the last 30 days, single chart, use Recharts, get the data from Supabase, the table is called signups." Hit send. Cursor went to work. I went back to the conversation.
Forty minutes later I shipped the demo. My wife had no idea I'd built anything.
That's the unlock. Voice doesn't just save typing time — it lets you integrate building into the rest of your life in a way typing simply doesn't allow. I built a feature without breaking eye contact. That's wild. We don't have language for it yet.
Where It Bites
You need a decent mic. Built-in laptop mic works in a quiet room. Open office or coffee shop, you'll fight it. Twenty bucks on a USB mic solves this forever.
Talking to your computer feels weird at first. I'm not going to pretend otherwise. In a shared office you'll be self-conscious. Most people adapt within a week, some don't. If "don't" is you, headphones with a boom mic fix the visual half of the problem.
Cloud processing. Your voice goes to their servers. For prompts about features, totally fine. For anything sensitive, think before speaking. Same rule as any cloud voice tool.
Imperfect on weird names. Library names with no vowels, internal codenames, your CEO's last name — you'll correct it sometimes. Annoying, not a dealbreaker.
Gavin's Rule: Talk Like You're Briefing A Coworker
The biggest unlock from Wispr Flow isn't speed. It's prompt quality.
Voice naturally produces longer, more conversational descriptions than typing. Use that. When you hit the hotkey, give the AI the full brief — what the feature is, what it connects to, what the edge cases are, what it should feel like. The way you'd actually explain it to a person sitting next to you.
A fifteen-second spoken prompt with full context will out-perform a typed prompt that abbreviates every time. You'll do fewer revision rounds. The AI will get it right on the first try more often.
Don't try to "be efficient" by talking in fragments. The whole point is that the cost of saying more just dropped to zero. Use that.
Bottom Line
This is the tool I forget to mention because I forget I'm using it. That's the highest compliment I can give a piece of software. It disappears into the workflow.
If you spend more than two hours a day prompting AI, install the free tier this week. The habit forms in three days. After that, going back to typing prompts will feel like writing with a pencil after using a keyboard.
— Gavin