Short answer: Replit runs in a browser, so it has no dictation of its own. On a Mac, install Voice Keyboard Pro, hold your hotkey, and speak into any Replit field: the agent prompt box, the editor, a README, or the shell. On iPhone, use the Voice Keyboard Pro keyboard.
Replit changed what a coding session looks like. A large share of the day is no longer typing syntax, it is typing English: describing what you want built, explaining what went wrong, writing the README, leaving yourself a note about the deployment. That work is prose, and prose is the exact thing dictation is good at.
This guide covers how to dictate into Replit on Mac and iPhone, what is worth speaking and what is not, and how to describe a feature out loud so the result comes back closer to what you meant.
Why Replit has no dictation of its own
Replit is a web application. The editor, the agent panel, the shell, and the deployment settings all live inside a browser tab, and browser text areas do not come with speech recognition attached. That is why people search for a way to dictate in Replit and find nothing in the product settings: the feature was never a Replit feature to build.
The consequence is that dictation has to come from the operating system layer instead, and once it does, it works in every part of Replit at once. There is nothing to configure per field, and the same setup keeps working when you switch to the Replit desktop app, which is a wrapper around the same web experience.
Setting up dictation for Replit on a Mac
Voice Keyboard Pro sits in the menu bar. You hold a hotkey, speak, release, and the text appears at your cursor, whatever app or browser tab is in front of you. Setup takes about a minute:
- Download and install the Mac app.
- Launch it. It appears in the menu bar rather than the Dock.
- Grant Microphone access when macOS asks.
- Grant Accessibility access. This is the permission that lets text land at your cursor. Skip it and the app will hear you but nothing will appear.
- Choose a hotkey you can hold with one hand, ideally not one your browser or window manager already uses.
- Open Replit, click into the agent prompt box, hold the hotkey, and describe what you want.
Because the text is inserted at the cursor rather than through a separate transcript window, it behaves the same in every Replit surface: the agent chat, a Markdown file, a code comment, the shell prompt, a commit message field, or the description box on a deployment.
Dictating into Replit on iPhone
Replit on mobile is where dictation feels least optional, because typing a paragraph-long specification on a phone keyboard is miserable enough that most people just do not do it, and the prompt suffers.
Install the Voice Keyboard Pro keyboard from the App Store, add it in Settings, General, Keyboard, Keyboards, and enable Full Access so the microphone can work. Then in Replit, tap into the prompt field, switch keyboards with the globe key, and tap the microphone button. The same keyboard works in Safari if you use Replit on the web from your phone.
Two features earn their keep in this context. Voice Edit lets you speak a change rather than tapping through text with a fingertip, which matters when the fix is "change the second sentence to say Postgres instead of SQLite". And two-way translation across 24 languages means a developer working in a second language can think and dictate in their first one while the prompt goes out in English.
What is actually worth dictating
Not everything in a coding session should be spoken. The split is fairly clean.
Agent prompts
This is the highest-value use by a distance. Agent output quality tracks prompt quality, and typed prompts are systematically worse than spoken ones for a mundane reason: typing is slow, so people stop early. They write "add auth" and hope. When you speak, the marginal cost of another sentence is nothing, so you naturally include the constraints you would have skipped.
Code comments and docstrings
Explaining why a piece of code exists is prose work wedged inside a code file. It is also the work developers skip most often. Dictating comments removes the friction, and we covered the mechanics of it in how to dictate code comments.
README files and project docs
Setup instructions, environment variable descriptions, and the "what this project is" paragraph are all easier said than typed. A README you can speak in four minutes tends to get written. One that needs twenty minutes of typing tends not to.
Commit messages and issue text
If you push from Replit to a Git host, the message field and the pull request body are both plain prose. The same applies to bug reports you file against your own project, which we broke down in dictating bug reports.
What to keep typing
Actual code syntax is still faster on a keyboard. Nested brackets, operators, and precise indentation are not natural speech, and editor autocomplete is working against you the whole time. Credentials and API keys should never be dictated either, not because of any specific risk in the transcription path, but because saying a secret out loud in a shared space is its own problem.
How to speak a specification
The skill worth learning is describing a feature out loud in a way an agent can act on. A useful spoken prompt has four parts, and saying them in order takes under a minute.
Context. What the project is and which part you are touching. "This is an Express app with a Postgres database. I am working in the routes folder."
Goal. The outcome, described as behavior rather than implementation. "I want users to be able to reset a forgotten password by email."
Constraints. The things you would otherwise have to fix in a second round. "Use the mailer module that already exists. Do not add a new dependency. Tokens should expire after one hour."
Acceptance. How you will know it worked. "When I request a reset for an address that is not registered, the response should look identical to a successful request."
Spoken, that is roughly forty seconds. Typed, most people would have written the first two parts and left the rest to a follow-up round of corrections. The prompt is better because dictation made length cheap, not because you thought harder.
Handling technical vocabulary
Every dictation system struggles with the same things in a coding context: library names, invented product nouns, and identifier casing. There are two levers that fix most of it.
The first is Smart Vocabulary, a personal dictionary with replacement rules. Add the names of the libraries, services, and internal terms you say every day, and they stop being rewritten into ordinary English words. This is a one-time setup that pays back within a week.
The second is deciding where casing gets applied. Speaking identifiers in camelCase or snake_case is possible, and we wrote a full guide to dictating camelCase and snake_case. In an agent-driven workflow, though, the simpler move is often to say the name in plain words and let the agent write the identifier in whatever convention the codebase already uses.
Speak punctuation as you go: "comma", "period", "new paragraph". In Markdown files and prompts, a spoken "new paragraph" between sections gives the agent structure to work with instead of one dense block.
Troubleshooting dictation in Replit
Nothing appears when you speak
Check Accessibility permission in System Settings, Privacy and Security. It is the step people skip. Toggling the app off and back on in that list forces macOS to re-register it.
Text lands in the wrong pane
Replit's layout puts several focusable areas on one screen, so the words go wherever the cursor actually is. Click into the field, watch for the cursor, then hold the hotkey. If the browser window itself was not focused, the text may land in an entirely different app.
Autocomplete swallows the text in the editor
Inserting a burst of text into a code editor while a completion popup is open can produce mangled results, because the editor tries to interpret the insertion. Dictate prose into comments, Markdown files, and the agent panel rather than into the middle of a live code line, and press Escape to dismiss any popup first.
The hotkey does nothing in the browser
Browsers and extensions claim keyboard shortcuts aggressively, and they do it silently. Pick a different hotkey instead of hunting for the conflict.
The shell behaves oddly
Replit's shell is a terminal, and terminals treat pasted or inserted text differently from typed text. Dictating a command works, but read it before pressing Return. Our guide to voice typing in a terminal covers the habits that keep this safe.
The speed argument for developers
The average adult types around 40 words per minute. Strong professional typists reach 80 to 100. Ordinary speech runs 130 to 150 words per minute with no training at all.
For syntax that gap is irrelevant, because code is not written at prose speed anyway. For everything around the code it is decisive. A developer who writes an eight-hundred-word specification, a README, and a handful of issue comments in a day is producing text where speaking is two to three times faster, and where the typed version is usually shorter than it should be. Teams that push work through Git see the same effect on review comments, which is why we wrote a separate guide to dictating in GitHub.
In agent-driven development, the bottleneck moved from how fast you type code to how clearly you can describe what you want. Speaking is the faster interface for that.
A workflow that works
Try one session in this shape.
Open your Replit project and, before touching the agent, dictate a short plan into a scratch Markdown file: what you are building today, in what order, and what you are deliberately not doing. Speaking a plan takes two minutes and prevents the drift that eats agent sessions.
Then work in loops. Dictate the specification, read what the agent produced, and dictate the correction as a full sentence rather than a terse fragment. "The reset link is going to the wrong host because it is using the request header instead of the configured base URL" gets a better result than "wrong URL", and it costs you five extra seconds when spoken.
At the end of the session, dictate the commit message and a two-line note in the README about anything you changed in the environment. Both of those get skipped constantly, and both are cheap when you say them.
Common questions
Does this work in the Replit desktop app as well as the browser? Yes. Because the text is inserted at the system cursor rather than injected into a specific web page, it does not matter whether Replit is open in Safari, Chrome, Arc, or the desktop wrapper. The same hotkey works in all of them, and it works in your terminal and your notes app in the same session.
Do I need an internet connection? Yes. Voice Keyboard Pro's transcription engine runs in the cloud, which is what keeps it fast and accurate without loading up your machine. Since Replit itself requires a connection, this is rarely a real constraint.
Will it work with other browser-based editors? It works anywhere you can put a cursor, so the same setup covers other online IDEs, the GitHub web editor, and cloud notebooks. Nothing is configured per site.
Can I dictate long prompts without the app cutting me off mid-thought? The hold-to-talk model means the recording lasts exactly as long as you hold the key. There is no fixed session timer to race against, which is the usual reason people abandon dictation for long specifications.
Does it handle accents? Yes, and this is worth testing rather than assuming. Advanced AI transcription is considerably more robust to accent variation than the built-in dictation most people formed their opinion on years ago. The free tier is enough to check with your own voice on a real prompt.
Privacy and cost
Dictated audio is not stored. The server keeps only operational pings, with no audio and no transcript content, and your local history stays on your own machine. That matters more than usual for developers, because prompts often describe internal systems.
There is a free tier with daily limits, which is enough to test whether the habit sticks on a real project. Pro is $4.99 per month or $34.99 per year, and covers both the Mac app and the iPhone keyboard.
Replit removed most of the reason to type code by hand. Dictation removes most of the reason to type the English around it. Set it up free and speak your next agent prompt instead of typing four words and hoping.