Voice typing has one real, growing use case for developers in 2026: dictating the natural-language text that surrounds code — prompts to AI coding assistants, commit messages, PR descriptions, code review comments, and documentation. It does not replace typing actual source code character-by-character, and any tool that claims otherwise is overselling. Here is exactly where the line sits, and how Hovor handles the parts that do work.
Describing a change to Claude Code, Cursor, or a similar assistant is writing, not typing code: you compose full sentences ("refactor the auth middleware to check the token expiry before hitting the database, add a test for the expired-token case"), exactly the long-form natural language dictation handles well. Speaking the prompt keeps your hands off the keyboard between edits.
For this use case, use Hovor's default cleanup tone rather than the coding tone. The cleanup tone removes filler words and fixes grammar and punctuation without altering your meaning, which is what an assistant expects as input. Trigger the global hotkey with your cursor in the assistant's chat box, speak the prompt as a normal sentence or two, and the cleaned text lands directly in the input field.
Commit messages, pull request descriptions, and code review comments are short, structured natural-language writing — exactly the profile dictation handles well, and a task most developers do dozens of times a day without ever touching a specialized tool. Speaking "fix race condition in the websocket reconnect handler when the client disconnects mid-handshake" into a commit message field produces the same result as typing it, once the cleanup pass removes filler words and fixes punctuation.
Hovor's formal tone-style behavior (structuring parallel items into numbered or bulleted lists when your dictation contains a clear enumeration) is useful specifically for PR descriptions with a "Changes:" section followed by several bullet points, and for review comments that list multiple issues in one pass.
Dictating a full function, class, or config file character-by-character does not work well with any voice tool, Hovor included. Source code is dense with exact punctuation, indentation, casing, and symbol placement that speech is a poor match for: saying every brace, bracket, colon, and underscore aloud is slower and more error-prone than typing it. This is a structural limit of voice as an input method for syntax-heavy text, not a Hovor-specific gap.
Hovor's coding tone-style is scoped narrowly around this limit rather than pretending it does not exist. It handles two specific cases: converting a short, fully spoken statement of code syntax ("select star from users where name equals Kyrylo") into the matching real syntax (SELECT * FROM users WHERE name = 'Kyrylo';), and generating a small function or snippet from a natural-language description ("create a Swift function that takes a user name and returns a greeting"). Output is formatted per the detected language's standard formatter — Prettier for JS/TS/JSON/HTML/CSS, Black for Python, gofmt for Go, rustfmt for Rust, swift-format for Swift, ktlint for Kotlin, google-java-format for Java, pgFormatter for SQL, shfmt for Bash — with string literals and comments preserved verbatim. It is not designed for long files, multi-file edits, or precise verbatim dictation of complex expressions.
Mechanically, the coding tone-style is sent from the client as a custom prompt rather than resolved from the server's built-in tone list, and the server always wraps it in the same safety scaffolding (language and dictionary handling) applied to every other tone, so it behaves predictably alongside your other settings rather than as a separate, unconstrained mode.
| Task | Good fit for dictation |
|---|---|
| Prompts to AI coding assistants (Claude Code, Cursor, etc.) | Yes — cleanup tone |
| Commit messages | Yes — cleanup or formal tone |
| PR descriptions with a bulleted "Changes" list | Yes — formal tone (auto-lists enumerations) |
| Code review comments | Yes — cleanup tone |
| Technical documentation, design docs | Yes — cleanup or formal tone |
| Slack / chat messages about your stack | Yes — cleanup tone, dictionary handles jargon |
| Short spoken SQL / API-shaped statements ("select star from...") | Yes, within limits — coding tone |
| Generating a small function from a description | Yes, within limits — coding tone |
| Writing a full source file character-by-character | No |
| Precise, punctuation-dense expressions (regex, nested generics) | No |
| Multi-file refactors | No |
Speech recognition models mishear rare technical terms because they are trained on general text where common words dominate: "Kubernetes" comes out as "cube ornetes," "PostgreSQL" as "post-gres-sequel," "OAuth" as "oh auth." Hovor's custom dictionary fixes this with mapping rules — list the variant mishearings, then the correct form, for example kubernetes|cube ornetes|cube ornate is → Kubernetes or oauth|oh auth|o-auth → OAuth.
The replacement runs on the raw transcription first. Then the same dictionary entries are passed to the AI cleanup pass as vocabulary context, so the model is aware of your preferred spellings and applies them consistently even inside sentences it has not seen that term in before, rather than only as a mechanical find-and-replace. Entries sync across devices, so a framework name you add on your phone is corrected on your Mac too. For the full setup walkthrough, see Hovor's custom dictionary setup guide.
Hovor's three relevant tone styles serve different developer text: cleanup removes filler and fixes grammar without restructuring, formal auto-lists clear enumerations (useful for a PR "Changes:" section), and coding converts short spoken syntax or a natural-language spec into formatted code. Pick the tone to match what you are dictating, not what feels safest by default.
| Tone style | What it does | Best for |
|---|---|---|
| Cleanup | Removes filler words, fixes grammar and punctuation, keeps every idea, no restructuring | AI assistant prompts, review comments, chat, docs |
| Formal | Same cleanup, plus auto-numbering or bulleting when the dictation contains a clear parallel enumeration | PR descriptions with a "Changes:" list, structured docs |
| Coding | Converts spoken syntax to real code, or generates a small snippet from a description, formatted per-language | Short SQL/API-shaped statements, one-off function generation |
Dictation does not eliminate keyboard use for a developer — code still gets typed — but it offloads the parts of the job that are pure writing: commit messages, PR text, review comments, docs, and chat, all natural-language tasks rather than character-precise syntax.
On macOS, Hovor's hands-free auto-stop mode starts recording with one trigger and stops automatically when you go quiet, while continuous mode keeps listening across multiple pauses so you are not holding a hotkey down through an entire paragraph — see the hands-free dictation guide for setup details. For developers managing RSI, removing the sustained key-hold is often as valuable as removing the typing itself. For the broader picture on dictation and RSI, see Hovor's guide to voice typing for RSI.
Hovor's macOS global hotkey works from any app — an IDE, a terminal, a browser tab, or an assistant's chat panel — and pastes cleaned text wherever the cursor currently sits, with no window switching or copy-paste. Four trigger modes, set in Settings → Hotkey, cover different work styles: Hold to talk, Toggle, Hands-free (auto-stop), and Hands-free (continuous).
Hold to talk records only while the key is held down and stops the moment you release it, best for short, precise insertions. Toggle starts recording on one press and stops on a second press, suited to full sentences. The two hands-free modes (Parakeet-only, macOS) free you from touching the key again after the first press: auto-stop ends the recording once it detects a pause, continuous keeps listening across multiple pauses until you stop it manually. This matters for a developer workflow specifically because the same session often moves between an editor, a terminal, and a browser several times a minute; a global hotkey means dictation follows you rather than requiring you to go find it.
Hovor ships today on iOS and macOS, the two platforms shown below as "Yes." A native Windows client, built with Tauri (Rust + React), is in active development and is not yet a shipping release — it is not a fit yet if you develop primarily on a Windows PC. There is no Linux client.
| Platform | Available today |
|---|---|
| iOS | Yes |
| macOS | Yes |
| Windows (Tauri) | In development, not yet shipped |
| Linux | No |
If you develop primarily on a Windows PC, Hovor is not yet available for your main machine. The global hotkey described above is a macOS feature today; it is planned for the Windows client once it ships.
Free tier: 2,000 words/week on any platform. Pro Monthly is $11.99/month, Pro Yearly is $89.99/year (unlimited words, 5 devices). Local Unlock ($49.99 / 1999 UAH, one-time) adds unlimited on-device dictation. BYOK Unlock ($24.99 / 999 UAH) lets you use your own OpenAI or Anthropic key. Bundle: $69.99 / 2799 UAH.
Not for full files. Hovor's coding tone can turn a short spoken statement like "select star from users where name equals Kyrylo" into SELECT * FROM users WHERE name = 'Kyrylo'; and can generate a small function from a natural-language description, formatted per-language (Prettier, Black, gofmt, rustfmt, swift-format, ktlint, google-java-format, pgFormatter, or shfmt depending on the detected language). It is not built for dictating long files character-by-character, precise punctuation-heavy expressions, or multi-file edits — for that, typing or an AI coding assistant remains faster and more exact.
Use Hovor's default cleanup tone, not the coding tone. AI coding assistants read natural language, so you want filler words removed and grammar fixed, not code-formatted output. Trigger Hovor's global hotkey (macOS) or keyboard extension (iOS) with your cursor already in the assistant's chat input, describe the change you want in plain sentences, and let the cleanup pass tidy punctuation before it lands in the input field. No special configuration is required.
Through the custom dictionary. You add mapping rules such as kubernetes|cube ornetes → Kubernetes or postgresql|post gres sequel → PostgreSQL. The replacement is applied to the raw transcription first, then your dictionary entries are also passed to the AI cleanup pass as vocabulary context, so the model applies your preferred spelling consistently even in sentences it has not seen the term in before. Entries sync across your devices.
It offloads the non-code text around your work: commit messages, PR descriptions, code review comments, Slack updates, and documentation, all of which are natural-language writing rather than character-precise syntax. On macOS, Hovor's hands-free auto-stop and continuous modes let you dictate without holding a key down, which removes the sustained key-hold that many RSI sufferers report as its own source of strain.
Not yet. Hovor ships today on iOS and macOS. A native Windows client, built with Tauri (Rust + React), is in active development. It is not a shipping product yet, and there is no Linux client. If you are on Windows or Linux today, Hovor is not a fit until the Windows release ships.
Hovor handles the natural-language writing around your codebase: AI assistant prompts, commit messages, PR descriptions, review comments. Free tier: 2,000 words/week. Available on iOS and macOS.
Get Hovor