A new way to use your computer. Instead of typing, clicking and searching, you say what you want. ahsk writes it, cleaned up, in whatever app you are in; one shortcut (⌘ Space on the Mac, Win + Space on Windows) opens any app, file, site or command; and on the Mac, Alan does tasks for you in Calendar, Reminders and Messages.

Who is Alan, and what can he do?

Alan is the part of ahsk that does things, not just types them. On the Mac, hold fn + Option and say what you want done: move a meeting, set a reminder, text someone, answer the email on your screen. He acts and then offers Undo. Anything that cannot be taken back, like sending a message, waits for your OK first. He also remembers how you work, such as how long your meetings run.

What can the launcher do?

Press ⌘ Space (Win + Space on Windows) over any app and type. Half a name opens an app or finds a file. A keyword and a search goes straight to the site: g for Google, yt for YouTube, gpt for ChatGPT, cl for Claude, ps for Perplexity. It does sums, runs commands like sleep, lock and empty trash, and on the Mac also clean, dark mode, awake 2h and remind. Anything else, it asks AI.

How is ahsk different from ChatGPT or Claude?

ChatGPT and Claude live in a tab: you open it, copy text in, and paste the answer back. ahsk works where you already are. It writes into the app you are using, answers from one shortcut, and on the Mac Alan acts in your own calendar and messages. And when you do want ChatGPT or Claude, the launcher takes you there with your question already typed.

How much does ahsk cost?

Your first 10,000 dictated words are free, with no card required, which is enough to find out whether ahsk fits how you work. Invite a friend and you get 10,000 more words while they start with 5,000 extra. Pro is €9.99 a month for unlimited dictation and every feature, and you can cancel any time.

What platforms is ahsk available on?

macOS 14 (Sonoma) and above, Windows 10 (1809) and Windows 11, and iPhone and iPad with iOS 17 and above. One account covers all three. Alan is on the Mac; the launcher is on Mac and Windows; dictation is everywhere. Linux is not planned.

Does ahsk work offline?

No. Speech is turned into text in the cloud, and Alan and the launcher’s AI answers need a connection too. It is built to stay quick on slow connections, and a dictation that fails is kept so you can retry it.

How does ahsk protect my privacy?

Your speech is sent to our transcription provider, turned into text, and then discarded. We never store your audio and we do not keep transcripts on our servers. Your dictionary, learned words and Alan’s memory stay on your device. Nothing you say is used to train AI models, and you can revoke permissions or delete your account at any time.

By Jithin Kumar Palepu · 2026-10-23 · 10 min read

Dictation for Developers: Voice Input in Cursor, Terminals, and PRs

Dictation for developers is speech-to-text configured to match technical software environments like Cursor, command line terminals, and pull request reviews.…

Dictation for developers is speech-to-text configured to match technical software environments like Cursor, command line terminals, and pull request reviews. Instead of transcribing raw spoken words into conversational text, developer-focused dictation detects the active application. It formats spoken input as code references, shell commands, or structured markdown comments without requiring manual cleanup, keyboard mode switches, or repetitive copy-pasting.

Key Takeaways

According to Coderfile, 84% of developers use or plan to use AI tools in development, up from 76% in 2024.
Dictation expands input throughput from an average typing speed of 40 words per minute up to 125 words per minute, as reported by Zapier.
A 60-word pull request description takes about 25 seconds to speak versus 2 to 3 minutes to type, according to VoltType.
Modern voice systems offload 60% to 80% of daily typing volume, helping developers avoid 40 hours lost annually to repetitive strain injuries, based on findings from Whispererapp.
Ahsk handles application-specific formatting on Mac and Windows, offering 10,000 words free to start.

What is developer dictation?

Developer dictation is the practice of using voice recognition software specifically adapted to handle programming syntax, technical vocabularies, file structures, and development tools. Unlike general speech-to-text engines designed for continuous prose or conversational messaging, developer dictation recognizes the context of code editors, terminal windows, and issue trackers. It accurately parses technical jargon, camelCase variables, Git commands, and file paths without converting them into standard spoken English prose or mangling syntax.

+-------------------------------------------------------------+
|                      Spoken Input                           |
|       "check git status then add user auth controller"      |
+-------------------------------------------------------------+
                               |
                               v
               [ Ahsk Target-App Detection ]
                               |
         +---------------------+---------------------+
         |                                           |
         v                                           v
  Inside Terminal:                            Inside Cursor:
  $ git status && git add \                   Look at `@user_auth_controller`
    controllers/user_auth_controller.ts       to handle user session updates.

Standard operating system dictation was built for email and text messaging. When a developer speaks into a source code file or a terminal prompt using generic tools, the output often breaks syntax. A command like npm run test gets transcribed as "npm sprint test" or capitalized as "Npm run test."

Developer dictation bridges this gap through context awareness. The software inspects the focused window. When you dictate inside an editor like Cursor or VS Code, it recognizes code structures, file paths, and syntax delimiters. When you switch to a terminal prompt, it preserves lowercase formatting, spaces, and flags.

In an analysis by Zapier, the OpenAI GPT-4o Transcribe model demonstrated a word error rate of 2.46% by using multimodal context to process and format speech. Applying contextual models to technical work allows developers to dictate documentation, code reviews, and commit messages naturally.

How does target-app detection change voice formatting?

Target-app detection inspects the active operating system window to adapt transcription rules before inserting text. When Ahsk detects Cursor or an integrated development environment, it maintains indentation, converts mentioned files into IDE symbols like @filename, and preserves exact syntax naming conventions. In contrast, when Ahsk detects an email client or Slack, it applies natural sentence capitalization, conversational punctuation, and paragraph breaks suitable for human discussion.

Without app detection, developers waste time reformatting transcribed text. If you dictate a quick explanation of a bug fix in Slack, you want complete sentences, commas, and friendly capitalization. If you speak that exact same explanation as a prompt inside Cursor Composer or an AI agent, you want concise file pointers and variable names.

Here is how context changes formatting across different developer targets:

Code editors (Cursor, VS Code): Words like "file auth controller dot ts" format directly as @auth_controller.ts. Function references use code formatting or backticks.
Terminal emulators (iTerm2, Terminal, PowerShell): Input defaults to lower case, removes trailing punctuation like periods, and preserves flags like --dry-run.
Git interfaces and pull requests: Text expands into structured markdown lists, ticket tags, and standard section headers like "Summary" and "Testing Steps."
Team chat (Slack, Discord): Output uses standard conversational punctuation and natural casing.

The Ahsk use cases document how app detection eliminates repetitive editing. By matching the target application automatically, the software removes the friction of switching typing habits between your shell, editor, and browser.

Target Application	Spoken Input	Standard Transcription	Context-Aware Output
Cursor / IDE	"look at auth service dot ts and fix token expiry"	Look at auth service. Ts and fix token expiry.	Look at `@auth_service.ts` and fix token expiry
Terminal	"git commit message fix user logout bug"	Git commit message fix user logout bug.	`git commit -m "fix user logout bug"`
GitHub PR	"summary update database schema and add test cases"	Summary update database schema and add test cases.	Summary - Update database schema - Add test cases
Slack / Teams	"hey team pr is ready for review"	hey team pr is ready for review	Hey team, PR is ready for review!

Why do developers use voice input for pull requests and code reviews?

Developers use voice input for pull requests and code reviews because spoken communication is three times faster than typing prose. While programmers type code at modest speeds due to syntax and logic constraints, technical prose creates an administrative bottleneck. Dictating code review summaries and pull request descriptions allows engineers to provide thorough, high-quality feedback in seconds without suffering keyboard fatigue or hand strain.

Typing documentation is often the most delayed part of software delivery. As measured by VoltType, a 60-word PR description takes about 25 seconds to say compared to 2 to 3 minutes to type manually. Over several pull requests each week, that difference adds up to hours of saved engineering time.

Furthermore, physical health drives developer adoption. Whispererapp notes that developers typically spend 40% to 60% of their day typing prose, documentation, and chat. Voice dictation offloads 60% to 80% of that keyboard volume. If repetitive strain injury (RSI) sidelines an engineer for one week per year, that represents 40 hours of lost productivity. Speaking technical prose directly into PR templates protects wrists while maintaining documentation standards.

Thorough code reviews also improve team code quality. When typing reviews feels slow, developers frequently leave terse comments like "LGTM" or vague suggestions. Speaking allows reviewers to explain edge cases, design trade-offs, and alternative architectures in depth without breaking their flow state.

How to use voice input in terminals and AI coding agents

Using voice input in command line terminals and AI coding agents requires mapping global push-to-talk shortcuts to transcription tools that feed the terminal prompt buffer. Developers hold a key, speak instructions or shell flags, and release the key to populate the input buffer. This workflow allows developers to prompt local coding agents and execute commands without moving their hands away from the keyboard or switching windows.

Developers Digest highlights this hands-free approach for tools like Claude Code, where users hold a hotkey, dictate a complex agent instruction, and release to fill the prompt buffer directly. This method avoids the awkward context switching of typing lengthy multi-sentence prompts into command line interfaces.

Follow these steps to configure an efficient developer dictation workflow:

Select an input key: Assign a single global shortcut that does not conflict with your IDE keybindings. Dedicated function keys (like F5) or double-tap modifiers (like double-tapping Control) work best.
Set up active app targeting: Ensure your speech engine detects window focus. In Ahsk, app detection for editors, terminals, and chat is enabled out of the box on Mac and Windows.
Dictate commands in phrases: Speak terminal commands in single cohesive phrases rather than isolated words. Speaking "docker compose up build" in one breath ensures the engine recognizes the compound technical command.
Use file references explicitly: When directing AI coding assistants like Cursor Composer or Claude Code, state the file extension clearly (e.g., "point to main dot ts"). Target-aware software converts this phrase into an IDE-recognized symbol like @main.ts.
Review before execution: For terminal commands, keep automatic carriage returns turned off. Always inspect the generated shell string before pressing Enter to prevent unintended executions.

What are the technical limits of voice coding today?

The technical limits of voice coding center on arbitrary syntax, exact variable capitalization, and complex nested punctuation. While voice dictation excels at prompts, commit messages, comments, and documentation, dictating raw character-by-character code like nested bracket structures or regular expressions remains slower than using a physical keyboard. Voice is most effective for driving AI agents and writing prose, not typing line-by-line syntax.

A developer rarely needs to speak individual semicolons or curly braces today. According to Coderfile, 51% of professional developers now use AI tools daily. Rather than dictating raw code line by line, engineers use voice to direct agents: "Refactor the payment processor to handle network timeouts." The AI generates the syntax while the developer supplies the intent.

However, edge cases still occur:

Lookalike menu items and UI buttons: Automated on-screen tools can struggle with identical symbols. In internal tests documented in the Ahsk release notes, an initial system failed five out of five times (0/5) when asked to find the shortcut for a lookalike tab icon; reading the application menu system directly brought success to 10 out of 10 (10/10).
Homophones in variable names: Words like "byte" and "bite" or "cache" and "cash" can confuse models lacking local repository context.
Microphone quality and noisy environments: Background office chatter degrades accuracy unless noise suppression is configured properly.

For these reasons, developers adopt hybrid workflows. You use the keyboard for precision syntax, structural editing, and refactoring, while using voice input for agent prompting, terminal commands, PR summaries, and issue updates.

How Ahsk fits the developer workflow

Ahsk is built as a complete computer interface driven by voice. Rather than acting as a simple text box transcription tool, Ahsk integrates across your operating system. Version 6.14 on Mac and Version 1.23 on Windows introduced target-app detection designed specifically for email, code editors, terminals, and chat.

Beyond dictation, Ahsk includes Alan, an on-screen assistant. If you are stuck inside complex developer settings or new software, you can ask Alan how to do something. Alan looks at your screen, points at the right UI control, and talks you through the steps until you finish or press Escape. Ahsk provides 20 interactive guides per month on the free tier, and 150 guides per month on Pro.

For developers working on long compiles, remote uploads, or downloads, Ahsk 6.13 added a Mac utility to keep your machine awake with the lid closed. Typing "awake 2h" into the Ahsk launcher keeps the laptop active for exactly two hours, automatically ending itself on time or when reaching a safe battery threshold. Ahsk also searches text inside screenshots locally on your device, making it easy to find error messages, logs, and technical diagrams captured during debugging sessions.

You can learn more about configuration and keyboard shortcuts in the Ahsk Docs. Ahsk is free to start with 10,000 words of transcription included, while the Pro plan is €9.99 per month.

Frequently Asked Questions

Can developer dictation write raw code syntax accurately?

Developer dictation handles comments, docstrings, prompts, and prose well, but speaking character-by-character syntax like brackets and semicolons is inefficient. Most developers use voice dictation to instruct AI coding agents (like Cursor Composer) rather than dictating raw punctuation.

Does voice dictation work with terminal commands?

Yes. When voice dictation software detects a terminal window, it strips out sentence capitalization and ending punctuation. This allows you to speak commands like "git status" or "npm run build" directly into the shell prompt.

How does dictation handle camelCase or snake_case variables?

Context-aware tools recognize programming terminology and adapt spoken words into standard variable naming conventions based on surrounding code. If ambiguity occurs, stating the phrase as a single continuous technical term helps the speech engine format it accurately.

Does dictating code send proprietary source files to third-party servers?

It depends on the tool's architecture. Some tools run Whisper models locally on your machine, while others send audio to high-speed transcription APIs. Check your tool's documentation to ensure its data policies match your employer's security requirements.

How much faster is dictating technical prose compared to typing?

Dictation allows most developers to speak at 80 to 125 words per minute, whereas typing prose averages around 40 words per minute. For PR descriptions, bug reports, and design docs, speaking is two to three times faster.

Next Step: Download Ahsk for Mac or Windows to get 10,000 free words of context-aware dictation inside Cursor, your terminal, and GitHub.