This is a submission for the GitHub Copilot CLI Challenge
What I Built
I built VoiceDev — a VS Code extension that lets you control your entire development workflow with your voice. Not just dictation. Real workflows. Git commits, file navigation, Copilot CLI queries, all of it — spoken.
It started with a simple question I kept asking myself: What if I could be even more productive?
I was tired of hunting down VS Code shortcuts. Tired of configuring keybindings just to do something simple. Tired of dictation tools that let you talk at your editor but never to it.
I wanted more.
I wanted to say "git commit message fixed the auth bug" and have it just... happen. No terminal. No command palette. No twelve-step shortcut sequence I'd forget by next week.
🗣️ "git status" → "git commit message fixed the auth bug" → "git push"
A full commit cycle. Just my voice.
🗣️ "open file server.ts" → "go to line 42" → "format document" → "save all"
Navigate, edit, save. Spoken.
🗣️ "ask copilot explain this file" → "copilot commit"
AI-assisted development, triggered by voice.
Here's the thing — I originally wanted to build a standalone voice app. But I spend most of my waking hours inside my IDE. I never want to leave it. So why would I build something that lives outside the place where I actually work?
That's how VoiceDev became a VS Code extension instead.
More Than Dictation
Most voice tools for developers are glorified speech-to-text. You talk, text appears. That's it.
VoiceDev is different. It understands intent.
Say "format the document" and it triggers format-document even though the wording isn't exact. Fuzzy matching with confidence scoring. If VoiceDev isn't sure what you meant, it falls back to dictation instead of guessing wrong. Nine commands support wildcard patterns — say "git commit message fixed the auth bug" and it extracts "fixed the auth bug" as your commit message.
Development using voice is always faster than typing. Voice lets you map your raw thoughts directly into action. No translation layer. No "let me remember the shortcut for that." Just think it, say it, done.
Privacy and Saving Money
Honestly? I'm tired of paying for things that should be free.
Privacy and saving money — that's the core philosophy of VoiceDev. Your voice, your rules.
Want speed? Use Groq or Voxtral (Mistral) — both blazing fast, both free tier.
Want your voice to never leave your machine? Switch to Local mode. It runs via faster-whisper, fully offline. Zero API keys. Zero cost. Unlimited usage. Your audio stays on your hardware.
That choice matters. A developer in a startup who can't expense API keys should still have voice-native workflows. A freelancer working from a cafe on public WiFi shouldn't have to stream their voice to the cloud. A developer who just values their privacy shouldn't have to compromise on features.
VoiceDev gives you cloud speed or local privacy. Always your choice.
Safety, Because Voice Is Powerful
Force push is blocked. Git push requires confirmation. VoiceDev doesn't guess on destructive commands.
When you give your voice the power to control your editor, you better make sure it can't blow things up by accident.
Demo
🔗 GitHub Repository: github.com/mohitSharma74/voicedev
🎨 Install from VS Code Marketplace: VoiceDev (Preview)
🎥 Watch VoiceDev in Action:
My Experience with GitHub Copilot CLI
This is my second submission for the Copilot CLI Challenge. My first was Ukiyo-Tone — a VS Code theme bundle inspired by Japanese woodblock art. That project taught me how to work with Copilot CLI. VoiceDev is where I pushed it to its absolute limits.
My Framework
I follow a pipeline for everything I build: Idea → Philosophy → Brand Identity → PRD → Milestones → Execution → Testing. (Hint: Im actually working on a blog series to talk about this framework in depth and how can someone leverage it with their own workflow)
Every product. Same framework. No exceptions.
GitHub Copilot CLI was my partner from Idea all the way through Milestones. Not just for writing code — for thinking through the product. During the Idea phase, I used it to research the landscape of voice-controlled dev tools and VS Code extension architecture. During Philosophy and Brand Identity, it helped me articulate VoiceDev's core beliefs. By PRD and Milestones, it was helping me break down a multi-provider speech-to-text architecture into shippable chunks.
But here's what made the difference.
Agent Skills: Giving AI Power and Boundaries
I created Agent Skills — guardrails that made sure Copilot CLI could never make a decision outside my philosophy and vision.
Think of it as a leash. Not because the AI is dangerous, but because AI agents drift. Without constraints, they optimize for "technically correct" instead of "aligned with what I'm actually building."
Privacy-first? Non-negotiable. Local offline mode? Must exist from day one. Progressive disclosure? Every suggestion had to fit the vision.
Agent Skills kept Copilot CLI building my product, not just a product. That's a distinction most people miss when they work with AI agents. You need to control its powers, not just unleash them.
Beyond Building: Reviews and Delegations
Copilot CLI wasn't just for building. Once code was written, I used it for code reviews on my own repository — catching patterns I'd miss after staring at the same codebase for days. And for cloud delegations — offloading deployment tasks and infrastructure decisions without context-switching out of my terminal.
Build, review, deploy. All from the CLI. All with context.
The Thing That Almost Killed It
I need to be real about something.
There aren't many audio libraries out there that work cross-platform and give you the fine-tuning you need for a product like this. In the early days, I was searching for a library that could handle recording, audio processing, and work on Windows, Mac, and Linux without falling apart.
I almost had to kill the idea.
Every library I tried had a dealbreaker. Wrong platform support. No fine-tuning. Bad audio quality. Limited format options. I spent days on this — not writing features, just trying to find the foundation.
The library I ended up with has a big compromise: size. It's heavier than I'd like. But it checks every other box — cross-platform, reliable, configurable. Sometimes shipping means accepting trade-offs and moving forward.
That's not the kind of thing you read about in product launches. But it's the truth of building something real.
The Connected Ecosystem
What surprised me most wasn't Copilot CLI by itself. It was the connected ecosystem.
I'm fully invested in the GitHub ecosystem, and with MCP servers plugged in, Copilot CLI always had the right context available. That's massive when you're building something complex.
The Microsoft Learn MCP server is a perfect example. When I was figuring out how to package and publish a VS Code extension — which has its own quirks and gotchas — I didn't have to tab out and dig through Microsoft's documentation. Copilot CLI already had that context through the MCP server. I could ask questions and steer my product based on real, up-to-date documentation without leaving my terminal.
Same with GitHub MCP. Issues, PRs, repo context — all available to the agent without me copy-pasting anything into a chat window.
That connected ecosystem is what made it possible to ship VoiceDev from idea to preview in 16 days.
Let me say that again. A voice-controlled VS Code extension with four speech-to-text providers, 30+ commands, fuzzy matching, wildcard pattern extraction, a local offline mode with faster-whisper, Copilot CLI integration, Copilot Chat integration, audio feedback, and a Command Center webview.
Sixteen days. I'll take that win.
The Moment It Clicked
There's a moment I keep coming back to.
I used GitHub Copilot CLI to build the feature inside VoiceDev that calls GitHub Copilot CLI by voice.
Say "ask copilot explain this error" and VoiceDev sends it to Copilot CLI. Say "copilot commit" and Copilot generates your commit message. Say "copilot suggest how to list docker containers" and you get an answer without touching your keyboard.
Seeing it work for the first time — my voice triggering the same tool that helped me build the feature — felt like a real sense of achievement. Not because it was technically impressive. Because it was following my vision. The thing I imagined during the Idea phase was now real and doing exactly what I designed it to do.
That's the feeling you build for.
Honestly...
I'm a TUI person. I live in the terminal. I want to do everything from my command line.
Copilot CLI meets me exactly where I already work. No new UI. No context switching. Just my terminal, my agent, and a product that shipped in 16 days because the ecosystem gave me superpowers without asking me to change how I work.
Copilot CLI + MCP servers + Agent Skills isn't just "AI writes code for you." It's an agent that builds with you, inside boundaries you define, with access to the documentation and context it needs. That's a development partner.
And VoiceDev is proof of what happens when that partnership works.
What's Next
VoiceDev is at v0.1.0-preview. Here's where it's heading:
Action Chaining — say "git workflow" to run diff → stage → commit → push in one voice command. Chain any combination of built-in commands.
Custom Macros — say "deploy staging" to trigger your own sequence of git pull → build → deploy.
AI-powered workflows — inline completion and coding assists, triggered by voice.
Real-time translation — speak in any language, VoiceDev understands in English.
The vision is simple. Your voice is your IDE.
That's exactly what I'd like you to take away from this.
Voice-native development isn't about replacing your keyboard. It's about giving developers a faster, more natural way to do the things that pull them out of flow. Voice maps your raw thoughts directly into action — no shortcuts to memorize, no menus to navigate, no flow to break.
And GitHub Copilot CLI made it possible to ship this in 16 days instead of 16 weeks. Not because it wrote the code for me. Because it thought through the product with me, stayed inside my guardrails, and had the context it needed through a connected ecosystem that just works.
If you want to try it: install VoiceDev from the marketplace. Press Ctrl+Shift+V. And just... talk to your editor.
🔗 Repo: github.com/mohitSharma74/voicedev
🎥 Demo: youtube.com/watch?v=65OpDcS4oHM
🎨 Marketplace: VoiceDev (Preview)
While you are at it, show me some love on these 🥰 -
🆇 Twitter/X: https://x.com/MohitSharm95396
⛓️💥 Linktree: https://linktr.ee/onlyonemohitsharma
📸 TerminalSnap: https://terminalsnap.online/
🎋 Ukiyo-tone: https://marketplace.visualstudio.com/items?itemName=OneMohitSharma.ukiyo-tone
Top comments (0)