What We Do

VoiceCoder lets developers write code by speaking. You describe a change out loud, an AI agent makes it in your own workspace, and you review the result in VS Code. It is built for developers who can’t use a keyboard or mouse reliably, and it works for anyone who would rather talk than type.

1. 1

   Speak

   Your microphone is on from the start. Say what you want to build or change, then pause. Your words appear as you talk.

2. 2

   The agent codes

   Your request is cleaned up and sent to a coding agent, which edits the files in your workspace while you watch.

3. 3

   Review and approve

   See the changes in VS Code next to a plain-language summary. Say “approve” to keep them, or ask to undo them.

• Hands-free from the first second
• Live transcript
• Results read aloud
• Your own Coder workspace

How to Set Up

Run VoiceCoder in your own Coder workspace, or try it on your computer with Docker. Everything you need is in our GitHub repo.

• Docker with Compose v2
• A Coder deployment (for workspaces)
• An Anthropic API key
• Chrome or Edge with a microphone

On Coder

1. 1

   Step 1: Clone the repo

   Clone it with its submodules (the bridge and the front end live in their own repos).

   git clone --recurse-submodules https://github.com/TheAtlantaDream/VoiceCoder.git Copy
2. 2

   Step 2: Push the Coder template

   From the repo folder, push the template in template/ to your Coder deployment.

   coder templates push voicecoder --directory ./template Copy
3. 3

   Step 3: Create a workspace

   In Coder, create a workspace from the VoiceCoder template and paste your key into the Anthropic API Key parameter. Without a key, VoiceCoder still transcribes and cleans your speech, but no coding agent runs.

4. 4

   Step 4: Wait for the first build

   The first boot builds the images and downloads the speech model (about 650 MB). After that, workspaces on the same host start quickly.

5. 5

   Step 5: Open VoiceCoder and talk

   Open the workspace’s VoiceCoder app in Chrome or Edge and allow the microphone. Say “hey VoiceCoder” or “Coder,” followed by your request. VS Code Web is on the same page.

On your computer

1. 1

   Step 1: Add your key

   After cloning, create your environment file and set ANTHROPIC_API_KEY in it. Never commit .env.

   cp .env.example .env Copy
2. 2

   Step 2: Start the stack

   This starts the Parakeet speech-to-text service, the bridge, the front end and VS Code Web with Docker Compose. Use logs to follow output and down to stop.

   ./scripts/run-local.sh Copy
3. 3

   Step 3: Open it

   Go to 127.0.0.1:8080 on the same computer and start talking.

Stuck? The README has the full details, including how to update the bridge and front-end submodules.

About Us

We’re a team of four students from Georgia Tech, Georgia State and UGA who built VoiceCoder so that developers who can’t rely on a keyboard or mouse can still write code.

• 

  Stephen Sulimani

  Coder template + Parakeet

  MS in Quantitative & Computational Finance at Georgia Tech, BS in Computer Science from UGA. A backend and quant engineer who built the Coder workspace template and speech-to-text that every VoiceCoder session runs on. suli.nyc

• 

  Carter Tierney

  Go bridge

  MS in Computer Science at Georgia State University. Built the Go bridge, the real-time link that streams your voice in, turns it into a clean request and hands it to the coding agent.

• 

  Uyiosa Nehikhuere

  Agent runner

  Computer Science at Georgia State University. Built the agent runner, which connects Claude Code to each workspace and sets the rules the agent follows when it edits your project.

• 

  Seth Voisine

  Front end + design

  Computer Science at the University of Georgia. Designed and built VoiceCoder’s hands-free, accessible interface, along with the project’s visual style and this site.