🎤 Voice Driven Development with Kiro Autonomous Agents

Convert spoken requirements into structured specifications or GitHub issues using AI transcription and LLM processing.

Speak your ideas, and this app turns them into actionable work. Record or upload audio, and the app transcribes your words and uses AI to generate structured requirements — or creates a GitHub issue that Kiro's Autonomous Agent picks up, implements, and delivers as a Pull Request. No typing required.

In Local mode, the generated specs drive Kiro IDE to implement the entire application through specs-driven development. In GitHub mode, Kiro's Autonomous Agent picks up the issue and raises a PR automatically.

🤖 Fun fact: This entire project was created using Kiro, an AI-powered development assistant that helped design, implement, and document the application from initial concept to deployment.

---

✨ Features

Feature	Description
🎙️ Browser Recording	Record requirements directly from your microphone
📁 File Upload	Upload .wav audio files
☁️ Amazon Transcribe	Cloud-based speech-to-text
🦜 Parakeet MLX	On-device transcription for Apple Silicon (optional)
📝 Local Mode	Generate requirements.md + tasks.md in Kiro spec format
🐙 GitHub Mode	Create GitHub issues directly from voice input
🤖 Configurable Model	Use any Bedrock model (Claude Haiku, Sonnet, etc.)

---

📋 Prerequisites

• uv package manager
• Python 3.12+
• AWS account with access to:
  • S3 — audio file storage
  • Amazon Transcribe — speech-to-text
  • Amazon Bedrock — LLM inference (Claude model access required)
• AWS credentials configured (~/.aws/credentials or environment variables)

---

🚀 Quick Start

1. Clone & install

git clone https://github.com/aws-samples/sample-voice-driven-development.git
cd sample-voice-driven-development
uv sync

2. Configure environment

cp .env.example .env

Edit .env with your values:

S3_BUCKET_NAME=your-s3-bucket-name
BEDROCK_MODEL_ID=eu.anthropic.claude-haiku-4-5-20251001-v1:0

# Optional: GitHub issue creation mode
GITHUB_TOKEN=ghp_your_token_here
GITHUB_REPO=owner/repo

3. Run

uv run streamlit run streamlit_app.py

Open http://localhost:8501 in your browser.

---

🐳 Docker

With AWS credentials volume mount:

docker build -t voice-driven-dev .
docker run -p 8501:8501 \
  -e S3_BUCKET_NAME=your-bucket \
  -v ~/.aws:/home/appuser/.aws:ro \
  -v $(pwd)/projects:/app/projects \
  voice-driven-dev

With environment variables:

docker run -p 8501:8501 \
  -e AWS_ACCESS_KEY_ID=your_key \
  -e AWS_SECRET_ACCESS_KEY=your_secret \
  -e S3_BUCKET_NAME=your-bucket \
  -v $(pwd)/projects:/app/projects \
  voice-driven-dev

---

🎯 Usage

1. Choose input — Record audio via microphone or upload a .wav file
2. Select output mode — Local (specs + tasks) or GitHub (issue creation)
3. Process — Click the process button to transcribe and generate output
4. Download — Get your requirements.md, tasks.md, or view the GitHub issue

Output Modes

Mode	Output	Requires
Local	projects/<name>/requirements.md + tasks.md	S3 bucket
GitHub	GitHub issue with kiro label — triggers Kiro Autonomous Agent to pick up and implement the task	GITHUB_TOKEN + GITHUB_REPO

---

🦜 Local Transcription (Optional)

For on-device transcription on Apple Silicon (skips S3 + Transcribe entirely):

brew install ffmpeg
uv sync --extra local-transcribe

Select Parakeet MLX in the sidebar transcription engine dropdown.

---

🤖 Kiro Autonomous Agent Mode (Optional)

Use Kiro's Autonomous Agent to close the loop entirely — speak your requirements, have them logged as GitHub issues, and let Kiro automatically pick them up, implement the code, and raise a Pull Request.

How it works

🎤 Voice Input ──▶ 🐙 GitHub Issue ──▶ 🤖 Kiro Agent ──▶ 📦 Pull Request

1. You speak your requirements into the Streamlit app (GitHub mode)
2. The app creates a GitHub issue with structured acceptance criteria
3. Kiro Autonomous Agent detects the new issue and starts working on it
4. Kiro implements the changes and opens a PR for your review

Setup

Step 1: Connect Kiro Autonomous Agent to your GitHub repo

Follow the Kiro Autonomous Agent docs to set up:

1. Open Kiro and navigate to Autonomous Agent
2. Connect your GitHub account and select the target repository
3. Configure which issue labels Kiro should pick up (e.g., kiro)

Step 2: Generate a GitHub Personal Access Token

1. Go to GitHub → Settings → Developer settings → Personal access tokens
2. Create a Fine-grained token scoped to your target repository with:
   • Issues: Read & Write
3. Copy the token

Step 3: Configure the app for GitHub mode

Add the token and repo to your .env:

GITHUB_TOKEN=github_pat_xxxxxxxxxxxx
GITHUB_REPO=owner/repo

Step 4: Use the app in GitHub mode

1. Run the app: uv run streamlit run streamlit_app.py
2. In the sidebar, select 🐙 GitHub — open issue directly
3. Record or upload your audio requirements
4. Click Process — a GitHub issue is created with the kiro label
5. Kiro Autonomous Agent picks up the issue, implements the code, and opens a PR

💡 Tip: The app automatically applies the kiro label to created issues. Make sure your Kiro Agent is configured to watch for this label so it triggers automatically.

---

🏗️ Architecture

flowchart LR
    A[🎤 Voice Input] --> B[🗣️ Transcribe]
    B --> C[🤖 Bedrock Claude]
    C --> D[📝 Local Specs → Kiro IDE]
    C --> E[🐙 GitHub Issue → Kiro Agent → PR]

Loading

---

🔐 AWS Permissions

Your IAM credentials need the following permissions:

Service	Actions
S3	s3:PutObject, s3:GetObject on your bucket
Transcribe	transcribe:StartTranscriptionJob, transcribe:GetTranscriptionJob
Bedrock	bedrock:InvokeModel for your configured model

---

⚙️ Configuration Reference

Variable	Required	Default	Description
S3_BUCKET_NAME	Yes*	—	S3 bucket for audio storage
BEDROCK_MODEL_ID	No	eu.anthropic.claude-haiku-4-5-20251001-v1:0	Bedrock model ID
AWS_REGION	No	Auto-resolved from model prefix	AWS region override
GITHUB_TOKEN	No	—	Enables GitHub issue mode
GITHUB_REPO	No	—	Target repo (owner/repo)
PARAKEET_MODEL_ID	No	mlx-community/parakeet-tdt-0.6b-v2	Local transcription model

* Not required when using Parakeet MLX engine in local-only mode.

---

📚 Resources

• Kiro IDE Docs — Specs-driven development with Kiro
• Kiro Autonomous Agent Docs — Let Kiro pick up GitHub issues and raise PRs
• Kiro Blog — News, tutorials, updates from the Kiro team - find it all here

---

🤝 Contributing

See CONTRIBUTING.md for guidelines.

---

🔒 Security

See CONTRIBUTING.md for reporting vulnerabilities.

---

📄 License

This project is licensed under the MIT-0 License.