Inspiration Changing a website price or image shouldn't cost $100 or take three days. We wanted to kill the "maintenance contract" for business owners. VocalCommit replaces complex dashboards with a simple voice command, democratizing web development for non-coders.
What it does VocalCommit is a Voice-Native DevOps Orchestrator that transforms spoken commands into production-ready code.
Talk to Code: Converts spoken intent ("Add a dark mode toggle") into production-ready commits with multi-file coordination.
GitOps Security: Manages the full Git lifecycle with a two-stage approval system:
Local commits happen automatically after AI validation Manual approval required before pushing to production GitHub repository Easy rollback options (soft/hard) for safety AI-Powered Planning: PM Agent analyzes requests and creates structured task plans with file targeting and dependency management.
How we built it We built a "Multi-Agent Architecture" for intelligent code generation:
The Brain (Gemini 2.5 Flash): We leverage Gemini 2.5 Flash for its speed-to-intelligence ratio. It powers:
PM Agent: Analyzes voice commands and creates structured task plans Dev Agent: Generates production-ready code with multi-file coordination Testing Agent: Validates changes before committing The Context System:
Need-to-know architecture: Each agent only sees relevant files Multi-file awareness: Agents coordinate changes across related files (e.g., theme systems require Context → Hook → Component → CSS) File sanitization: Prevents sentence-like filenames and ensures proper naming conventions The Body: Python FastAPI orchestrator with:
WebSocket real-time communication GitPython for local repository management GitHub API integration for production deployment Rate limiting to manage API quotas The Face: React + TypeScript + Vite UI with:
Voice and text input support Real-time workflow status updates Commit approval modal with three options (Approve/Soft Rollback/Hard Rollback) Production monitoring dashboard Two-Repository System:
Local development in orchestrator/todo-ui/ Production deployment to separate GitHub repository File synchronization on approval Challenges we ran into The "Blind" AI: The AI could write code but couldn't verify results. We built a testing agent that validates syntax and functionality before committing.
Multi-File Coordination: Theme systems and complex features require multiple files working together. We implemented context-aware file targeting where the PM Agent identifies all necessary files and the Dev Agent coordinates changes across them.
Filename Sanitization: AI sometimes suggested sentence-like filenames ("create a theme toggle component"). We built intelligent sanitization that extracts meaningful names (ThemeToggle.tsx) while preserving proper conventions.
Rate Limiting: Gemini API has quota limits. We implemented smart rate limiting with status tracking and wait times to prevent failures.
Production Safety: Needed approval workflow to prevent accidental deployments. We built a two-stage commit system with local commits first, then manual approval for production push.
Accomplishments that we're proud of Zero-Touch Code Generation: From voice command to production-ready code without manual file editing.
Intelligent Multi-File Coordination: Successfully implementing theme systems that require 4+ coordinated files (Context, Hook, Component, CSS) from a single voice command.
Safe Production Workflow: Built a robust approval system with three rollback options and easy revert capability for production commits.
Real-Time Updates: WebSocket-based communication provides instant feedback on task progress, approvals, and production pushes.
Smart Context Management: "Need-to-know" architecture keeps AI focused on relevant files, reducing token usage and improving accuracy.
What we learned Speed Matters: For voice interfaces, latency kills the vibe. Gemini 2.5 Flash's sub-second response time is essential for maintaining flow state.
Context is King: Providing related file context to the AI dramatically improves multi-file coordination and reduces errors.
Safety First: Manual approval before production deployment is crucial. Users need confidence that changes won't break their live site.
Business Impact: This isn't just a dev tool; it's an economic tool for small business owners who are currently locked out of their own websites.
What's next for VocalCommit Visual Verification: Adding screenshot capture and AI-powered visual review before committing UI changes.
Expanded File Support: Beyond React/TypeScript to support more frameworks and file types.
Team Collaboration: Multi-user support with role-based approvals and collaborative workflows.
CI/CD Integration: Automatic testing and deployment pipelines triggered by approved commits.
Mobile App: A pocket command center for business owners to update their sites on the go.
Log in or sign up for Devpost to join the conversation.