Codex-Vitae

• 

  presentation file

• 

  achitecture

• 

  create-page

Project Story: Codex Vitae 🌱 Inspiration

I’ve always been fascinated by the idea of AI systems that can improve themselves. During this hackathon, I wanted to explore whether AI agents could evolve autonomously — learning from feedback and redeploying improved versions without human intervention. That curiosity led me to create Codex Vitae, a self-evolving AI agent platform.

🧠 What I Built

Codex Vitae is a multi-agent system powered by Google Cloud Run and Gemini AI, designed to manage the full lifecycle of AI agents — from deployment to evolution. It consists of:

An Orchestrator Agent that acts as the brain, deploying and evolving other agents.

Specialist Agents that perform specific AI tasks, each configurable through a genome-like structure.

A Feedback Agent that analyzes user input to guide evolution.

Each version learns from user ratings, automatically mutates its configuration, and redeploys itself to Cloud Run — simulating digital evolution.

🔧 How I Built It

I developed and deployed each service using FastAPI, Express and Google Cloud Run. I used Firestore to store agent genomes and feedback, and Artifact Registry to manage container images. The system communicates entirely through HTTP APIs, making it modular and easily scalable.

🚀 Challenges & Learnings

Setting up autonomous deployment pipelines through Cloud Run’s Admin API was complex but rewarding. I learned to manage container orchestration programmatically, handle Firestore concurrency, and design an agent system capable of adapting on its own.

💡 What I Learned

I learned how powerful feedback loops can be when combined with automation. Codex Vitae taught me that AI development doesn’t have to be static — it can evolve dynamically, just like life.

Built With

• express.js
• firebase
• google-adk
• google-artifact-registry
• google-cloud-run
• javascript-(react-for-frontend)
• node.js
• python
• vertexai