DEV Community

1433_IBM
1433_IBM

Posted on

Codex Vitae: Building a Self-Evolving AI Agent Platform on Google Cloud Run

Codex Vitae: Building a Self-Evolving AI Agent Platform on Google Cloud Run

Author: Shantanu Awate

Introduction

At the Google Cloud Run Hackathon 2025, I set out to solve a common challenge: rapidly deploying, managing, and evolving AI agents at scale. Traditional AI agent development often requires manual redeployment, complicated monitoring, and slow iteration cycles — a bottleneck for small teams and solo developers.

Enter Codex Vitae, a multi-agent platform that automates the full lifecycle of AI agents, from creation to evolution.

Project Overview

Codex Vitae allows anyone to build their own AI agents through two modes:

Quick Create: Write a prompt, deploy in 1–3 minutes, and get a fully functional AI agent.

Architect Mode: Define the agent’s style, tone, beliefs, and rules. Enable auto-mutation to evolve the AI automatically based on feedback.

The platform runs entirely on Google Cloud Run, making it serverless, scalable, and modular.

The Three-Agent Architecture

Codex Vitae leverages a robust multi-agent system:

Orchestrator Agent (“Brain”)

Central command hub deployed on Cloud Run.

Manages deployment, stores agent genomes, and triggers evolution.

Specialist Agent (“Product”)

Template for user-defined AI agents.

Runs independently on Cloud Run for each deployed instance.

Feedback Agent (“Consultant”)

Evaluates user feedback and suggests improvements.

Feeds into the auto-mutation loop for self-evolving AI.

Each agent communicates seamlessly to create a continuous feedback-driven evolution loop.

Tech Stack

Languages & Frameworks: Python, React, Node.js

Cloud & Deployment: Google Cloud Run, Firebase Hosting, Google Artifact Registry

AI & ML: Google Vertex AI (Gemini 2.5 Flash), Google Agent Development Kit (ADK)

Database & Storage: Google Firestore

Containerization: Docker

Challenges & Learnings

Automating Evolution: Designing the feedback loop to generate actionable AI prompt updates was challenging.

Multi-Agent Communication: Ensuring reliable, stateless communication between agents on Cloud Run required careful orchestration.

Rapid Deployment: Cloud Run enabled near-instant deployment, which drastically accelerated development.

I learned a lot about serverless AI architectures, automated prompt engineering, and full-stack AI deployment workflows.

How to Try It

GitHub: https://github.com/Shantanu-00/Codex-Vitae

Demo: https://codex-vitae-hackathon.web.app
(Quick Create mode ready to test!)

Conclusion

Codex Vitae empowers anyone — solo developers or small businesses — to own intelligent AI assistants without infrastructure overhead. From idea, to deployment, to autonomous evolution, everything happens seamlessly on Cloud Run.

It’s a self-evolving AI maker for everyone, built to push the boundaries of multi-agent AI systems.

Top comments (0)