CodeGuardian AI: Multi-Agent Code Automation

CodeGuardian AI is a fully autonomous multi-agent code modernization system built on the Band.ai platform. Five specialized AI agents (Conductor, Planner, Coder, Reviewer, and Merger) form a directed collaboration chain that takes a GitHub repository from analysis to pull request without human intervention. What makes CodeGuardian unique is true cross-provider diversity: each agent runs on a different LLM provider (OpenAI GPT-5, Anthropic Haiku, DeepSeek V4, OpenAI GPT-4o, Google Gemini) and a different agent framework (PydanticAI, AnthropicAdapter, LangGraph, CrewAI). All traffic routes through the AI/ML API unified gateway with a single key. The system supports five operational modes: Modernizer (upgrade deprecated patterns), CodePruner (remove dead code), Security (fix vulnerabilities), Suggestion (read-only analysis via GitHub Issues), and OnDemand (custom refactoring instructions). Users simply paste a repository link, select a mode, and the agent chain handles the rest, from cloning and scanning to implementing changes, reviewing diffs, and creating pull requests. Reliability is baked in through loop guards that prevent non-terminating agent ping-pong, mention auto-fill patches for CrewAI edge cases, and automatic room cleanup to avoid rate-limit storms. The system runs on Azure with Docker Compose, Nginx reverse proxy, and Dozzle for real-time log streaming. Built during the Band of Agents Hackathon, CodeGuardian demonstrates how Band.ai's collaborative agent infrastructure enables complex multi-framework orchestration where agents with different strengths hand off work, request revisions, and converge on terminal artifacts autonomously.