CodeSentry

CodeSentry is an enterprise-grade, agentic AI-powered security and performance copilot built for modern AI and cloud-native workflows. It helps developers automatically analyze codebases, detect vulnerabilities, optimize AI/ML performance, and accelerate CUDA → ROCm migration for AMD hardware. The platform uses a multi-agent architecture consisting of specialized AI agents working together in real time. The Security Agent combines static analysis with semantic LLM reasoning to identify prompt injections, unsafe deserialization, exposed secrets, insecure API usage, and dangerous execution patterns. The Performance Agent focuses on ML optimization by detecting GPU bottlenecks, inefficient loops, repeated model loading, and hardware optimization opportunities for AMD MI300X accelerators. The Fix Agent automatically generates Git-style diffs and intelligent patch recommendations. In addition, the AMD Migration Advisor scans projects for CUDA-specific dependencies such as nvidia-smi, CUDA_VISIBLE_DEVICES, cuDNN, Flash Attention, FP16 usage, and BitsAndBytes configurations, then generates ROCm migration guidance with an AMD Compatibility Score. A major highlight of CodeSentry is its real-time AMD MI300X GPU metrics dashboard. During scans, the system streams GPU utilization, VRAM usage, memory bandwidth, temperature, power draw, and inference speed to the frontend using Server-Sent Events (SSE). The platform follows a strict Zero Data Retention (ZDR) architecture where every analysis session generates a privacy certificate, blocks outgoing network calls during scans, and wipes all data immediately after completion. Built using FastAPI, React, Vite, ROCm, Groq, and Qwen models, CodeSentry transforms traditional static analysis into a fully agentic AI workflow that not only detects problems but actively generates solutions for developers and enterprises.