Informatik student & Founder of Aegisora. Building runtime security and governance layers for AI agents.

medium

devto

productHunt

github

projects

Eray Özer • Peerlist

Aegisora is an open-source runtime security and governance layer for AI agents.As AI agents gain the ability to call tools, APIs, databases, and production systems, traditional allow-or-block security models become difficult to apply safely.Aegisora sits between the AI agent and the systems it can act on, evaluating actions before execution.Every intercepted action can result in:✅ ALLOW — safe, policy-compliant actions continue immediately.⛔ BLOCK — clear policy violations and known attack patterns are rejected.⏸️ ESCALATE — ambiguous or high-risk actions can be routed to human review.Aegisora is built for developers working on autonomous agents, agent security, tool-call governance, prompt-injection defense, policy enforcement, observability, and auditability.The project is fully open source and welcomes contributions across:• Security policies and detection rules• Agent framework integrations• Security testing and attack cases• Benchmarks• Documentation• Developer toolingExplore the source, run Aegisora locally, and help shape an open-source security layer for the next generation of AI agents.GitHub:<a target="_blank" rel="noopener noreferrer nofollow ugc" class="break-all link-break text-gray-500 hover:underline" href="https://github.com/aegisora-ai/aegisora.ai">https://github.com/aegisora-ai/aegisora.ai</a>Live Demo:<a target="_blank" rel="noopener noreferrer nofollow ugc" class="break-all link-break text-gray-500 hover:underline" href="https://aegisora-ai.vercel.app">https://aegisora-ai.vercel.app</a>

Aegisora Open Source

Aegisora provides a narrow control plane for AI agents, enforcing zero-trust, least-privilege access to tools and APIs. It deploys as a sidecar proxy inside your VPC to intercept reasoning traces, block unauthorized calls, mask PII, and produce human-readable, immutable audit logs. Use the runtime governance console to orchestrate multi-agent swarms, define dynamic policies, and integrate alerts with Slack, Microsoft Teams, and PagerDuty. Choose an open-source sandbox, managed cloud, or dedicated enterprise VPC.<h3></h3>

Aegisora

Analyzed the Baidu Apollo stack's integration of HD Maps, AI-driven Perception, and control systems. This architectural study provides a technical foundation for my VYNO project. These principles directly support my goal of architecting a global autonomous mobility ecosystem.

Autonomous Vehicle Architecture: Apollo Frame

Over a rigorous 5-month period, I independently completed the full curriculum of Stanford University’s CS229: Machine Learning, originally taught by Andrew Ng. This project represents a deep dive into the mathematical foundations and architectural implementations of modern AI and Autonomous Systems.

Stanford CS229 Machine Learning

Developed a decentralized framework for resilient and scalable swarm intelligence, focusing on high-fidelity decision-making for autonomous multi-agent systems. Decentralized Consensus Protocol (DCP): Implemented a protocol to ensure collective agreement among agents while maintaining system stability. Byzantine Fault Tolerance (BFT): Engineered outlier rejection mechanisms using Global Error Coefficients ($\epsilon_{global}$) to ensure safety even when individual sensors fail. Physical Intelligence (PI): Integrated mathematical modeling to validate swarm perceptions and execute real-time autonomous actions in complex environments