GreatAegis: Post-Quantum Secure Enterprise AI Gateway Integrating PQC Encryption and AMD ROCm Acceleration

This article introduces GreatAegis, an open-source enterprise AI gateway. Its core features include integrating post-quantum cryptography (PQC) encryption, dynamic hybrid workload routing, and AMD ROCm-accelerated isolated open-source large model inference, aiming to provide dual guarantees for AI security and performance. The original author/maintainer of the project is zulroxx, and the source platform is GitHub (link: https://github.com/zulroxx/GreatAegis). The update time is 2026-07-12T17:22:59Z.

With the widespread application of large language models (LLMs) in enterprise scenarios, data security and privacy protection have become core issues. Enterprise AI applications face two major risks: data theft and tampering during transmission and storage, and the potential threat of quantum computing to traditional encryption systems. Traditional encryption algorithms (such as RSA and ECC) are vulnerable to quantum computer attacks, so post-quantum cryptography (PQC) has become a research hotspot to address quantum threats. GreatAegis was born in this context, combining three key elements: post-quantum security, enterprise AI gateway, and high-performance inference acceleration, to provide a comprehensive solution that balances security and performance.

The core technologies of GreatAegis include three parts:

1. Client-side post-quantum encryption: Data is encrypted before leaving the user's device, featuring zero-trust architecture (data remains encrypted even if the gateway is compromised), quantum attack resistance (using NIST-standardized algorithms), and end-to-end protection (eliminating the risk of man-in-the-middle attacks);
2. Dynamic hybrid workload routing: Intelligently allocate resources based on task requirements—privacy-sensitive tasks are routed to local open-source models, high-performance tasks use AMD ROCm-accelerated nodes, and cost-optimized tasks select cost-effective resources;
3. Isolated open-source LLM inference: Built on AMD ROCm, including hardware-level isolation (GPU virtualization), model isolation (independent containers to prevent leakage and contamination), open-source transparency (auditable models like Llama/Mistral), and ROCm acceleration (compatible with open-source ecosystems and performance close to CUDA).

The main application scenarios of GreatAegis include:

• Financial services: PQC encryption and local deployment meet the security and compliance requirements for sensitive data, supporting applications such as AI customer service and risk assessment;
• Healthcare: End-to-end encryption and isolated inference protect patient privacy (compliant with HIPAA regulations), supporting medical Q&A and medical record summarization;
• Government agencies: Post-quantum encryption ensures data sovereignty and long-term security, resisting future quantum threats;
• Enterprise knowledge management: Private AI assistants process internal documents and business secrets to avoid data leakage.

The technical highlights of GreatAegis are:

1. Entry for AMD Developer Cloud Hackathon (Track3), making full use of AMD GPU features to achieve efficient ROCm inference acceleration;
2. Modular design: Encryption, routing, and inference modules can be upgraded or replaced independently, facilitating enterprise customization and community contributions;
3. Open-source ecosystem compatibility: Compatible with mainstream frameworks like Hugging Face Transformers and vLLM, reducing the migration threshold for enterprises.

GreatAegis represents the evolutionary direction of enterprise AI infrastructure that balances security and performance. With the implementation of NIST post-quantum cryptography standards and enterprises' increasing attention to AI data privacy, such solutions will be more popular in the market. The open-source nature promotes community collaboration, accelerates the popularization of post-quantum secure AI technologies, and provides reference cases for practitioners in AI security, quantum threats, and enterprise AI deployment. Early deployment of post-quantum security architecture will become an important part of enterprise AI strategies, and GreatAegis provides a feasible technical path for this.