Leveraging AWS AgentCore for Prototyping and Secure Multi-Cloud Deployment in Apex Engine
Introduction to AWS AI Bedrock AgentCore and Apex Engine
In yesterday’s (05 August, 2025) AWS AI Bedrock AgentCore podcast episode, the team offered an insightful overview of how Amazon is approaching the development and deployment of AI agents at scale. The episode covered a range of high- to mid-level concepts around modular agent infrastructure, secure execution environments, memory management, and observability. For teams building future-facing products, it provided a clear window into how AWS envisions the agentic era evolving within the enterprise space.
We want to thank AWS and the Bedrock team for sharing such valuable context and making these capabilities accessible to a broad development audience. As we explore how agents will support real-time collaboration and intelligent automation inside our own platform, these kinds of open discussions help validate and refine our approach.
At TGS Tech, we are in active design and development of Apex Engine, a fully cloud-native, real-time 3D platform designed to support industries such as game development, simulations and training, architecture, digital twins, and smart infrastructure. Agentic workflows will play a significant role across this spectrum, from procedural content generation and automation to intelligent in-editor assistance and adaptive simulation control.
We see AWS AgentCore as a valuable short-term accelerator, particularly during the prototyping phase. It gives us the ability to explore and validate agent behavior, memory models, tool integration, and planning logic in a secure, modular environment. This early validation helps us define interface boundaries, observe real-world interaction patterns, and shorten our iteration cycle before investing heavily in a full-scale, platform-native implementation.
However, Apex Engine is being designed to meet more rigorous demands. Our clients require server and host flexibility, full control over AI model behavior, and assurance that their proprietary workflows and data will remain private and secure. AWS AgentCore, while powerful, is tied to Bedrock and the broader AWS ecosystem. It is not intended to support hybrid or multi-cloud deployment, and its key components such as memory, code interpretation, and tool access are not open or portable.
To mitigate these long-term risks and ensure both platform sovereignty and client trust, we are building our own internal LLMs and SLMs. These models will be trained, hosted, and versioned entirely in-house. This approach gives us control over everything from inference behavior to fine-tuning and optimization, while eliminating external dependencies that could otherwise impact IP ownership, latency, or compliance.
In parallel, we are developing Apex Engine’s agent infrastructure to be fully modular and deployment agnostic. Clients will be able to run the platform, including all agent services, on AWS, Azure, Alibaba Cloud, private servers, or air-gapped infrastructure. By investing early in cross-platform abstraction and isolating cloud-specific tooling, we are designing a system that is both forward-compatible and robust across verticals.
This paper outlines the strategic role AWS AgentCore plays in our early prototyping, the benefits and boundaries of that approach, and how we are leveraging it to build a flexible, secure, and scalable multi-cloud agent system within Apex Engine.
The Benefits of AWS AgentCore for Prototyping and Emerging Products
AWS AgentCore is a modular framework within Amazon Bedrock that provides developers with the tools to build, test, and scale intelligent AI agents. It includes key components such as:
- AgentCore Runtime: A serverless execution environment with tenant and session isolation.
- Memory Modules: Short- and long-term context storage for persistent, context-aware agents.
- Gateway and Identity: Secure access control mechanisms for third-party APIs and internal systems.
- Code Interpreter: A sandboxed execution engine for running agent-generated code.
- Browser Tool: A secure interface that allows agents to fetch information from the web.
- Observability: Built-in telemetry, tracing, and debugging for analyzing agent behavior in production.
These features allow small teams and early-stage companies to prototype complex agent behavior quickly. Without building custom infrastructure, developers can test autonomous workflows, secure API access, memory design, and tool integration inside a robust, scalable runtime.
For startups that are already within the AWS ecosystem, AgentCore dramatically reduces the barrier to entry. It enables real-world experimentation with AI-driven features and provides native compatibility with other AWS services, including S3, Lambda, Bedrock models, and Step Functions.
At TGS Tech, we are leveraging AgentCore in the early stages of Apex Engine to experiment with agent-driven automation, live collaboration assistants, and procedural content tools. The fast iteration cycle helps us refine our technical direction before committing resources to a long-term custom framework.
Strategic Considerations for Apex Engine
While AgentCore provides an efficient starting point, the long-term design of Apex Engine requires more flexibility and control than a proprietary service can offer. Apex Engine must serve a wide range of use cases, including:
- Large-scale multiplayer simulation environments
- Interactive architectural walkthroughs and BIM-integrated digital twins
- Real-time collaborative educational spaces
- Smart city infrastructure dashboards
- Custom in-editor tools powered by AI
These use cases often include requirements that fall outside of what AgentCore currently supports, such as:
1. Deployment Flexibility
Many enterprise and government clients require their solutions to run on infrastructure they control. AgentCore is locked to the AWS Bedrock environment and cannot be deployed on Azure, Alibaba Cloud, or private servers.
2. IP and Data Sovereignty
Clients in regulated sectors such as healthcare, defense, or finance must ensure that model inference, logging, and memory are contained within their jurisdiction. Third-party LLMs or cloud-managed components introduce legal and technical risks.
3. Fine-Grained Optimization
Some features of Apex Engine require deep integration between the agent system and the rendering pipeline, physics engine, or editor UI. This level of customization is not possible with managed services like AgentCore.
For these reasons, we have made the strategic decision to build our own AI infrastructure, both to support the advanced functionality Apex Engine demands and to protect our clients’ interests.
Mitigating Risk with Internal LLMs and SLMs
To maintain control over data, behavior, and long-term cost, we are developing and training our own internal Large Language Models (LLMs) and Small Language Models (SLMs). These models will power the agent systems inside Apex Engine without requiring connection to external APIs or hosted inference endpoints.
Benefits:
- Data Privacy: No third-party sees our prompts, outputs, or fine-tuning datasets
- Performance Tuning: We can train and compress models specifically for 3D workflows, editor assistance, and simulation logic
- Offline Support: For edge deployments or air-gapped systems, our models can run locally without external dependencies
- IP Ownership: All model weights, training procedures, and behavior remain entirely proprietary
In addition to internal use, this also provides a strategic advantage for future B2B partnerships and licensing opportunities. Clients can benefit from Apex Engine’s intelligence layer without having to rely on external vendors or accept external compliance risks.
Toolchain Integration and Technical Alignment
Apex Engine is designed from the ground up to function as a modular, extensible real-time 3D development platform, capable of operating as both a creation environment and a runtime framework across industries. This vision requires deep integration with a wide range of modern and legacy toolchains to support rendering, physics, animation, UI/UX, and procedural generation in real-time.
Core Technologies and Middleware
We are actively integrating a series of proven and highly optimized technologies into Apex Engine, including:
- Rendering and Graphics: DirectX 11/12, Vulkan, and cross-platform fallback options
- UI/UX: Qt for tool development and in-editor interfaces, with both visual scripting and native code control
- Physics and Simulation: Havok Physics, Cloth, and Navigation modules, integrated with our internal state system and collision detection layers
- Animation and Rigging: Apex3D for skeletal animation pipelines, customized for live retargeting and editor-time playback
- Terrain and Procedural Generation: Voxel Terrain, integrated into our terrain editor for dynamic world streaming and generation at runtime
Agent-Driven Tooling and Workflows
Our agent framework is designed to interface directly with these toolchains, allowing AI agents to:
- Adjust physics simulation parameters in real-time for training or experimentation
- Generate, edit, or validate UI components via the Qt layer
- Modify shaders, lighting setups, or environmental parameters
- Assist with animation retargeting, state transitions, and character scripting
- Support intelligent procedural worldbuilding and simulation adjustments
Technical Benefits for Developers and Partners
- Deep Interoperability with critical systems
- Modular extension paths for custom pipelines and workflows
- Shared debugging and observability layers across editor and runtime environments
This approach provides a robust foundation for engineering partnerships and long-term scalability across both creative and simulation-focused use cases.
TLDR Summary
- In the August 5, 2025 AWS AI Bedrock AgentCore podcast, AWS provided a thoughtful and informative overview of their modular agent framework. Their solution offers a powerful foundation for building secure, scalable AI agents in the cloud and is an excellent fit for many companies building directly within the AWS ecosystem.
- While AgentCore is ideal for cloud-native teams, it is limited to the AWS environment and does not support multi-cloud, on-premise, or sovereign deployments. It also restricts deep customization and visibility into critical systems such as runtime, memory, and tool orchestration.
- At TGS Tech, we are building Apex Engine, a modular, real-time 3D development platform designed for gaming, simulations, architecture, digital twins, and smart infrastructure. Agentic workflows will play a major role across these industries.
- We are using AWS AgentCore for early-stage prototyping to rapidly validate agent workflows, memory systems, and tool integrations in a secure and efficient way.
- To meet the long-term needs of our clients, we are building our own internal LLMs and SLMs, trained and hosted in-house. These models are optimized for real-time collaboration, procedural workflows, and AI-assisted development inside Apex Engine.
- Apex Engine is being developed as a server- and host-agnostic platform, allowing deployment on AWS, Azure, Alibaba Cloud, private infrastructure, or air-gapped environments. This flexibility supports strict security, compliance, and performance requirements across industries.
- The platform is deeply integrated with key toolchains, including Qt, Vulkan, DirectX, Havok, Voxel Farm, and Granny3D. Our AI agents interface directly with these systems to drive intelligent behaviors such as UI generation, simulation control, procedural content creation, and runtime state management.
- We recognize that building a hybrid, modular, real-time runtime system is a complex challenge. It requires careful abstraction, dual-path development, and sustained AI infrastructure investment. But this approach prevents lock-in, increases IP control, and supports long-term innovation.
- The result is a platform that serves our company, our partners, and our clients with greater freedom, security, and extensibility, while remaining open and aligned with future advancements in agent-based infrastructure.