KTern.AI builds an agentic platform for SAP on Amazon Bedrock AgentCore

KTern.AI presented an architecture for an agentic AI platform for SAP built on Amazon Bedrock AgentCore and Strands Agents SDK: the system coordinates multiple specialized agents for long-term enterprise programs while preserving context and controlling access to tools.

How the platform works

KTern.AI is evolving its previous SaaS platform into an agentic system where tasks are not resolved by a single universal chatbot. Instead, specialized agents handle separate parts of the enterprise process. This approach is important for the SAP environment: requests are typically not about a single answer but about a long-term program where previously made decisions, process states, and available enterprise data must be considered.

Amazon Bedrock AgentCore—AWS's managed platform for deploying and operating AI agents—forms the foundation of the infrastructure. To orchestrate agent behavior, KTern.AI uses Strands Agents SDK. Together, these components enable work distribution across agents without building the infrastructure from scratch for state management, tool integration, and reliable service execution.

• KTern.AI uses Amazon Bedrock AgentCore as the infrastructure for an agentic platform oriented toward SAP.
• Strands Agents SDK handles the coordination of specialized AI agents.
• Agents are designed for long-term enterprise programs, not just one-off chat requests.
• The architecture provides persistent context for each agent.
• Access to external tools must remain secure and managed.

Why One Agent Is Not Enough?

A single agent can be a convenient entry point for users, but in an enterprise system, it must simultaneously understand program context, access business tools, execute sequences of actions, and enforce access rules. As the number of scenarios grows, this architecture becomes harder to maintain: it is difficult to separate responsibilities, audit actions, and restrict access to sensitive operations.

KTern.AI distributes work across multiple agents with specific roles. This simplifies orchestration: one agent can be responsible for a separate process stage, while another handles a specific tool or continues an already-started task. As a result, the system can preserve not only chat history but also the working context needed for the next steps in the enterprise program.

Persistent context is particularly critical for SAP implementations and support, where processes rarely conclude with a single request. Agents need to understand what has already been done, which data and constraints apply to a specific client, and what stage the task is at. Without this, every new session becomes a repetition of initial data gathering, and automation quality deteriorates.

What Changed for Operations

Amazon Bedrock AgentCore handles tasks that typically must be implemented within the product itself: running agent components, managing their state, and ensuring reliable operation in production. For KTern.AI, this means the team can focus on agent logic and SAP scenario integration rather than building the entire base platform from scratch.

A separate security layer is necessary because agents do not merely generate text but also access tools. In a corporate environment, such access must be limited, controlled, and tied to specific tasks. In the described architecture, secure tool access is considered a mandatory part of the system, not an afterthought added after launch.

Production reliability also becomes a key requirement. A demo agent can operate in short sessions and be manually handled on errors. For an enterprise platform, this is insufficient: agents must reliably execute long-running processes, preserve state across interactions, and interact correctly with each other.

What This Means

The KTern.AI case demonstrates a practical shift from standalone AI assistants to managed systems of multiple agents. For enterprise platforms, the value of such an architecture lies not in the dialog itself but in the ability to safely conduct long processes, use business tools, and maintain context across work stages.