The Runtime Decision Engine

This explains the AgentCore Runtime decision engine — how the agent decides what to use and why. The runtime doesn't just react; it plans, decides what it needs, gets it from the right place, then generates the best answer.

How it works

1. 1The user asks a question; a planner analyses intent and context.
2. 2Can I answer directly? If yes, use existing knowledge and context.
3. 3Need customer info? Use the gateway for real-time/enterprise data.
4. 4Need documentation? Search the knowledge base. Need past conversation? Retrieve from memory.
5. 5Synthesize all retrieved information to generate the best possible answer.

Key points

• Smarter decisions: the agent only calls external systems when it actually needs to, saving time and cost.
• Accurate answers: it combines real-time data, trusted knowledge and past context.
• Efficient and scalable: fewer unnecessary calls, better latency.
• The runtime doesn't just react — it thinks, plans and fetches from the right place.

Why it matters

The decision engine is what makes AgentCore efficient and accurate — it fetches only what's needed, when it's needed, instead of blindly calling every system.

Frequently asked questions

How does it avoid unnecessary calls?

It first checks if it can answer directly, only reaching for tools, knowledge or memory when the question requires them.

What can it draw on?

Real-time data via the gateway, verified facts from the knowledge base, and past context from memory.