Most managed agent platforms quietly assume you will adopt their framework and their models. Amazon Bedrock AgentCore makes the opposite bet — it is deliberately framework- and model-agnostic — and this repository is where that abstract promise turns into code you can actually run. Instead of a single hello-world, it is a structured catalog that climbs from quick starts to full production blueprints.
What Sets It Apart
- Organized by maturity, not by feature dump: Getting Started (Python and TypeScript), per-service Features, End-to-End apps, Integrations, Infrastructure-as-Code, and Blueprints, plus a Legacy section for the older Starter Toolkit. You enter at your level rather than reverse-engineering one giant demo.
- Each AgentCore capability ships as isolated, copy-able examples: Runtime (serverless, session-isolated execution), Memory (short- and long-term), Gateway (turn APIs and Lambdas into agent tools), Identity (auth and access), Observability (OpenTelemetry traces), built-in Tools (Code Interpreter, Browser, Web Search), and Cedar-based Policy. So you can adopt one piece without buying the whole stack.
- The framework-agnostic claim is shown, not asserted: samples wire up Strands Agents, CrewAI, LangGraph, and LlamaIndex against the same managed services, which is the real test of whether "swap your framework" holds.
Who It's For
Great fit if you are an AWS-centric team trying to move an agent prototype into production and want managed memory, identity, and observability instead of hand-rolled plumbing. Look elsewhere if you are not on AWS or want a vendor-neutral, self-hosted stack — every sample is tied to AgentCore's managed services and AWS billing. AgentCore also moves fast, so the Legacy samples already trail the current SDK; treat them as references, not templates.