AgentCore Runtime¶
Work in Progress
This page is currently under development.
Overview¶
Amazon Bedrock AgentCore Runtime is a managed service that enables running containerized AI agents with extended execution times, enhanced payloads, and isolated environments.
Key Features¶
Extended Execution¶
- Long-Running Sessions: Up to 8 hours of execution time
- Complex Reasoning: Support for multi-step, sophisticated reasoning chains
- Multi-Agent Collaboration: Communication between agent instances
- Stateful Processing: Maintain context across multiple invocations
Enhanced Payloads¶
- Large Content Support: Up to 100MB payload size
- Multi-Modal Input: Support for text, images, audio, and video
- Structured Data: JSON, XML, and binary data handling
- File Processing: Document analysis and processing capabilities
Session Isolation¶
- Dedicated microVMs: Isolated execution environment for each agent
- Resource Allocation: Configurable CPU and memory resources
- Filesystem Isolation: Separate disk storage per session
- Network Controls: Configurable network access policies
Framework Compatibility¶
AgentCore Runtime works with popular agent frameworks:
- LangGraph: Directed graphs for complex reasoning flows
- CrewAI: Multi-agent collaborative frameworks
- Strands Agents: Structured reasoning agents
- Custom Implementations: Any containerized agent system
Built-in Services¶
- Authentication: Integration with identity providers
- Observability: Tracing, logging, and metrics
- Tool Access: Browser automation, code interpretation
- Memory Management: Short and long-term memory systems
Deployment Model¶
Details on how AgentCore CLI deploys agents to Amazon Bedrock AgentCore Runtime will be added in a future update.
Best Practices¶
Recommended patterns for designing and implementing agents for AgentCore Runtime will be covered in a future update.