Amazon Bedrock AgentCore now offers runtime instances, a new persistent compute option built on managed EC2 infrastructure for production AI agents. Unlike the existing microVM option (limited to 8-hour invocations), runtime instances support sessions lasting up to 14 days, GPU acceleration, shared file systems for multi-agent collaboration, and session hibernate/resume. Multiple agents can run on the same host, share a session directory, and collaborate without direct API calls between them. The service supports Python 3.11–3.14, container images, and popular frameworks like CrewAI, LangGraph, LlamaIndex, and Strands. Pricing is standard EC2 rates plus an AgentCore management fee, and the feature is available in multiple AWS regions.
Questions this post answers
What is the maximum session duration for Amazon Bedrock AgentCore runtime instances?
Runtime instances support shared sessions that persist for up to 14 days. This contrasts with AgentCore Runtime microVMs, which support invocations lasting up to 8 hours. Sessions can also be hibernated and resumed, so agents can stop overnight and pick up where they left off without losing state. Teams running long-horizon agent workflows track infrastructure options like these on daily.dev.
How do multiple agents collaborate in Amazon Bedrock AgentCore runtime instances without calling each other's APIs?
Agents deployed on the same runtime instance share a session-scoped file system directory. One agent writes output to a path like /tmp/agentcore-session/{session-id}/file, and another agent reads from the same path using the same session ID — no inter-agent API calls or message passing required. Any number of agents can share the same working directory within a session. Developers designing multi-agent architectures find patterns like this on daily.dev.
What Python versions and frameworks does Amazon Bedrock AgentCore runtime instances support?
Runtime instances support Python 3.11 through 3.14 with native code support, plus container images for teams that need custom environments. Supported agent frameworks include CrewAI, LangGraph, LlamaIndex, and Strands. Packaging requires only an @app.entrypoint decorator and a zip file or container image. Developers choosing an agent framework for AWS deployments can compare options on daily.dev.