Industry story
Amazon Bedrock AgentCore memory, policy, and harness expand to GovCloud
agent-framework guardrails memory orchestration
GovCloud availability matters more than it looks. Most enterprise AI agent frameworks assume you can spin up memory and orchestration infrastructure wherever you like; regulated buyers can't, and that gap has kept serious agentic workloads out of government and defense. Bringing AgentCore's memory, plain-English policy controls compiled into Cedar, and the managed execution harness to GovCloud (US-West) removes that blocker, and the teams building in those environments now have one fewer reason to roll their own.
Full analysis
AWS has brought three Amazon Bedrock AgentCore capabilities to its AWS GovCloud (US-West) region: memory, policy controls, and a managed execution harness, targeting teams building AI agents in regulated environments such as government and defense. AgentCore Memory gives agents both short-term (within a conversation) and long-term (cross-session) memory without requiring teams to build their own memory infrastructure. The Policy feature lets developers write access-control rules in plain English that are automatically compiled into Cedar (AWS's open-source policy language), then enforced at an AgentCore gateway before any tool call executes. The managed harness lets builders declare an agent's model, tools, and instructions via configuration and launch it in a few API calls. AWS handles compute, identity, observability, and orchestration plumbing.
Comments