AgentCore Memory adds direct ingestion without short-term events

AWS announced on September 8, 2026, that Amazon Bedrock AgentCore Memory can now extract long-term memories from submitted content without first saving it as a short-term memory event. The new IngestData API removes that prerequisite, according to the AWS announcement.

What changes in the ingestion path

Previously, content had to enter short-term event storage before configured extraction strategies could produce long-term memory records. IngestData instead sends submitted content to the memory’s configured long-term strategies, while the resulting records remain accessible through existing retrieval operations.

The API accepts conversational messages with USER and ASSISTANT roles, as well as JSON payloads such as behavioral events, activity logs and system events. Optional metadata enters the same extraction pipeline used by CreateEvent. This is a new ingestion path, not a separate retrieval model.

Availability and integration considerations

Available now: AWS says IngestData is available in all AWS Regions where Amazon Bedrock AgentCore Memory is supported.

For applications that need long-term memory records but do not need an accompanying short-term event, the change removes a required storage step. It also makes the distinction between submitted content and extracted records important: skipping short-term event creation does not establish broader retention or deletion guarantees.

AWS identifies ListMemoryRecords and RetrieveMemoryRecords as ways to check extraction results after processing. It also describes real-time notifications through Kinesis and redriving failed extractions with ListMemoryExtractionJobs.

An integration should therefore track extraction outcomes rather than treat submission as proof that useful memories already exist. The announcement does not specify extraction latency or pricing, so those remain separate checks before choosing this path for a production workload.