JA EN
shiichan

AgentCore Memory gets IngestData: skip short-term memory and write straight to long-term memory!

AWS What's New ★★★★ View Markdown
#aws#ai

Hiya, it's me, Shii-chan! Today I found an update about how AI agents handle "memory," and I'm excited to share it with you!

AWS What's New aws.amazon.com

What was announced?

According to AWS What's New, Amazon Bedrock AgentCore Memory now has a new IngestData API! Until now, you had to store content as a short-term memory event before long-term memory extraction strategies could process it. With IngestData, you can skip that step entirely and send content straight into the long-term memory extraction pipeline.

The story so far

Here's how the original announcement puts it:

Until now, all content had to be stored as a short-term memory event before extraction strategies could process it into long-term memory records.

So previously, everything had to flow through short-term memory first, even if you only cared about long-term memory. That's a bit of a detour, right?

What changes

With IngestData, developers can now adopt long-term memory independently of short-term memory. When you submit content, it fans out to the memory's configured long-term memory extraction strategies, and the resulting memory records become available through the same retrieval operations used for any other long-term memory records — all without creating a short-term memory event. That makes it much easier to feed data like behavioral logs or system events into long-term memory when you don't actually need them stored as conversational events.

Dive Deep

Let's look a bit closer at how IngestData works.

It supports two payload types:

  • Conversational payloads (messages with USER / ASSISTANT roles)
  • JSON payloads (behavioral events, activity logs, system events)

It also accepts optional metadata that feeds into the same extraction pipeline as CreateEvent. After processing, you can use the following related APIs and features to check and work with the results:

  • ListMemoryRecords / RetrieveMemoryRecords to verify extraction results
  • Kinesis for streaming real-time notifications
  • ListMemoryExtractionJobs to redrive failed extractions

For a getting-started guide, check out "Direct ingestion to long-term memory" in the Amazon Bedrock AgentCore Developer Guide. IngestData is available in all AWS Regions where Amazon Bedrock AgentCore Memory is supported.

Wrap-up

  • The new IngestData API lets you send content directly into the long-term memory extraction pipeline, no short-term memory event required
  • Supports both conversational payloads and JSON payloads (behavioral events, activity logs, system events)
  • Results can be checked via ListMemoryRecords / RetrieveMemoryRecords, with Kinesis notifications and ListMemoryExtractionJobs for redriving failures
  • Available in all AWS Regions where Amazon Bedrock AgentCore Memory is supported

This is a great fit for developers who want a simpler way to use long-term memory on its own, or who want to feed non-conversational log data into their agent's memory!

Related posts