What’s New in Oracle AI Agent Memory Release 26.8

This article highlights the key changes in Oracle AI Agent Memory Release 26.8.

Features and Improvements

• End-user memory isolation with Deep Data Security

  Oracle AI Agent Memory now integrates with Oracle Deep Data Security to control which signed-in user can access each memory. This enables applications to enforce user-level memory isolation directly in Oracle AI Database.

  For more information, see Enforce end-user memory isolation with Deep Data Security and Deep Data Security API Reference.

• Images and multimodal messages

  Oracle AI Database-backed memory stores can now store images and use them in search and automatic memory extraction. This enables agents to remember and retrieve information from both text and images.

  For more information, see Use Images and Multimodal Messages and Records API Reference.

• Memory links and graph-expanded search

  Memories can now be linked to related memories. Search can follow these links to include additional related information, and automatic memory extraction can create links between related memories.

  For more information, see Create and Search Linked Memory Graphs and Records API Reference.

• Configurable search post-processing

  Search results can now be refined after retrieval. Applications can limit the number of results, reorder them with reranking, or use an LLM to prune less relevant results.

  For more information, see How To Improve Search Result Relevance and Search API Reference.

• Record listing APIs

  New APIs are available for listing stored threads, users, agents, messages, and memories. Results can also be filtered or limited to make it easier to find specific records.

  For more information, see Agent Memory API Reference.

• Observability events and listeners

  Applications can now listen for events from Agent Memory operations, including calls to memory stores, embedders, and LLMs. This makes it easier to monitor Agent Memory activity and troubleshoot application responses.

  For more information, see Frequently Asked Questions (FAQs) and Troubleshooting.

Deprecations and Compatibility Changes

• Automatic extraction now defaults to background mode

  Automatic memory extraction now runs in background mode by default. This allows add_messages() to return sooner, but extracted memories might not be available immediately.

  Applications that need extraction to finish before continuing should call wait_for_memory_extraction().

  For more information, see Customize Automatic Memory Extraction and Threads API Reference.

• Python 3.14 is now supported

  The SDK now supports Python 3.14. Supported Python versions are now 3.10 through 3.14.

  For more information, see Frequently Asked Questions (FAQs) and Troubleshooting.

• Text-named store write parameters

  The text parameter on store update() and update_async() is deprecated; use content instead. The texts constructor parameter on PendingRecordBatch is also deprecated; use contents instead.

  For more information, see Stores and Schema API Reference.