Source: Oracle Developers Blog — 2026-07-07
Summary
Oracle shipped a round of updates to its AI Agent Memory product inside Oracle AI Database, focused on practical developer controls: faster extraction of memories from conversation history, hybrid (vector plus keyword) search over stored memories, custom extraction rules, "context cards" for surfacing relevant memory at a glance, and lifecycle controls governing what gets stored, retrieved, updated, or expired.
Key Takeaways
- Adds hybrid search, combining vector similarity with keyword search over an agent's stored memories.
- Introduces custom extraction, letting developers define their own rules for what gets pulled out of conversation history into memory.
- Adds "context cards" and expanded lifecycle controls for what memory content is injected back into an agent's context window.
- Builds on Oracle AI Agent Memory 26.6 ("Memory With Receipts"), Oracle's governed, unified memory core for enterprise AI agents running on Oracle AI Database.