Your Agent's Memory Might Be On The Wrong Database

Most agent memory in production sits on the team's database of choice, picked long before agents arrived: an in-memory cache because it was fast, a document store because memories looked like documents, a relational database with a vector extension because it was one command away, or a specialized vector database because the tutorial used one. Each was chosen for one good reason. Agent memory in production needs all of those reasons at once, along with memory that is shared across agents, governed per user and kept for years.

In this session we start with the requirements of an AI database and the requirements of agent memory workloads, then map both to the database capabilities that we will implement and assemble live.

You will leave with:

• A requirements checklist for agent memory that you can score any database against, including the one you run today

• Working notebooks that implement each memory capability from the session, ready to rerun

• An understanding of where the common database choices run out for agent memory, and why

• A reference agent memory implementation on Oracle AI Database, assembled step by step during the session

Back with your team, you will be able to score the database behind your agents against a clear set of requirements and show exactly where it will hold you back as your agentic workloads grow. You will have the code to replace a stack of stitched-together services with a single engine, and the argument for why that engine should be a converged database, with Oracle AI Database as the best contender on the market today.

Duration: 90 minutes