Managed agent memory and sessions in Databricks

Use cases

Imagine you are building an AI assistant that needs to remember user preferences across different conversations and keep a durable history of interactions. You can use this feature to:

• Store persistent long-term memory with semantic search.
• Keep durable conversation session history.
• Integrate with agents built on any framework.

How to use

import os
from databricks_agents import AgentMemory

memory = AgentMemory()
memory.add_memory(
    agent_id="my-agent",
    user_id="user-123",
    content="The user prefers short answers in Portuguese."
)

What happens in this code:

1. Imports the Databricks agents library.
2. Initializes the agent memory manager.
3. Adds a durable fact associated with the user for future semantic search.

Limitations and requirements

• The feature is in Beta.
• Backed by managed Lakebase (Postgres) storage.
• Compatible with agents built on any framework.