Databricks Agent Memory Pipelines are in Beta

Memory pipelines extract durable facts, preferences, and decisions from stored conversations and write them to a long-term memory store for your AI agents.

Use cases

Imagine you have a support assistant handling hundreds of customers and needing to recall recurring preferences without cluttering the conversation context. With this feature you can:

• Build long-term memory from conversations without adding memory-writing tools to your agent.
• Backfill history from an existing session store.
• Keep memory organized by consolidating entries and removing duplicates.

How to use

You can create and manage pipelines using the Agent Bricks CLI or the REST API.

agentbricks memory pipeline create \
  --session-store support-agent-sessions \
  --memory-store support-agent-memory \
  --model system.ai.claude-sonnet-5

What happens in this code:

1. The command triggers the creation of a new pipeline.
2. The --session-store parameter defines the source of conversations.
3. The --memory-store parameter defines the destination for distilled memories.
4. Optionally, the --model parameter sets the AI model.

To start a manual run:

agentbricks memory pipeline run <pipeline-id>

What happens in this code:

1. The command initiates the execution of the specified pipeline.
2. The system returns the resource name and the RUNNING state.

What it returns

Field	Description
RUNNING	The run is pending or in progress.
SUCCEEDED	The run processed every actor successfully.
FAILED	The run finished with an error for one or more actors.
CANCELLED	The run was cancelled before it completed.

Limitations and requirements

• The feature is in Beta.
• Requires a session store with conversations and a memory store for writing.
• The identity creating the pipeline needs access permissions to both stores and permission to create tables in the memory store's database.
• To use the CLI, you must install the databricks-agentbricks package version 0.5.0 or higher.
• Serverless general compute pricing and Foundation Model Serving pricing apply to pipeline runs.