Persist context beyond one run
Store durable memory that can support later interactions. Keep conversation history in its own tables so past exchanges and reusable memory serve their respective purposes.
ART Datalake / Agent memory
Carry useful context into the next conversation.
A useful interaction should not have to start from zero every time. Give agents a persistent home for memory and conversation history, while keeping control over what remains relevant.
Agent memory in practice
Built for the way your data works.
Keep memory current
Review, update, and delete memory as circumstances change. Version-aware mutations help prevent one update from silently overwriting another.
Be deliberate about what stays
Memory retention is configured through the agent. Pin context that needs to remain; expired, unpinned memory is removed by lifecycle cleanup.
Illustrative workflow
Remember a preference, then keep it accurate
A supplier's preference for delivery updates can be retained as useful context for subsequent interactions. When that preference changes, update the memory rather than continuing to rely on an outdated detail.
- Identify reusable context
- Retain it within scope
- Review, update, or remove it
Understand the lifecycle.
Durable memory is different from temporary tool-result storage. Pinned memories remain until explicitly unpinned or deleted. Conversation-history retention follows its table policy.
A few useful details.
Does every response become permanent memory?
No. Memory behavior is agent-configured. Conversation history, durable memories, and temporary execution results are separate forms of persistence.
Can retained context be removed?
Yes. Memory supports updates and deletion. Unpinned memories can also expire under the agent's configured retention policy.
Start with your workflow
What should your data make possible?
Bring the records, files, and context behind your process. We’ll explore how Datalake fits.