Getting Started
A short guide to thinking about memory in AI products.
Start with the user journey
Before choosing a memory tool, define what your AI system should remember and why. Good memory is not unlimited storage. It is selective context that improves future interaction.
Common memory needs include:
- User preferences and profile facts.
- Prior decisions, tasks, and project context.
- Conversation summaries and thread state.
- Domain knowledge retrieved at the right moment.
Choose the memory shape
Most products need more than one memory shape:
- Working memory for the current task.
- Episodic memory for prior events and conversations.
- Semantic memory for facts and stable knowledge.
- Procedural memory for learned behavior and workflows.
Evaluate tools
When comparing tools, look for ingestion controls, retrieval quality, data ownership, observability, deletion support, and how easily the memory layer fits your agent or application framework.