durable memory
A system where AI agents can store and recall information across multiple sessions, enhancing their ability to maintain context over time.
Learn
When to use it
Use durable memory when an AI agent needs to retain information across multiple interactions, beyond a single session's context. Durable memory systems combine long-term storage and retrieval capabilities to enable agents to perform tasks like remembering user preferences or tracking ongoing projects.
Quick example
In Microsoft's integration with Azure Cosmos DB, AI agents can store session data persistently, allowing them to recall information from previous interactions. This setup makes Azure Cosmos DB an instance of durable memory, where the database acts as the long-term storage layer for the agent's memory. By using this system, agents can maintain continuity across sessions, improving their ability to handle tasks that require historical context.
prompt → durable memory → recall → update → stop
Ecosystem
Durable memory interacts with various components to maintain and retrieve context. It sits alongside the model and tools, providing a persistent storage layer that agents can access as needed.
┌─ tools ─┐
prompt →│ durable memory │→ model → stop
└─ recall ──┘
Misconceptions
| Misconception | Rebuttal |
|---|---|
| Durable memory is just a large cache | It persists data beyond session limits |
| Any database can be durable memory | It requires specific integration for context recall |
| It automatically manages context relevance | Developers must define relevance and lifecycle policies |
Trade-offs
- Context continuity — requires careful memory management
- Improved personalization — demands robust data privacy measures
- Session persistence — can increase storage costs