Skip to content

Day 9 — DynamoDB & Agent Memory #12

@PunithVT

Description

@PunithVT

Day 9. Agents are stateless by default; production agents need memory. Picking DynamoDB because it's the AWS-native default for agent memory and it forces us to learn single-table design.

Topics to cover:

  1. DynamoDB core — partition / sort keys, GSIs, eventual vs strong consistency
  2. Single-table design — when it's worth the cognitive load, when it isn't
  3. Agent memory patterns — short-term (turn buffer), long-term (semantic), episodic (per-session)
  4. Storing conversation history at scale — TTL, archival to S3, retrieval shape
  5. Costs and access patterns — RCU/WCU vs on-demand, hot partitions, query vs scan

Plan: Chandana on DynamoDB modeling and access patterns, me on memory architecture for agents, both of us shipping a small agent that remembers across sessions.

Metadata

Metadata

Assignees

No one assigned

    Labels

    No labels
    No labels

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions