Skip to main content

​
Agent Memory Retrieval Starter

​
Summary

This starter shows one clean way to separate active notes, working memory, retrieval inputs, and durable artifacts in a small agent loop.

​
Status

starter

​
Why It Exists

Memory examples often collapse everything into one vague history object. This starter is intentionally narrower: it highlights the boundary between what the agent is currently holding, what it can retrieve, and what it should preserve as an artifact.

​
Folder Structure

agent-memory-retrieval-starter/
├── index.mdx
├── SOURCE_NOTES.mdx
└── src/
    ├── artifact_policy.py
    ├── memory_flow.py
    └── retrieval_trace.py

​
Quick Start

This is a starter, not a runnable application. Read src/memory_flow.py for the minimal state shape and use it as a base for a fuller example later. For a repo-level smoke check, run python3 scripts/verify_example_projects.py from the repository root.

​
Included Sample Files

  • src/memory_flow.py: the smallest useful state container for active notes, retrieval inputs, and durable artifacts
  • src/retrieval_trace.py: a tiny ranking and trace surface for making retrieval decisions inspectable
  • src/artifact_policy.py: one place to show how a starter can separate artifact-promotion rules from raw note capture

​
Constraints

  • No storage backend is wired yet.
  • No embedding or vector-store integration is included.
  • The example favors clear boundaries over completeness.

​
Next Steps

  • Add a simple persistence layer.
  • Add an artifact promotion rule and a retrieval trace.