Memory and Feedback Loops

Memory and Feedback Loops

Learning from mistakes within a session.

Memory and Feedback Loops

An agent that repeats the same mistake is useless.

The Scratchpad

Your state should include a scratchpad or critique field. When a node runs, it append its thoughts.

  • Iter 1: "I tried searching for 'Apple' but got fruit results. I need to search for 'Apple Inc'."
  • Iter 2: The agent sees this note and adjusts.

This "Short Term Memory" allows the agent to navigate a problem space without getting stuck in a local minimum.

Subscribe to our newsletter

Get the latest posts delivered right to your inbox.

Subscribe on LinkedIn