Skip to content
#

experience-memory

Here are 5 public repositories matching this topic...

Experience memory for LLM agents that act, tested one rung at a time against controls. First finding (ALFWorld, 4B): a token-matched placebo block alone raised the solve rate 13 points, so much of the apparent memory gain was not memory. Lend a 12 GB GPU for a 2-hour cell, or pick a CPU task.

  • Updated Nov 18, 2026
  • Python

Add this topic to your repo

To associate your repository with the experience-memory topic, visit your repo's landing page and select "manage topics."

Learn more