Glossary · Term

context condenser

← all terms

Definition

A separate model that watches an AI agent's running notebook and rewrites it into a clean summary when it gets too full.

A dedicated module in long-horizon agent systems (e.g., QUEST, AgentFugue) that compresses accumulated rollout context into structured state — typically partitioned into verified, contradicted, and uncertain claims with next actions — to keep working memory within budget without losing epistemic state.

Mentioned in 1 episode

  1. 082
    Training a Deep Research Agent on 8,000 Synthetic Tasks: The Rubric Tree Trick