In LLM agent trajectories, delaying KV cache compaction by a few turns and using the agent's later queries as the compression guide preserves most accuracy while cutting KV cache by 80%.
You MUST call` get_document`AT LEAST 5 times across the investigation to read full document content (snippets are truncated at ~512 tokens and the answer often sits past them)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Practical Online KV Cache Compaction for LLM Agents: An Empirical Study
In LLM agent trajectories, delaying KV cache compaction by a few turns and using the agent's later queries as the compression guide preserves most accuracy while cutting KV cache by 80%.