Parallel Context Compaction for Long-Horizon LLM Agent Serving
\IEEEauthorblockN Musa Cim, Burak Topcu, Chita Das, Mahmut Kandemir \IEEEauthorblockA The Pennsylvania State University {mtc5693, bvt5283, cxd12, mtk2}@psu.edu · 2026
Read Parallel Context Compaction for Long-Horizon LLM Agent Serving by \IEEEauthorblockN Musa Cim, Burak Topcu, Chita Das, Mahmut Kandemir \IEEEauthorblockA The Pennsylvania State University {mtc5693, bvt5283, cxd12, mtk2}@psu.edu (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2605-23296.txt.
Description
Cim and three coauthors present side-by-side shrinking for long-horizon LLM agent serving. Long-horizon agents gather chat logs that exceed the context window. LLM-based digests keep the chat bounded, but digests are lossy and the blocking call stalls work for tens of seconds. Side-by-side shrinking gives the user set control over summary volume and enables better prompt craft per block. At matched decode volume, it cuts end-to-end wall time and improves throughput over the serial base.
Catalog record
| CREATOR | \IEEEauthorblockN Musa Cim, Burak Topcu, Chita Das, Mahmut Kandemir \IEEEauthorblockA The Pennsylvania State University {mtc5693, bvt5283, cxd12, mtk2}@psu.edu |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 43955 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.23296