AgentIR: A Workload-Adaptive Cascade Retrieval Substrate for Long-Term Conversational Memory CCS: Information systems Retrieval efficiency CCS: Computing methodologies Parallel computing methodologies
Aojie Yuan, Haiyue Zhang, Shahin Nazarian · 2026
Read AgentIR: A Workload-Adaptive Cascade Retrieval Substrate for Long-Term Conversational Memory CCS: Information systems Retrieval efficiency CCS: Computing methodologies Parallel computing methodologies by Aojie Yuan, Haiyue Zhang, Shahin Nazarian (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-2605-25092.txt.
Introduction
Aojie Yuan, Haiyue Zhang, and Shahin Nazarian build AgentIR, a retrieval system tuned for chat agents that must recall long conversations under tight time limits. For each query it chooses among fusion methods. It also decides, from the cheap BM25 ranking alone, whether a slower dense search is worth running. On one benchmark it skips 63 percent of dense searches without losing judged accuracy. On another it skips them all, lifting shared-server capacity from about 154 to roughly 1,400 agents.
Catalog record
| CREATOR | Aojie Yuan, Haiyue Zhang, Shahin Nazarian |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 145223 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2605.25092