arXiv cs.CL· Shai Nahum Gefen, Vered Chassida, Dmitry Basin, Guy Feigenblat·· 11 小时前AI 评分39
CLEO：为自托管 LLM 拟合最小化智能体框架，而非优化模型本身
Optimize the Harness, Not the Model: Fitting a Minimal Agent to a Self-Hostable LLM for Investigative Analysis
AI 导读
CLEO 是一种为固定的自托管 LLM 拟合最小化智能体框架的方法，只保留能提升准确率且保持通用、可审计的工具、护栏、记忆策略与循环组件。在留出数据集上，CLEO 框架通过率达 83.8%，与最强的开箱即用 Claude Code SDK 基线相差 1.2 个百分点，而规模小两个数量级。
正文 · 原文
Abstract:Government intelligence teams need LLM agents that can analyze large, messy evidence collections, but security and cost constraints often rule out hosted frontier models and broad, hard-to-audit tool surfaces. We report CLEO: Constrained Loop for Environment Optimization, a method for fitting a minimal agent harness to a fixed, self-hostable LLM. Given an expert-validated dataset, CLEO admits only the tools, guardrails, memory policies, and loop components that improve accuracy while remaining general and auditable. On our held-out dataset, the CLEO harness reaches an 83.8% pass rate, within 1.2 points of the strongest out-of-the-box Claude Code SDK baseline, while being two orders of magnitude smaller.
| Comments: | 3 pages, 1 table. Industry Day talk at CIKM 2026, Rome, 7 November 2026 |
| Subjects: | Artificial Intelligence (cs.AI); Computation and Language (cs.CL); Multiagent Systems (cs.MA) |
| Cite as: | arXiv:2611.03434 [cs.AI] |
| (or arXiv:2611.03434v1 [cs.AI] for this version) | |
| https://doi.org/10.48550/arXiv.2611.03434 arXiv-issued DOI via DataCite (pending registration) |
|
| Related DOI: | https://doi.org/10.1145/3799682.3839256
DOI(s) linking to related resources |
Submission history
From: Shai Nahum-Gefen [view email]
[v1]
Mon, 2 Nov 2026 16:53:29 UTC (7 KB)
来源：arXiv cs.CL · arxiv.org