# OpenSquilla发布开源AI代理，主打高性价比长时任务

- 来源：🚨 AI News | TestingCatalog (@testingcatalog)
- 发布时间：2026-05-14 17:25
- AIHOT 分数：70
- AIHOT 链接：https://aihot.virxact.com/items/cmp5api000cpisljxjo1h5wzk
- 原文链接：https://x.com/testingcatalog/status/2054855675847585978

## AI 摘要

OpenSquilla近日推出了一款开源Python AI代理，专为执行成本效益高的长时任务而设计。其核心创新是内容感知模型路由技术，能自动将简单任务分配给廉价模型，复杂任务分配给高性能模型，基准测试显示相比单一模型方案可节省60%-80%的token成本。该代理还集成四层持久化记忆系统，支持本地嵌入以提升效率，并提供不依赖Docker的系统调用级沙箱隔离，确保运行安全。项目采用Apache-2.0许可证开源，强调本地优先和Python基础。为验证成本节省效果，OpenSquilla启动了“1000万Token账单挑战”，邀请用户对比账单，设立奖项包括忠实复现、最佳节省案例和质量错误报告三类，共30名获奖者各获得1000万OpenRouter积分，旨在推动高效AI代理的实践应用。

## 正文

OpenSquilla released a new open-source AI agent designed for executing cost-efficient， long-running tasks. It offers 60-80% token savings and is available under the Apache-2.0 license！

> ML classifier sends simple tasks to cheap models， complex ones to stronger models

> 60-80% token savings vs a single flat-model setup， per OpenSquilla benchmarks

> Four-tier persistent memory with local embeddings

> Syscall-level sandbox isolation without Docker dependency

### 引用推文

> OpenSquilla：Long-running agents shouldn't pay frontier-model prices for every turn. We've been quietly building our agent with content-aware model routing, memory consolida...