Pai 的 AI 雷达

今日必看

今天这份简报把信息聚在“让代理真正做事”这条主线上：一端是边缘设备上的多步代理基准，以及更可控的编排、记忆和执行工具；另一端是代码协作中的预算、扫描、Rust GPU 内核与版权议题，同时补上多模态创作、工作价值和一批可组合的开源项目。阅读时先用主条目抓住主线，再按行动提示挑能落到工作流、接口、权限边界或本地实验的条目；长列表更适合当作组件清单来拆解。

日期
2026/09/17
今日条目
101
往期2026/09/17
全部101编码代理21Agent与工具11个人助理5模型与评测31创作成片8开发与开源9产品接入16
领域机器人具身1硬件边缘2生信医疗5法律合规1金融3安全治理1行业观察5创业者1

更多资讯

#33

Show HN: Deiko – point at your screen, talk, and hand the brief to your agent

macOS, 0.4.5, free to try: curl -fsSL https://deiko.app/install.sh | sh I kept losing time explaining my own screen to Claude Code. Screenshot, crop, paste, then type three paragraphs describing which dropdown I meant. The context was all right in front of me and I was transcribi

推荐理由在 macOS 上用 Deiko 0.4.5，把“指向屏幕并交给代理”作为交互步骤，减少截图、裁剪、粘贴和长段描述某个下拉框的往返；当前版本可免费试用。
编码代理HN MCP/Claude Code/agent
deiko.app打开原文
#43

Show HN: A Coding Agent from Scratch

I saw a post about building a coding agent from "first principles" and that inspired me to create my own. I used my own custom Rust-based ML DSL to implement this. The blog post includes link to the repos (MIT Licensed), a video of the agent in action, sample code produced, and a

推荐理由从“第一性原理”出发搭一个编码代理，可以用自定义 Rust ML DSL 实现，并通过博客中的仓库、MIT 许可、演示视频和样例代码拆解实现与产物。
编码代理HN MCP/Claude Code/agent
blog.softwarewrighter.com打开原文
#46

SCIM user responses now include a profileUrl attribute

SCIM user responses from GitHub now include the standard profileUrl attribute defined by RFC 7643, containing the absolute URL of the GitHub account linked to the external identity. Previously, matching a SCIM record to the corresponding GitHub account required extra lookups or i

推荐理由处理 GitHub 的 SCIM 用户响应时，直接读取 RFC 7643 定义的 profileUrl，用绝对 URL 对应外部身份与 GitHub 账户；这样可减少为匹配账户而进行的额外查找。
产品接入GitHub Changelog
github.blog打开原文
#48

ANASSA: An Agentic AI Orchestration Framework for Spatial Intelligence

arXiv:2609.14824v1 Announce Type: new Abstract: The emergence of large language models (LLMs) and large multimodal models (LMMs) has enabled a new class of agentic systems capable of integrating natural language understanding with tool-based execution. In geographic information s

推荐理由ANASSA: An Agentic AI Orchestration：arXiv:2609.14824v1 Announce Type: new Abstract: The emergence of large language models。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#49

AcquireBound: Runtime Authorization for Resources Acquired by AI Agents

arXiv:2609.14744v1 Announce Type: new Abstract: By acquiring compute, credentials, accounts, services, and other agents, autonomous AI agents can introduce new authority into a task. Payment, budget, OAuth, mandate, and fulfillment checks can validate transaction conditions witho

推荐理由AcquireBound: Runtime Authorization：arXiv:2609.14744v1 Announce Type: new Abstract: By acquiring compute, credentials。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#50

Assessing the Applicability of Existing Design Recommendations to AI Companion Design: A Multi-Method Study

arXiv:2609.14236v1 Announce Type: cross Abstract: With the rapid proliferation of large language model (LLM)-based systems, AI companions have emerged as conversational agents designed to cultivate emotional connection rather than primarily to support humans in instrumental tasks

推荐理由Assessing the Applicability of Existing：arXiv:2609.14236v1 Announce Type: cross Abstract: With the rapid proliferation of large。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#51

Automating Attack Graph Construction for Agentic Pentesting. Towards Neuro-Symbolic Vulnerability Hunting

arXiv:2609.15523v1 Announce Type: cross Abstract: Logic attack graphs grounded in scanner output provide explicit and auditable attack path reasoning LLM-based agents lack. Integrating symbolic frameworks such as MulVAL to contemporary security workflows or agentic pipelines, how

推荐理由Automating Attack Graph Construction：arXiv:2609.15523v1 Announce Type: cross Abstract: Logic attack graphs grounded。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#52

Bridging Thought and Action: Taming Long-Horizon Instability in Open-Source LLM Agents with a MetaTool-Enhanced ROS Framework

arXiv:2609.13335v1 Announce Type: cross Abstract: Large Language Models (LLMs) have enabled more natural human-robot interaction, but open-source models often exhibit unstable long-horizon reasoning and inefficient action execution when deployed in agentic robotic frameworks. Thi

推荐理由Bridging Thought and Action: Taming：arXiv:2609.13335v1 Announce Type: cross Abstract: Large Language Models (LLMs) have。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#53

CIDERS: Cloud-Edge LLM Collaborative Learning via Accelerating Personalized Bilevel Optimization

arXiv:2609.15664v1 Announce Type: cross Abstract: Amid the rapid advancement of physical-world intelligence, cloud-edge collaborative large language models (LLMs) have emerged as a promising roadmap for practical LLM deployment. However, existing cloud-edge paradigms struggle to

推荐理由CIDERS: Cloud-Edge LLM Collaborative：arXiv:2609.15664v1 Announce Type: cross Abstract: Amid the rapid advancement。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文
#54

Carbon-Aware Routing for Function Calling in Edge-Cloud LLM Systems

arXiv:2609.13559v1 Announce Type: new Abstract: Large Language Models (LLMs) with function-calling capabilities are becoming critical for modern agentic AI systems. Nevertheless, current deployments typically route inferences to powerful cloud-based models, incurring significant

推荐理由Carbon-Aware Routing for Function：arXiv:2609.13559v1 Announce Type: new Abstract: Large Language Models (LLMs)。
模型与评测arXiv cs.AI 官方
arxiv.org打开原文