开源
开源项目推荐、技术周刊与开发者博客:HelloGitHub、Python官方、JavaScript Weekly、OSChina 等。
排序
标签

端侧 AI Agent 利器:Swift 原生框架 Swarm
Swarm 是一个 Swift 原生的 AI Agent 运行时框架,提供类型安全的工具定义、Apple Foundation Models 设备端推理、可组合工作流、记忆管理与护栏机制。基于 Swift 6.2 严格并发模型构建,支持 iOS、macOS、watchOS 与 Linux,用同一套 Agent 循环贯穿端侧与云端,帮助苹果生态开发者构建离线优先的生产级智能体应用。

Python Language Summit 2026
The 2026 Python Language Summit was hosted in Kraków, Poland as part of EuroPython 2026. There were 15 talks covering free-threading, Rust, garbage collection, type annotations, and more.

Lightning Talks (Python Language Summit 2026)
Lightning talks on a one-time ABI break, safer interruptions, EktuPy (Scratch but Python), an AGENTS.md file for CPython, and a call to read PEP 836.

PEP 827: Type Manipulation (Python Language Summit 2026)
Michael J. Sullivan presents PEP 827 and discusses a key design decision: how to store type annotations?

Free-Threaded Python Post-Era (Python Language Summit 2026)
Tobias Wrigstad, Fridtjof Stoldt, and Donghee Na propose a safe and performant, high-level concurrency model for free-threaded Python

Developer-in-Residence Update & Future (Python Language Summit 2026)
Petr Viktorin gives an update on the Developer-in-Residence role and asks Python core developers for projects to prioritize

Memory Snapshots for CPython (Python Language Summit 2026)
Hood Chatham proposes memory snapshots and an initialization phase for speedier Python startups

Memory Buffer Protocol (Python Language Summit 2026)
Nathan Goldbaum proposes safe concurrent access through buffer leases and custom data types for the Python Buffer Protocol.

Garbage Collection: Generational? Incremental? Both! (Python Language Summit 2026)
Mark Shannon proposes a future garbage collection strategy for Python following the revert of the incremental garbage collector in Python 3.14

One namespace to namespace them all (Python Language Summit 2026)
Pablo Galindo Salgado proposes a top-level `std` namespace for the Python standard library to prevent module shadowing and free up module names.

Anthropic 红队盯上 GLM-5.3:能力追平 Mythos,却连拒绝护栏都是摆设
Anthropic 的前沿红色团队(Frontier Red Team)盯上了智谱的 GLM-5.3,结论有点重:这是第一个能像五个月前的 Claude Mythos Preview 那样自主构建端到端漏洞利用、却几乎光着身子就被放出来的模型。 五个月前,Anthropic 发布 Claude Mythos Preview,自称是第一个能自主构建复杂端到端漏洞利用的 AI 模型。当时他们判...

...
...
加载中...
已经到底啦