A compression layer for LLM context — it squeezes tool outputs, logs, and RAG chunks down by 60-95% before they ever reach the model, without changing the answers. 专为大模型上下文设计的压缩层，把工具输出、日志和 RAG 片段在送入模型前削减 60-95% 的体积，答案质量却不受影响。
Google Research's pretrained foundation model for time-series forecasting — zero-shot predictions across domains without training a model from scratch. 谷歌研究院推出的时间序列预测基础模型，支持零样本跨领域预测，无需为每个场景重新训练。
A battle-tested methodology for turning loose agent instructions into composable, reusable skills — already pulling in 1,100+ stars in a single day. 一套把零散的智能体指令打磨成可复用技能的方法论，单日狂揽 1100+ 星标，实战检验颇受好评。
An MCP server that turns an entire codebase into a persistent knowledge graph — 158 languages, sub-millisecond queries, and 99% fewer tokens spent on context. 一个把整个代码库构建成持久知识图谱的 MCP 服务器，支持 158 种语言、亚毫秒级查询，上下文 token 消耗减少 99%。
A macOS video editor designed from the ground up for AI-assisted workflows — built in Swift, and already pulling 750+ stars on day one. 一款从底层为 AI 协同剪辑而生的 macOS 视频编辑器，用 Swift 打造，首日狂揽 750+ 星标。
Zhipu AI's GLM-5 marks the shift from vibe-coding demos to genuinely agentic engineering — a model built to plan and act, not just autocomplete. 智谱 AI 的 GLM-5 标志着从"氛围编程"演示走向真正的智能体工程，模型不仅能补全代码，更能规划与执行任务。
The Astro team's new sandbox framework for running AI coding agents safely — isolated execution without sacrificing developer ergonomics. Astro 团队推出的 AI 编码智能体沙盒框架，让隔离执行与开发体验两者兼得。
A networking stack that swaps brittle IP addresses for cryptographic dial keys — peer-to-peer connectivity that just works behind NATs and firewalls. 用密钥拨号取代脆弱的 IP 地址寻址，这套 Rust 网络栈让点对点连接在 NAT 和防火墙背后也能即连即用。
Kong's open-source API client covers GraphQL, REST, WebSockets, SSE and gRPC in one tool — with cloud, local, or Git-based storage for your collections. Kong 出品的开源 API 客户端一站式覆盖 GraphQL、REST、WebSocket、SSE 与 gRPC，集合存储支持云端、本地或 Git 多种方式。
The official inference and LoRA training package for Lightricks' LTX-2 — a generative model that produces synchronized audio and video together. Lightricks LTX-2 的官方推理与 LoRA 训练工具包，这一生成模型能够同步产出音频与视频内容。