OPEN SOURCE OBSERVER

逛逛 GitHub

从项目热度走向工程判断:追踪开源工具、模型与开发者生态。

12热门项目
15热门模型
+11,311今日新增 Star

01 / CODE

GitHub Trending

排名项目语言今日新增总 Star
01tester-army/e2eNext generation e2e testing framework for web and mobile apps.TypeScript+1,7206,172
02sponsors/mattpocockSkills for Real Engineers. Straight from my .agents directory.Shell+972278,035
03earthtojake/text-to-cadGive your agent CAD superpowers.Python+62017,915
04boykopovar/AnyPS5Tool for automatic PS5 executables porting to Linux and WindowsC+++9436,283
05pbakaus/impeccableThe design language that makes your AI harness better at design.JavaScript+60977,622
06sponsors/thedotmackPersistent Context Across Sessions for Every Agent – Captures everything your agent does during sessions, compresses it with AI, and injects relevant context back into future sessions. Works with Claude Code, OpenClaw, Codex, Gemini, Hermes, Copilot, OpenCode + MoreTypeScript+53697,101
07ayghri/i-have-adhdA skill to stop your coding agent from burying the answer. ADHD-friendly output.Python+31854,336
08morluto/reaReverse engineer anything with agents, from app behavior down to native binaries.TypeScript+2,9638,801
09deepseek-ai/DeepGEMMDeepGEMM: clean and efficient BLAS kernel library on GPUCuda+3638,665
10sponsors/msitarzewskiA complete AI agency at your fingertips - From frontend wizards to Reddit community ninjas, from whimsy injectors to reality checkers. Each agent is a specialized expert with personality, processes, and proven deliverables.Shell+621157,757

02 / MODELS

ModelScope 热门模型

排名模型任务收藏下载
01Z-Image-Turbo文本生成图片2,565837,531
02MiniMax-H3—2,056194,871
03Qwen3.8-27B视觉多模态理解1,564612,837
04DeepSeek-R1—1,399927,457
05Qwen3.8-Flash-Next视觉多模态理解1,35378,603
06Qwen-Image-2.1文本生成图片1,21042,596
07Qwen3-Coder-480B-A35B-Instruct文本生成1,100138,536
08IndexTTS-2语音合成998212,311
09chatglm3-6b—8321,591,143
10Qwen3.8-2.4T-A95B文本生成7812,149

DEVELOPER VIEW

开发者视角

性能:DeepGEMM 是 DeepSeek 放出的 GPU BLAS kernel 库,关心训练/推理吞吐的人可以直接读实现;Qwen3-Coder-480B-A35B 的 "A35B" 说明激活参数远小于总参数,部署成本相对可控。

免费:榜单上的模型基本都可直接下载,视觉理解、文生图、TTS 三条线都有对应选择,个人开发者不必从零训练。

新范式:i-have-adhd 代表"输出格式约束"这一类技能——让 Agent 别把答案埋起来;thedotmack 的跨会话记忆则把上下文管理变成工程资产。AnyPS5 属于更激进的方向:用自动化补齐平台生态缺口。

WORTH A STAR

本周值得 Star

1. rea:今日涨星最高,把逆向工程(从应用行为一路下探到原生二进制)交给 Agent,是少数能直接产出交付物的方向。涉及逆向时请自行评估合规与安全边界。

2. text-to-cad:CAD 长期被专业软件垄断,给 Agent 接上 CAD 能力,想象空间在机械与硬件设计的自动化,值得关注它如何定义"文本→几何"的接口。

3. thedotmack 的持久上下文项目:明确列出兼容 Claude Code、Codex、Gemini、Copilot 等多家宿主。"换个会话就失忆"是几乎所有 Agent 用户的共同痛点,通用性使其最有可能被长期留在工具链里。

SOURCES & METHOD

完整信源

GitHub Trending(项目热度、语言、Star) · ModelScope(模型任务、收藏与下载)

抓取日期:2026-10-07。排名是当日快照,指标可能随平台更新而变化。