AI资讯大全AIPPXP.CN搜索 ↗
AI TOOLS · READY TO DOWNLOAD

AI ToolsAI工具

Desktop and command-line tools, ready to download.

桌面与命令行工具,直接下载。

全部内容47 篇

工具介绍 · 附官方安装包

browser-use: The browser automation champion with 119,000 stars releases its stable version 1.0. AI-controlled browsers finally have a de facto standard.browser-use:11.9 万星的浏览器自动化王者发布 1.0 稳定版,AI 操控浏览器这件事终于有了事实标准

Enabling an AI Agent to open a browser, click buttons, fill out forms, and scrape data is something that browser-use has achieved as the most widely cited open-source implementation in the entire field – with 119,000 stars on GitHub. Over the past year, almost all "AI-controlled browser" projects have been built on top of it. The project has now released the 1.0.4 stable version: rewritten based on Pydantic AI, with a persistent Python REPL running directly on the native CDP protocol. It can take over your local Chrome and also supports cloud browsers, with both speed and stability improved.

让 AI Agent 自己开浏览器、点按钮、填表单、抓数据,这件事 browser-use 已经做成了整个领域引用最多的开源实现——GitHub 11.9 万星,一年来几乎所有「AI 操控浏览器」项目都在它之上搭建。现在项目发布 1.0.4 稳定版:基于 Pydantic AI 重写,一条持久的 Python REPL 直接跑在原生 CDP 协议上,可接管你本地的 Chrome,也支持云端浏览器,速度和稳定性都上了一个台阶。

MIT 许可 · 119,000+ 星 · PyPI v1.0.4 · 纯 Python查看 →
Vercel Labs · 当周增速第一

Vercel agent-desktop: A command-line tool that enables AI agents to truly "operate computers"Vercel agent-desktop:让 AI Agent 真正「操作电脑」的命令行工具

No screenshots, no pixel matching: It directly reads the accessibility tree of the operating system, transforming desktop applications into JSON structures with ref numbers. AI clicks, inputs, and verifies everything with precision down to the control level. Written in Rust, it works across three platforms – macOS, Windows, and Linux – and topped GitHub's growth chart the week it was released. It marks the beginning of the era of "agents controlling real computers".

不截图、不像素匹配:它直接读取操作系统的无障碍树,把桌面应用变成带 ref 编号的 JSON 结构,AI 点击、输入、验证全部精确到控件级别。Rust 编写,macOS / Windows / Linux 三平台通吃,发布当周登顶 GitHub 增速榜,是「Agent 操控真实计算机」元年的开山之作。

工具介绍 · 附官方安装包

iFixAi: 41,000-star AI Agent Independent Audit Tool, 49 checks 120 seconds to uncover Agent's unauthorized actionsiFixAi:4.1 万星 AI Agent 独立审计工具,49 项检查 120 秒揪出 Agent 的越权操作

Your AI Agent is working for you every day, but do you know if there is anything it shouldn't do? The Python tool that rushed to the top of the GitHub Hot List this week and won the Product Hunt Daily Ranking gives a complete set of answers: Put the Agent into a sandbox that simulates your real business, send out daily requests and requests that intentionally cross the border, and then be scored one by one by an independent referee from different vendors. 2300 + stars this Sunday, from AI Red Team, Governance, Compliance to Philosophy, turning "Agent Can't Be Trusted at All" into a referenced scorecard.

你的 AI Agent 每天都在替你干活,但你知道它有没有偷偷干不该干的吗?这个本周冲上 GitHub 热榜、拿下 Product Hunt 日榜第一的 Python 工具给出了一套完整答案:把 Agent 放进模拟你真实业务的沙盒,发去日常请求和故意越界的请求,再由一个来自不同厂商的独立裁判逐条打分。本周日增 2300+ 星,从 AI 红队、治理、合规到哲学,把「Agent 到底可不可信」变成一张可引用的评分卡。

Apache 2.0 · 4.1 万星 · Product Hunt 日榜 Top 1查看 →
工具介绍 · 附安装包

DeepTutor: 52,000 stars self-hosted AI private education, turning your materials into a learning platform that will ask questions, ask questions, and keep an eye on progressDeepTutor:5.2 万星自托管 AI 私教,把你的资料变成会出题、会追问、会盯进度的学习平台

A collection of essays and textbooks you've never read? This open source project from HKUDS (Hong Kong Big Data Intelligence Laboratory) has made "learning" a complete product: multi-agent collaboration plus rag, turn your documents into a knowledge base, generate structured learning "Book", with quizzes, mastery tracking and learning paths, and remember where you have learned throughout the memory layer. With one-click self-hosting with Docker, your data is all yours.

收藏了一堆论文和教材,从来没读完过?这个来自 HKUDS(港大数据智能实验室)的开源项目把「学习」这件事做成了完整产品:多代理协作加 RAG,把你的文档变成知识库,生成结构化的学习「Book」,配测验、掌握度追踪和学习路径,还有贯穿始终的记忆层记住你学到哪了。可以 Docker 一键自托管,数据全在自己手里。

5.2 万星 · Apache-2.0 · 自托管查看 →
工具介绍 · 附源码包

Open Duet: Raise a team of AI colleagues on your own computer or server, with a name, a position, and a memory, and you need to nod before you startOpen Duet:在自己电脑或服务器上养一支 AI 同事团队,有名字、有岗位、有记忆,动手前还要你点头

The logic of most proxy tools is “one person facing a chat box”. Open Duet has a different model: it is an open source workspace, you can create a bunch of AI colleagues with names, assign each position, specify the workspace, assign the model, set the memory, and let them assign each other - leading the agent to meet the needs, disassemble the tasks, hand the work to the expert agent, and then return to the chat. The data stays on your own machine or server, and when it needs approval is up to you. 6,100 + stars, rushing to the GitHub Trends this week.

大多数代理工具的逻辑是「一个人对着一个聊天框」。Open Duet 换了个模型:它是个开源工作区,你可以创建一堆有名字的 AI 同事,给每个安排岗位、指定工作区、分配模型、设置记忆,让他们互相派活——主导代理接需求、拆任务,把活交给专家代理,结果回到聊天里。数据留在你自己的机器或服务器上,什么时候需要人批准由你决定。6,100+ 星,本周冲上 GitHub 趋势榜。

自托管 AI 同事团队 · Apache-2.0 · 免费 · 用你自己的模型查看 →
工具介绍 · 附安装包

Yoinks: Paste the link directly under the video in the terminal, 1,800 + sites are omnipresent, there is no pop-up window and no fake download buttonYoinks:在终端里粘贴链接直接下视频,1,800+ 站点通吃,没有弹窗也没有假下载按钮

The experience of finding a video download site is well known: the full-screen "Download Now" button is clicked on which is an advertisement. Yoinks (6,000 + stars, this week's GitHub trend list, mit permission) moved the matter back to the terminal - paste the link, choose the clarity (or only pull the mp3 track), enter, the file fell into the download directory, no pop-ups and no jumps throughout. The bottom layer is yt-dlp, which is supported by more than 1,800 sites such as YouTube, X/Twitter, Instagram, Threads, TikTok, etc. Node 18 + one command is installed.

找视频下载网站的经历大家都懂:满屏「立即下载」按钮点哪个都是广告。Yoinks(6,000+ 星,本周 GitHub 趋势榜,MIT 许可)把这件事搬回终端——粘贴链接、选清晰度(或只抽 mp3 音轨)、回车,文件落进下载目录,全程无弹窗无跳转。底层是 yt-dlp,YouTube、X/Twitter、Instagram、Threads、TikTok 等 1,800 多个站点都支持,Node 18+ 一个命令装好。

终端视频下载器 · MIT · 6,000+ stars查看 →
工具介绍 · 附源码包

Graphify: 133,000-star "Code Knowledge Atlas", so that AI no longer relies on grep to flip files, and the token of reading files is cut by 80%Graphify:133,000 星的「代码知识图谱」,让 AI 不再靠 grep 翻文件,读文件的 token 直接砍八成

Agents change a function, often need to read a dozen files before they dare to do it - this is the most expensive part of the token bill. Graphify parses the entire project (code, documentation, SQL, configuration, PDF) into a traversable knowledge graph at once, and then proxies look up the graph instead of flipping the file: ask a sentence to get the relevant subgraph, ask the two concepts to get the shortest path between them. The code part uses tree-sitter to do pure local AST parsing, but the model does not come out of the machine, 133,000 + stars, Apache-2.0.

代理改一个函数,往往要先读十几个文件才敢动手——这才是 token 账单里最贵的部分。Graphify 把整个项目(代码、文档、SQL、配置、PDF)一次性解析成一张可遍历的知识图谱,之后代理查图而不是翻文件:问一句话拿到相关子图,问两个概念拿到它们之间的最短路径。代码部分用 tree-sitter 做纯本地 AST 解析,不过模型、不出机器,133,000+ stars,Apache-2.0。

Code knowledge graph · Apache-2.0 · 133,000+ stars查看 →
工具介绍 · 附 Windows 安装包

TUIOS terminal manager: "split-screen cockpit" built for AI programming agents, whose agents are doing at a glanceTUIOS 终端管理器:专给 AI 编程代理造的「分屏驾驶舱」,谁的代理在干嘛一眼看清

Opening three or four AI programming agents at the same time, the biggest headache is not knowing what the "employee" in each terminal is doing at the moment: waiting for you to confirm, or has run away. TUIOS is a terminal window manager written in Go. It has split-screen, workspace, and session keepalive. The most important thing is that it knows agents such as Claude Code and Codex. The sidebar displays the status of each agent in real time. The inbox centrally manages all confirmation requests that require you to take a picture of the board.

同时开三四个 AI 编程代理干活,最头疼的是不知道每个终端里那位「员工」此刻在干嘛:是在等你确认,还是已经跑挂了。TUIOS 是一个用 Go 写的终端窗口管理器,分屏、工作区、会话保活全都有,最关键的是它认识 Claude Code、Codex 这类智能体——侧栏实时显示每个代理的状态,收件箱集中管理所有需要你拍板的确认请求。

智能体终端驾驶舱 · Go 编写 · 5,500+ stars查看 →查看 →
工具介绍 · 附安装包

Scrapling crawling framework: 88,000 star open source crawler, automatic adaptation of webpage revision also comes with MCP for AIScrapling 抓取框架:8.8 万星开源爬虫,网页改版自动适配还自带 MCP 给 AI 用

Scrapling is GitHub's 88,000-star adaptive web scraping framework: the selector remembers element characteristics and automatically repositions the target after the website is revised; anti-crawling fingerprint camouflage, request concealment, and concurrent scheduling are all built-in, from a single request to crawling an API throughout the station. It also comes with an MCP server, and the AI agent can directly call the webpage fetching as an ordinary tool.

Scrapling 是 GitHub 8.8 万星的自适应网页抓取框架:选择器会记住元素特征,网站改版后自动重新定位目标;反爬指纹伪装、请求隐匿、并发调度全部内置,从单次请求到整站爬取一个 API 搞定。它还自带 MCP 服务器,AI 智能体可以把网页抓取当普通工具直接调用。

自适应网页抓取 · 内置 MCP · 88,000+ stars查看 →查看 →
工具介绍 · 附安装包

GitHub official Spec Kit: Specification-driven development tool v1.2.1 released, first write the specification and then write the code, AI programming no longer deviatesGitHub 官方 Spec Kit:规格驱动开发工具 v1.2.1 发布,先写规格再写码,AI 编程不再跑偏

GitHub's official open source Spec Kit (github/spec-kit) turns "write specification first, write implementation" into a set of executable commands: specify init initialization project,/specify write specification,/clarify to clarify ambiguity,/plan to define technical scheme,/tasks to disassemble tasks,/implement generation implementation. v1.2.1 was just released on October 16, 2026. With the addition of an extended directory, output capture, and multiple fixes, it is the most popular methodology tool for making AI programming agents obedient.

GitHub 官方开源的 Spec Kit(github/spec-kit)把「先写规格、再写实现」变成一套可执行命令:specify init 初始化项目,/specify 写规格、/clarify 澄清歧义、/plan 定技术方案、/tasks 拆任务、/implement 生成实现。v1.2.1 于 2026 年 10 月 16 日刚发布,新增扩展目录、输出捕获与多个修复,是让 AI 编程智能体听话的最热门方法论工具。

GitHub 官方 · Spec-Driven Development · v1.2.1查看 →查看 →
工具介绍 · 免费好用

Skills Manager: one app to run the skill libraries of 50+ coding tools — batch install, sync and back upSkills Manager:一个界面管好 50 多个编程工具的技能库,批量安装、同步、备份全都有

Skills end up scattered across each tool's own folders, and nobody remembers what is installed where or what has gone stale. Skills Manager (v1.41.0, 6,000+ stars) gathers everything into one central library: batch-install skills from the skills.sh ecosystem, enable and sync them across 50+ tools including Claude Code, Codex and Cursor, split workspaces per agent, and back it all up.

技能散落在每个工具自己的目录里,装了什么、哪边过期了根本记不住。Skills Manager(v1.41.0、6,000+ 星)把所有技能收进一个「中央库」:一键批量安装 skills.sh 生态里的技能,在 Claude Code、Codex、Cursor 等 50+ 工具之间同步启用,按智能体划分工作区,还带备份恢复。

One skill library for 50+ coding tools · 6,000+ stars查看 →查看 →
工具介绍 · 免费好用

markitdown: Microsoft's open-source "anything to Markdown" tool — clean up PDF, Word, Excel and PPT before feeding your LLMmarkitdown:微软开源「万物转 Markdown」,PDF、Word、Excel、PPT 喂给大模型前先转干净文本

LLMs only consume text, but your material is usually PDF, Word, PPT, Excel, images or even audio. markitdown is Microsoft's open-source conversion CLI (MIT, 190,000+ stars) that turns common documents into clean Markdown — preserving headings, tables and structure — and is a classic preprocessing step for RAG, knowledge bases and agent pipelines.

大模型只吃文本,可资料往往是 PDF、Word、PPT、Excel、图片甚至音频。markitdown 是微软开源(MIT、190,000+ 星)的格式转换命令行,一条命令把常见文档转成保留标题、表格与结构的干净 Markdown,是搭 RAG、做知识库、给智能体喂数据的经典前置工具。

Turn any document into Markdown · MIT · 190,000+ stars查看 →查看 →
工具介绍 · 附下载文件

REA: one MCP that lets agents reverse engineer any software, from app behaviour down to native binariesREA:一个 MCP 让智能体逆向任何软件,从 App 行为一路挖到二进制

Seen a feature in someone else's product and wondered how it works — or whether you could build your own? REA turns reverse engineering into a set of MCP tools (about 169,000 stars, MIT licence): your agent analyses an app without source code — native binaries, JavaScript/Electron, .NET assemblies and websites — and every conclusion comes with its evidence and limitations. All analysis runs locally.

看到别人产品里一个好功能,想知道它怎么做的、能不能自己做一个?REA 把逆向工程做成了一套 MCP 工具(约 17 万星、MIT 许可):智能体在没有源码的情况下分析应用——原生二进制、JavaScript/Electron、.NET 程序集、网站都能查,每条结论都带证据与局限说明,分析全程在本机完成。

Reverse engineer anything · MCP · 169,000+ stars查看 →查看 →
工具介绍 · 免费好用

Headroom: compress the logs and documents your agent reads before they reach the model, cutting tokens by up to 95%Headroom:把智能体读到的日志与文档先压缩再喂给模型,token 最高省 95%

A context window clogged with tool output is a daily pain when you write code. Headroom is a context compression layer (about 75,000 stars, Apache-2.0): it compresses logs, tool output, RAG chunks, files and conversation history before they reach the model. Officially that means roughly 20% fewer tokens for coding work and 60%–95% fewer for JSON-like content, while critical lines (such as a FATAL log entry) survive byte for byte. Compression runs entirely on your machine.

上下文被工具输出塞满,是写代码时的日常痛点。Headroom 是一层「上下文压缩层」(约 7.5 万星、Apache-2.0 许可):在日志、工具输出、RAG 片段、文件与会话历史进入模型之前先压缩,官方称编码场景平均少约 20% token,JSON 类内容最高可省 60%–95%,而关键行(比如日志里的 FATAL)逐字节保留。压缩全在本机完成。

Context compression for agents · Apache-2.0 · 75,000+ stars查看 →查看 →
工具介绍 · 附下载文件

AI Job Search: Transform Claude Code into a job search pipeline to evaluate jobs, customize resumes, write cover letters, prepare for interviewsAI Job Search:把 Claude Code 改造成一条求职流水线,评估岗位、定制简历、写求职信、备面试

This is a job search framework running in your native, built on Claude Code. The author himself is a graduate of the geophysical science class. After his post was cut at the end of 2025, he used this workflow to find himself a job: 69 custom applications, 20 scenes, and finally got an offer. In June 2026, he joined as an AI engineer - and then open-sourced it. After erasing the personal story, it is actually a very clear engineering process:/setup filing,/scrape job grabbing,/rank bulk grading,/apply custom application,/interview interview preparation. Licensed by mit, the number of stars has exceeded 46,000.

这是一个跑在你本机的求职框架,基于 Claude Code 构建。作者本人是地球物理学科班出身,2025 年底岗位被裁掉之后,他用这套工作流给自己找工作:69 份定制申请、20 场一面、最后拿到offer,2026 年 6 月入职成为一名 AI 工程师——然后把它开源了。抹掉个人故事之后它其实是一套很清晰的工程化流程:/setup 建档、/scrape 抓岗位、/rank 批量打分、/apply 定制申请、/interview 面试准备。MIT 许可,star 数已经超过 4.6 万。

Job search that runs on your machine · MIT · 46,000+ stars查看 →查看 →
工具介绍 · 开箱即用

Pixelle MCP: Brush that turns ComfyUI workflows into AI agents with one clickPixelle MCP:把 ComfyUI 工作流一键变成 AI 智能体的画笔

Want agents like Claude to generate pictures, audio, video directly? Pixelle MCP automatically converts any workflow in ComfyUI into an MCP tool, and the agent can be plotted in one call. Open source agreement mit, 1,128 stars, with complete Chinese documents, this site packages 5.82 MB source code package can be downloaded directly.

想让 Claude 这类智能体直接生成图片、音频、视频?Pixelle MCP 把 ComfyUI 里的任意工作流自动转成 MCP 工具,智能体调用一次就能出图。开源协议 MIT,1,128 星,附完整中文文档,本站打包 5.82MB 源码包可直接下载。

ComfyUI workflows as MCP tools · MIT · Chinese docs included查看 →查看 →
工具介绍 · 免费好用

FreeLLMAPI: 34 free big model manufacturers twisted into one OpenAI interface, about 7.4 billion tokens per month freeFreeLLMAPI:把 34 家免费大模型厂商拧成一个 OpenAI 接口,每月约 74 亿 token 免费用

Every AI vendor is giving away free credits - millions of tokens a month, thousands of requests a day. A single look is a toy, but stacked is the available inference capacity of about 7.4 billion tokens per month, covering 474 model families and 635 free endpoints. The problem is that manually stacking is too painful: 34 SDKs, 34 throttling rules, 34 places that will fail. FreeLLMAPI collapses all of this into a local OpenAI-compatible endpoint: a/v1 at the same time says OpenAI, Anthropic, Gemini, Ollama four dialects, automatically picks the fastest model, cuts down the one after being throttled, encrypts the key locally, and helps you stay within the free quota of each house according to the key statistical usage.

每个 AI 厂商都在送免费额度——几百万 token 一个月、几千次请求一天。单看一家是玩具,但叠起来是每月约 74 亿 token 的可用推理能力,覆盖 474 个模型家族、635 个免费端点。问题是手动叠太痛苦:34 套 SDK、34 套限流规则、34 个会失败的地方。FreeLLMAPI 把这一切收成一个本地 OpenAI 兼容端点:一个 /v1 同时说 OpenAI、Anthropic、Gemini、Ollama 四种方言,自动挑最快的模型,被限流就切下一家,密钥加密存在本地,还按 key 统计用量帮你守在每家免费额度内。

工具介绍 · 免费好用

Aurelio: a local, self-custodial AI wealth manager that collects bank deposits, broker positions, real estate and liabilities into a net asset statementAurelio:本地自托管的 AI 理财管家,把银行存款、券商持仓、房产与负债收进一张净资产表

Take back “how much money do I have” from seven or eight apps. Aurelio is an open-source self-custodian personal finance board: bank accounts, broker positions, funds, real estate, and liabilities are all summarized into a net asset statement and cash flow curve, and then through OpenRouter to take a large model as a consultant, using natural language to ask it "How can I save 1200 euros per month?" All data is stored on your own machine, not uploaded by third parties.

把「我到底有多少钱」这件事从七八个 App 里收回来。Aurelio 是一个开源自托管的个人财务看板:银行账户、券商持仓、基金、房产、负债全部汇总成一张净资产表与现金流曲线,再通过 OpenRouter 接一个大模型当顾问,用自然语言问它「每个月能多存 1200 欧该怎么分」。数据全部存在你自己的机器上,不上传第三方。

工具介绍 · 免费好用

RTK (Rust Token Killer): Cut the command output read by the AI assistant by up to 90%, single file Rust binary, 80,000 starsRTK(Rust Token Killer):把 AI 助手读到的命令输出砍掉最多九成,单文件 Rust 二进制、8 万星

People who write code with Claude Code, Codex, and Cursor are distressed by tokens: a git status, a test failure log, thousands of tokens are poured into the context, the money is spent, and the window is full. RTK is an "output scissor" written in Rust - it inserts a layer of proxy between the command and the model, and filters, groups, truncates, deduplicates and hands over the output of more than 100 commands such as ls, grep, git, test, build, Docker, kubectl, etc. to the model. Officially, it can reduce the bash output by up to 90%, with a single overhead of less than 10 milliseconds.

用 Claude Code、Codex、Cursor 写代码的人都心疼过 token:一条 git status、一次测试失败日志,动辄几千上万 token 灌进上下文,钱花了、窗口也被占满。RTK 是一把用 Rust 写的「输出剪刀」——它在命令与模型之间插一层代理,把 ls、grep、git、测试、构建、Docker、kubectl 等 100 多个命令的输出做过滤、分组、截断、去重后再交给模型,官方称最高可减少约九成的 bash 输出,单次开销低于 10 毫秒。

工具介绍 · 免费好用

OpenCut: 93,000 Star Open Source Clipper, Free Clipping/CapCut Alternative Running LocallyOpenCut:9.3 万 Star 的开源剪辑器,本地运行的免费剪映 / CapCut 替代方案

Editing software is getting more and more expensive, and the export is also watermarked? OpenCut is currently the most popular video clip project in the open source circle. With 93,000 Star and mit licenses, you can run a complete timeline clip in the browser, and the material is not uploaded to a third-party cloud. It turned the "timeline, transition, special effects, text overlay, audio mixing" set of professional tools into a web application, self-hosted one is equivalent to having your own clipping, and easily solved the problem of privacy and subscription fees together.

剪辑软件越来越贵、导出还带水印?OpenCut 是目前开源圈热度最高的视频剪辑项目,9.3 万 Star、MIT 许可,浏览器里就能跑完整时间线剪辑,素材不上传第三方云端。它把「时间线、转场、特效、文字叠加、音频混音」这套专业工具做成了网页应用,自托管一份就等于拥有了自己的剪映,还顺手把隐私和订阅费问题一起解决了。

工具介绍 · 免费好用

e2e: Open Source AI Testing Framework for Apps Powered by Natural Language, # 1 on GitHub Trends (6300 + Star)e2e:用自然语言驱动 App 的开源 AI 测试框架,GitHub 趋势日榜第一(6300+ Star)

TesterArmy's next-generation end-to-end test framework: use one sentence to describe the goal (such as "upgrade the workspace to the Pro package"), the agent will go to the page, fill out the form, go through the process, and then verify the result with the locator and assertion in the same test. The web and mobile end are omnipresent, and the built-in agent steps to record and playback - the application does not change and does not burn the model repeatedly, and ordinary test cases do not need to call AI at all.

TesterArmy 出品的下一代端到端测试框架:用一句人话描述目标(比如「把工作区升级到 Pro 套餐」),智能体自己去点页面、填表单、走完流程,再用同一条测试里的定位器和断言验证结果。Web 和移动端通吃,自带智能体步骤录制回放——应用没变就不重复烧模型,普通测试用例更是完全不需要调用 AI。

工具介绍 · 免费好用

Strix: an open source AI penetration testing tool that allows AI to run applications like a real hacker, find vulnerabilities and provide repair patchesStrix:开源 AI 渗透测试工具,让 AI 像真黑客一样把应用跑起来、找出漏洞并给出修复补丁

An autonomous penetration testing agent with 66,000 stars. The biggest difference between it and traditional scanners is "verification": AI does not just report a suspicious point, but actually runs the application, constructs a reproducible PoC to prove the existence of the vulnerability, and then provides patches and compliance-level reports. The entire attack toolchain—HTTP interception proxy, browser exploits, shell, Python vulnerability runtime, reconnaissance and OSINT, SAST+DAST—is available out of the box.

6.6 万 Star 的自主渗透测试智能体。它和传统扫描器最大的区别是「验证」:AI 不是只报一个可疑点,而是真的把应用跑起来、构造可复现的 PoC 证明漏洞存在,再给出补丁与合规级报告。整套攻击工具链——HTTP 拦截代理、浏览器利用、Shell、Python 漏洞运行时、侦察与 OSINT、SAST+DAST——都是开箱即用。

工具介绍 · 免费好用

Camofox Browser: An "invisible browser" for AI agents, an open source self-hosted solution that passes Cloudflare and anti-crawling detectionCamofox Browser:给 AI 智能体用的「隐身浏览器」,过 Cloudflare 与反爬检测的开源自托管方案

11,400+ stars. It encapsulates the anti-fingerprint browser Camoufox into an HTTP service that is directly compatible with Puppeteer / Playwright: the page that the agent wants to visit must be used first. Cloudflare queries, robot detection, and anti-crawling verification are all solved locally. It can be self-hosted with a single Docker command, and the data does not leave its own server.

11,400+ Star。它把反指纹浏览器 Camoufox 封装成一个对 Puppeteer / Playwright 直接兼容的 HTTP 服务:智能体要访问的页面先用它过一遍,Cloudflare 质询、机器人检测、反爬验证都在本地解决,一条 Docker 命令即可自托管,数据不出自己服务器。

工具介绍 · 免费好用

Pi Programming Agent: A minimalist agent framework with 110,000 stars. Extensions, skills, and themes are all decided by you.Pi 编程智能体:11 万 Star 的极简智能体框架,扩展、技能、主题全由你定

GitHub continues to dominate the AI ​​agent toolkit list this week, with about 112,000 stars. Pi is a "minimalist but buildable" programming agent: a unified LLM interface, agent loop, TUI terminal interface and programming CLI are all available, but it deliberately does not have sub-agent and planning mode - whatever you need, let Pi write an extension by itself, or install a package made by others from npm.

GitHub 本周持续霸榜的 AI agent toolkit,约 11.2 万 Star。Pi 是一个「极简但可造」的编程智能体:统一 LLM 接口、agent 循环、TUI 终端界面和编程 CLI 一应俱全,但故意不带子智能体和计划模式——你需要什么,让 Pi 自己写一个扩展,或从 npm 装一个别人做好的包。

工具介绍 · 附下载文件

PageIndex v0.2.21: Do not index “vectorless rag” documents for vector databasesPageIndex v0.2.21:不要向量数据库的「无向量 RAG」文档索引

VectifyAI open source, 38,000 Star rag new idea: do not cut the document, do not do Embedding, do not build a vector library, but let LLM directly generate a tree-like directory index for the document, retrieved like a human flip book according to directory reasoning positioning, accuracy in FinanceBench and other benchmarks inverse traditional vector rag, indexing cost is also one order of magnitude cheaper.

VectifyAI 开源、3.8 万 Star 的 RAG 新思路:不给文档切块、不做 Embedding、不建向量库,而是让 LLM 直接给文档生成一棵树状目录索引,检索时像人翻书一样按目录推理定位,准确率在 FinanceBench 等基准上反超传统向量 RAG,索引成本还便宜一个量级。

PageIndex is open-sourced by VectifyAI (mit protocol), and this site transfers the official Release original files. If you are connecting the agent to the knowledge base, try using it instead of the vector library first, there is a high probability that you will not be able to go back.查看 →查看 →
工具介绍 · 附下载文件

NVIDIA OpenShell v0.1.2: Give the agent a "locked room", the official Linux version is about 9.96 MBNVIDIA OpenShell v0.1.2:给智能体一间「上锁的房间」,官方 Linux 版约 9.96MB

When Nvidia's open source agents are running safely, GitHub's hot list is the first echelon on the day. It solves a very real problem: the more capable the agent is, the more it needs to read files, load dependencies, call APIs, use credentials - and these permissions are difficult to take back once they are released. OpenShell allows agents to retain these capabilities without giving them unlimited access to your data, keys, and network.

英伟达开源的智能体安全运行时,当天 GitHub 热榜第一梯队。它要解决的是一个很现实的问题:智能体越能干,就越需要读文件、装依赖、调 API、用凭据——而这些权限一旦放开就很难收回。OpenShell 让智能体保有这些能力,同时不给它对你数据、密钥和网络的无限访问权。

Project address: github.com/NVIDIA/OpenShell (Apache 2.0). This is a very new project (0.1.x is just beginning to go steady release rhythm), the document is at docs.nvidia.com/openshell, it is recommended to read the support-matrix to confirm the platform availability before getting started.查看 →查看 →
工具介绍 · 附下载文件

Hindsight v0.10.2: Long-term memory that lets your agent actually learnHindsight v0.10.2:给智能体一套会「学」的长期记忆

One of the fastest-rising memory systems today, around 43,500 stars. It stores facts, entity relationships and temporal reasoning in PostgreSQL plus pgvector, so an agent remembers people, events and time, and forms opinions under configurable disposition traits. One docker run starts the service, and an MCP endpoint comes built in.

当天涨星最猛的记忆系统之一,约 4.35 万 Star。它把事实、实体关系和时序推理存进 PostgreSQL + pgvector:智能体能记住人、事、时间,还能按可配置的性格倾向形成判断。一条 docker run 起服务,自带 MCP 端点。

Project address: github.com/vectorize-io/hindsight (MIT). The service default port is 8888, and the memory is stored in the native PostgreSQL; the large model of the call is configured by yourself, and the local ollama can be connected completely offline.查看 →查看 →
工具介绍 · 附下载文件

Tencent Cloud open source Octop 1.0.2b4: a multi-user, multi-agent self-hosted AI assistant installed in one process腾讯云开源 Octop 1.0.2b4:一个进程装下多用户、多智能体的自托管 AI 助手

This week's GitHub hot list project, about 5.5k stars. A single process startup provides a web console, command line, eight major IM channels (Feishu, DingTalk, QQ, WeChat, Telegram, Discord, Enterprise WeChat, etc.) and natural language scheduled tasks. All data falls in the ~/.octop/ directory of the local machine.

本周 GitHub 热榜项目,约 5.5k Star。单进程启动就同时提供 Web 控制台、命令行、八大 IM 渠道(飞书、钉钉、QQ、微信、Telegram、Discord、企业微信等)与自然语言定时任务,所有数据落在本机 ~/.octop/ 目录里。

Project address: github.com/TencentCloud/Octop (Tencent Cloud official open source, MIT license; self-hosted deployment, please keep your local credentials and data properly).查看 →查看 →
工具介绍 · 附下载文件

OpenSpec v1.13.2: Write specifications first and then code, and put an "anti-deviation" constraint on AI programmingOpenSpec v1.13.2:先写规格再写代码,给 AI 编程上个「防跑偏」约束

A specification-driven development tool with about 70.5k stars, a GitHub weekly list project in September 2026. It allows AI to first write requirements into verifiable specification documents, and then write code according to the specifications, fundamentally solving the old problem of "changing to the fifth round is no longer the original requirement."

约 70.5k Star 的规格驱动开发工具,2026 年 9 月 GitHub 周榜项目。它让 AI 先把需求写成可验证的规格文件,再照着规格写代码,从根上解决「改到第五轮已经不是原来的需求」这个老毛病。

Project address: github.com/Fission-AI/OpenSpec (MIT-like open source license, can be used for personal and commercial projects).查看 →查看 →
工具介绍 · 附下载文件

PR Lens open source tool: automatically generate architecture and data flow walkthrough diagrams during code reviewPR Lens 开源工具:在代码评审里自动生成架构与数据流走查图

Can’t understand the code review, and the PR backlog is unattended? PR Lens generates animated architecture diagrams and data flow walkthrough diagrams directly in Pull Requests, allowing reviewers to understand the change path at a glance. Approximately 1,300 stars on GitHub.

代码评审看不懂、PR 积压没人盯?PR Lens 直接在 Pull Request 里生成动画版架构图与数据流走查图,让评审者一眼看懂改动路径,GitHub 约 1300 星。

Project address: github.com/coldteadotai/pr-lens. The sample image is taken from the official showcase of the warehouse.查看 →查看 →
工具介绍 · 附下载文件

LiteLLM v1.102.1: One gateway to access more than 100 large models, new streaming guardrail upgradeLiteLLM v1.102.1:一个网关接入 100 多家大模型,流式护栏全新升级

LiteLLM, a star open source project with more than 20,000 stars on GitHub, has released version v1.102: Guardrails can finally take effect in real-time in streaming responses, and can be rewritten while being generated without waiting for the entire output to end.

GitHub 2 万多星的明星开源项目 LiteLLM 发布 v1.102 版本:护栏(Guardrails)终于能在流式响应中实时生效,边生成边改写,不用等整段输出结束。

Source: github.com/BerriAI/litellm (MIT License). For the construction steps, see the "LiteLLM Gateway Construction Graphical Tutorial" on this site.查看 →查看 →
工具介绍 · 附下载文件

Google open source agent orchestrator ax v0.3.0: Manage AI agents like a cluster谷歌开源智能体编排器 ax v0.3.0:像管集群一样管 AI 智能体

Google's official open-source agent orchestration runtime is on the GitHub trend list: a YAML file declares a task, a sandbox starts the agent, and SSHs in to watch it work. It can be called the "Kubernetes of the agent world."

谷歌官方开源的智能体编排运行时登上 GitHub 趋势榜:一个 YAML 文件声明任务,沙箱启动智能体,SSH 进去围观它干活,堪称「智能体界的 Kubernetes」。

For more AI developer tools, please see the "AI Tools" column of this site, all of which can be downloaded for free.查看 →查看 →
工具介绍 · 附下载文件

AI-Infra-Guard: AI infrastructure security scanner produced by Tencent ZhuqueAI-Infra-Guard:腾讯朱雀出品的AI基础设施安全扫描器

Scan your Agent, Skill, MCP server and AI infrastructure with one click to identify jailbreak risks and security vulnerabilities. Produced by Tencent Suzaku Labs, with 6,000+ stars on GitHub.

一键扫描你的 Agent、Skill、MCP 服务器和AI基础设施,找出越狱风险与安全漏洞,腾讯朱雀实验室出品,GitHub 6千+星。

AI security · Tencent Zhuque Lab · Open source and free查看 →查看 →
工具介绍 · 附下载文件

Aichat: command line all-round AI assistant (direct download for Windows version)aichat:命令行全能AI助手(Windows版直接下载)

A command line AI assistant of less than 5MB: dialogue, translation, summary, and even let AI help you execute commands, all in one tool, open source and free.

一个不到5MB的命令行AI助手:对话、翻译、总结、甚至让AI帮你执行命令,一个工具全包,开源免费。

Command Line AI · 5MB All-Purpose查看 →查看 →
工具介绍 · 附下载文件

Aider: AI pair programming assistant, directly help you change the codeAider:AI结对编程助手,直接帮你改代码

Pair up with AI in the terminal to write code: describe the requirements, and Aider will directly read your project, modify files, submit to git, and run after modifications are made.

在终端里和AI结对写代码:描述需求,Aider直接读你的项目、改文件、提交git,改完能跑。

AI programming · Direct code modification查看 →查看 →
工具介绍 · 附下载文件

ChatGPT desktop version: Direct download of Windows client installation packageChatGPT桌面版:Windows客户端安装包直接下载

There is no need to open a browser, the desktop directly communicates with ChatGPT: global shortcut keys can be called up at any time, permanent tray is available, and multiple windows are supported.

不用开浏览器,桌面直接和ChatGPT对话:全局快捷键随时唤出、常驻托盘、支持多开窗口。

Desktop client · Quick call查看 →查看 →
工具介绍 · 附下载文件

Chrome DevTools MCP: Let AI really know how to open the browser, look at the console, and check performanceChrome DevTools MCP:让AI真的会开浏览器、看控制台、查性能

MCP server officially maintained by the Google Chrome team. After installation, the AI ​​assistant no longer relies on screenshots to guess pages, but can directly open a real browser, read console errors, capture network requests, and run performance tracking - front-end debugging has changed from "describing the problem" to "doing it yourself."

谷歌 Chrome 团队官方维护的 MCP 服务器。装上之后,AI 助手不再靠截图猜页面,而是能直接开真实浏览器、读控制台报错、抓网络请求、跑性能追踪——前端调试从「描述问题」变成「自己动手验证」。

Official open source · Browser automated debugging查看 →查看 →
工具介绍 · 附下载文件

Fabric: Load hundreds of AI prompt word templates into the command line (Windows version)fabric:把上百个AI提示词模板装进命令行(Windows版)

Open source AI prompt word template library + executor: summarize videos, refine articles, write weekly reports, analyze threat intelligence... hundreds of ready-made patterns, called with one command.

开源的AI提示词模板库+执行器:总结视频、提炼文章、写周报、分析威胁情报……上百个现成模式,一条命令调用。

Prompt word template library · Advanced for masters查看 →查看 →
工具介绍 · 附下载文件

GPT Engineer: Let AI generate the entire project in one sentenceGPT Engineer:一句话让AI生成整个项目

Describe the application you want, and AI will automatically generate the complete project code and continuously iterate based on your feedback, making it suitable for rapid prototyping.

描述你想要的应用,AI自动生成完整项目代码并不断按你的反馈迭代,适合快速做原型。

AI website building · rapid prototyping查看 →查看 →
工具介绍 · 附下载文件

Llmfit: One command to find out which large model your computer can runllmfit:一条命令查出你的电脑能跑哪个大模型

No more guessing "whether my graphics card can run 70B". llmfit reads hardware and configures models with one command. It has 37,000 stars on GitHub and is one of the top five popular tools in terms of growth rate this month.

不用再瞎猜“我这显卡能不能跑 70B”。llmfit 一条命令读硬件、配模型,GitHub 3.7万星,本月增速排名前五的爆款工具。

Local deployment · Model selection · Open source and free查看 →查看 →
工具介绍 · 附下载文件

MarkItDown: Microsoft’s open source universal converter, convert PDF/Word/PPT into Markdown with one clickMarkItDown:微软开源的万能转换器,PDF/Word/PPT 一键变 Markdown

If the document format fed to AI is incorrect, the model will lose content when reading it. Microsoft's open source MarkItDown specializes in this: converting PDF, Word, Excel, PPT, pictures, audio, and HTML into clean Markdown. The structure, titles, and tables are all preserved, and can be directly inserted into large models.

喂给 AI 的文档格式不对,模型读起来就会丢内容。微软开源的 MarkItDown 专治这件事:把 PDF、Word、Excel、PPT、图片、音频、HTML 统统转成干净的 Markdown,结构、标题、表格都保留下来,直接就能塞给大模型。

Official open source · Document format conversion查看 →查看 →
工具介绍 · 附下载文件

Desktop version of Zhuici Translator: Direct download of Windows installation package (open source and free)划词翻译桌面版:Windows安装包直接下载(开源免费)

Don't want to install a browser plug-in? This desktop version is globally available on Windows: select text in any software and press the shortcut key to translate it, and it also supports OCR screenshot translation.

不想装浏览器插件?这个桌面版在Windows上全局可用:任何软件里选中文字按快捷键就出翻译,还支持OCR截图翻译。

Global word delineation · Windows installation package查看 →查看 →
工具介绍 · 附下载文件

Open Interpreter: Let AI perform tasks directly on your computerOpen Interpreter:让AI直接在你电脑上执行任务

Use natural language to command the computer: let AI write code and run it, modify files in batches, and process table data. Open source star project, the authority is in your hands.

用自然语言指挥电脑:让AI写代码并运行、批量改文件、处理表格数据。开源明星项目,权限在你手里。

AI Automation · Local Execution查看 →查看 →
工具介绍 · 附下载文件

Repomix: One-click packaging of code repositories into files that can be read by AIRepomix:一键把代码仓库打包成AI能读的文件

Package the entire project into a structured text file and feed it directly to AI analysis: an artifact to modify old projects and understand other people’s code.

把整个项目打包成一个结构化文本文件,直接喂给AI分析:改老项目、读懂别人代码的神器。

Code understanding · AI analysis查看 →查看 →
工具介绍 · 附下载文件

SkillSpector: Packing a security gate for AI skills (NVIDIA open source)SkillSpector:给AI技能包装一道安检门(英伟达开源)

The AI ​​skill package you download from the Internet may contain prompt word injection, data external transmission, and supply chain poisoning. Nvidia's open source SkillSpector is designed to handle this: before installing any skills, let it unpack the package and scan it.

你从网上下载的AI技能包里,可能藏着提示词注入、数据外传、供应链投毒。英伟达开源的 SkillSpector 就是专治这个的:安装任何技能之前,先让它把包拆开过一遍扫描。

AI security · Skill package scanning · NVIDIA open source查看 →查看 →