2026年7月4日 技术热点总结

项目简介：Let Claude (or any LLM) actually watch a video — scene-aware, deduplicated frames + transcript, from a URL or local file. Runs locally, MIT.

技术栈：Python

核心介绍：> Same 58-second clip: fixed 1 fps sampling = 58 frames. crv keeps the 26 that actually differ — and –grid packs them into 3 contact sheets. Fewer tokens, nothing missed. Most AI tools don’t really *see* a video. Paste a YouTube link into ChatGPT and it reads the transcript, not the picture. Claude won’t take a video file at all. Even Gemini, which *can* read video natively, has to send it up to Googl…

项目简介：小隐寺投资百科官方公开索引：美股、期权与加密货币知识框架

技术栈：chinese、cryptocurrency、investing、options、xiaoyinsi

核心介绍：这是一份面向中文投资者的公开投资入门指南，适合想从零开始了解美股、期权和加密货币，却不知道应该先学什么、怎样建立完整知识框架的人。 指南不要求读者具备金融专业背景。内容从交易时间、订单类型、财务报表和基础估值讲起，再逐步延伸到期权定价、仓位管理、钱包安全、链上交易与常见骗局，帮助初学者看懂投资过程中真正会遇到的术语和操作问题。 你可以使用本指南： 这份指南的目标不是提供“买什么”的答案，而是帮助你理解自己正在买什么、风险来自哪里，以及在投入资金前应该核对哪些信息。每篇内容尽量保持简短、实用，并给出可继续学习的官方词条入口。 普通人的收入主要来自工作，但买房、养老、子女教育和应对意外，都需要跨越很长时间。 大部分（尤其父母那一辈人）只依赖当期收入和现金储蓄，可能受到通胀、收入中断和购买力下降的影响，晚年生活非常凄惨，而中国是不进行任何投资者教育的，类似「电诈为什么无法彻底根除」、「骚扰电话为什么屡禁不止」，这里面有着太多的利益冲突和对普通人额外的限制。 学习投资，本质上是学习怎样把今天的劳动成果，更合理地分配到未来，而不是追求一夜暴富。 不学习投资，并不意味着可以避…

项目简介：Native iPhone app for your Hermes agent

技术栈：Swift、hermes、hermes-agent、hermex、ios、llm、self-hosted、swift、swiftui

核心介绍：Your server. Your iPhone. No middleman. Website · App Store · Report a bug · Contributing Hermex is a native SwiftUI iPhone app for driving a self-hosted hermes-webui server — a mobile cockpit for an AI agent that lives on a machine you control. The phone is the control plane, not the compute plane: the agent, its tools, and your data stay on your own hardware.

关键特性：Chat with your agent — send messages with model, reasoning-effort, workspace,…

项目简介：在 Mac（Apple Silicon / Intel）上用 UTM 跑 Linux 虚拟机，把大疆 4G 模块（EG25-G）伪装成移远 Quectel EC25 并部署 vohive 平台的完整步骤

核心介绍：> 在 Mac（Apple Silicon / Intel 通用）上，用 UTM 跑一个 Linux 虚拟机，把大疆 4G 模块（1 代，本质移远 Quectel EG25-G）的 USB 身份从大疆私有 2ca3:4006 永久改成移远 Quectel EC25 的 2C7C:0125，并在该 Linux 里一键部署 vohive 短信/网络/eSIM 管理平台的全套步骤。 本仓库本身只是一份操作手册（README），实际起作用的是下面这个上游项目，部署步骤会调用它的一键安装脚本： curl -fsSL https://raw.githubusercontent.com/iniwex5/vohive-release/master/install.sh | bash 安装后：二进制 /opt/vohive/bin/vohive，配置 /opt/vohive/config/config.yaml，systemd 服务 vohive，后台 http://:7575（默认 admin/admin）。

**项目数…

项目简介：Codex CLI instruction-file installer for local configuration

技术栈：Python

核心介绍：Codex CLI instruction-file installer for local configuration. 简体中文 · English · License > Status boundary / 状态边界

项目数据：⭐ 411 Stars，🍴 70 Forks

项目简介：UPI redeem only extension

技术栈：JavaScript

核心介绍：这是 CDK 兑换专用版 Chrome 扩展。当前版本保留 UPI 主流程：邮箱注册、邮箱验证码、设置 GPT 密码、第 7 步开通 2FA、读取 AT、Free/Plus 分组、CDK 兑换、Plus 识别/验证、导入导出。 完整配置说明见 docs/CONFIG-USAGE.md。 邮箱验证码能力保留，因为注册和设置 GPT 密码仍需要邮箱取码。

关键特性：自动注册邮箱账号并读取邮箱验证码。；设置 GPT 登录密码。；第 7 步开通 TOTP 2FA、读取 access token、检测是否有试用资格。；资格通过后保存到 Free 组。；Free 组导入、导出、补充 AT、一键识别 Plus、一键兑换 CDK。。

项目数据：⭐ 398 Stars，🍴 170 Forks

项目简介：多线程全自动注册free 强上K12空间

技术栈：Python

核心介绍：aBaiAutoplus 是一个以 ChatGPT free 账号注册、管理和本地配置为主的 Web 面板。当前公开前端侧栏只暴露三个顶层入口：总览、chatgpt free、设置。 本 README 仅按当前前端可见菜单描述项目能力。代码中可能仍存在历史页面、内部路由或实验组件，但它们不属于当前公开入口。 用于查看系统和账号整体状态： 点击链接加入群聊【GPT PLUS交流群aBaiAutoplus】：https://qm.qq.com/q/vbHPiYpqUg 用于管理 ChatGPT free 账号： 设置页的子菜单当前全部保留： ..venvScriptsActivate.ps1 pip install -r requirements.txt python -m uvicorn main:app –host 0.0.0.0 –port 8000 启动后访问 http://localhost:8000。 python3 -m venv .venv source .venv/bin/activate pip install -r requirements.txt

项目数据：⭐ 377 Stars，🍴 154 Forks

项目简介：A css framework recreating the DS / DS Lite’s UI

技术栈：CSS

核心介绍：A css framework recreating the DS Lite’s UI. Preview what’s included here! to use it, you can copy and use the contents of /css, then include it in your html’s npm i @spiritov/ds.css

项目数据：⭐ 373 Stars，🍴 2 Forks

项目简介：Ackem — 本地优先 AI 桌面陪伴 · Local-first AI desktop companion. 隐私数据不上传，支持记忆/情绪/扩展/ 。AGPL-3.0

技术栈：TypeScript

核心介绍：> Source: GitHub · Gitee mirror > Download: GitHub Releases · Gitee Releases > Build: npm run dist:green → dist/release/Ackem-1.0.0-win-x64/ · Path map · Docs languages > Status： Ackem is still in active testing. As a solo-maintained project, test coverage is limited — you may hit unexpected behavior or rough edges. Frequent crashes and severe lag are uncommon, but imperfections still happen. Open an Issue if something feels off. 中文文档 · Privacy & …

Grid-based approaches to approximate nearest neighbor (ANN) search have been absent from modern scaling analyses. We present a systematic characterization of a multiprobe grid algorithm with respect to dataset size N and dimensionality d. Our experiments reveal a previously unreported d-scaling crossover on the GloVe embedding family, in which multiprobe grid search maintains an approximately constant dimensional scaling exponent while other graph-, tree-, and partitioning-based methods exhibit degrading throughput. The advantage comes with near-linear query scaling in N, but also with lowe…

Traffic matrices (TMs) capture network-wide origin-destination demand and are central to traffic engineering, yet accurate whole-matrix forecasting remains challenging when prediction must be performed under the memory, update, and training-budget constraints of online network control. This paper investigates whether compact quantum-inspired recurrent models can provide effective TM forecasts without relying on dedicated graph, transformer, or diffusion modules. We adapt gated quantum-inspired Kolmogorov-Arnold network fast-weight programmers (QKAN-FWPs) to direct multi-step Abilene TM fore…

Foundation models are routinely released to the public, yet the data recipes used to train them — such as domain mixture weights that determine how different sources are sampled — are rarely disclosed. This creates an access asymmetry: researchers study the resulting models but lack visibility into the training distribution that produces them. Prior works for inferring training data, such as membership inference, detect at the level of individual samples and thus cannot characterize the global composition of the training corpus. We introduce WARP, a framework that recovers a fine-tuned mo…

Memory expertise is a learned skill: knowing what to encode, when to retrieve, and how to organize knowledge–a capacity known in cognitive science as metamemory. We bring this perspective to LLMs by treating memory management as a trainable skill. We promote file-system operations to first-class memory actions alongside task actions, letting the model itself decide how to manage its memory. This memory skill improves along two axes: the structure that supports it (prompts, file schemas, action vocabulary), and the proficiency of the model exercising it. Both axes resist manual optimization…

Large Language Model (LLM)-based agents can solve complex procedural tasks by interacting with environments over multiple turns, but this ability typically depends on large models, long contexts, and repeated inference calls. This makes advanced memory-augmented agents difficult to deploy on resource-constrained devices. We introduce DuoMem, a dual-space distillation framework that transfers procedural problem-solving ability from a large teacher model to compact student models. DuoMem distils in two complementary spaces: (1)context-space distillation, which replaces student-generated memor…

In long-context use, large language models frequently synthesize answers from the meaning of a relevant context span rather than literally copy-pasting them. Identifying which attention heads perform this synthesis matters for interpreting long-context model behavior. Yet existing detectors miss these heads by construction: they reward heads whose attended token matches the generated token, a literal-copy criterion that captures where a head reads but not what it writes through its output-value (OV) circuit, the very mechanism that carries non-literal retrieval. We introduce Logit-Contribut…