Search

GitHub - alexwkleung/web-search-skill: Web search (Brave) and page reading (defuddle) skill for coding agents through Chrome CDP · GitHub

A Chrome-based web search and page-reading skill for coding agents uses a real browser via DevTools, avoids search APIs, and outputs pages as markdown with defuddle. It is distributed as a small, install-from-source tool with environment-driven runtime selection, primarily tested on macOS and Linux support indicated as untested; updates and maintenance are managed via vendor bundles and versioned scripts.

GitHub - abi/screenshot-to-code: Drop in a screenshot and convert it to clean code (HTML/Tailwind/React/Vue) · GitHub

Screenshot to Code turns screenshots, design files, and screen recordings into web code for several frameworks and styling systems. The GitHub project documents hosted and local setup, model-provider API keys, and optional Docker and browser-preview support.

Welcome to CodeRail | CodeRail

CodeRail is a self-hosted web app that runs Claude Code and Codex on your own machines via a single Docker container, with per-computer roles and persistent terminals, enabling browser or phone access to code, Git, and device dev servers without third-party servers. It is open source under Apache 2.0 by Wiva, Inc.

REA MCP & CLI: Reverse Engineer Anything with AI Agents | AI/TLDR

REA is an open-source toolchain that lets AI agents decompile and understand software by running locally with Hopper or Ghidra, producing reproducible evidence bundles; it covers native and web/JS targets and does not upload apps to a service.

GitHub - microsoft/azure-dev-tools: Developer experience tools for Azure (e.g. Canvases, skills) · GitHub

Azure Dev Tools for GitHub Copilot adds a suite of Azure-focused plugins for Copilot to manage hosted skills, resource inspection, SRE diagnostics, cost/usage, storage, containers, and canvas apps; installation is via the marketplace with version-pinned releases and explicit setup instructions.

GitHub - OpenDocBot/OpenDocBot: The client-side, provider-agnostic M365 AI agent. No lock-in, no hidden telemetry. · GitHub

OpenDocBot is a client-side, provider-agnostic Microsoft 365 add-in for Word, Excel, and PowerPoint that runs offline or online, connects to multiple AI providers, and emphasizes zero telemetry, human oversight, and self-hosting options for teams. It supports in-app reading, writing, analysis, formulas, and slide creation, with sandboxed execution and configurable provider/model setups including SSO and self-hosted deployments; compatibility is limited to Microsoft 365 apps, with detailed documentation available.

Trying DeepSeek Search API・Fetch API - Web search usable with Local LLM | npaka

Translated title. Original: DeepSeek Search API・Fetch APIを試す - ローカルLLMで使えるWeb検索｜npaka

DeepSeek publicly offers Search API and Fetch API for low-cost web search and page extraction; pricing is $0.75 per 1,000 calls with no charge for empty results or errors, and no permanent free tier. The document outlines setup steps, pricing comparisons, and quick integration methods for Pi, OpenCode, and MCP environments, aimed at enabling quick deployment and testing of the APIs.

GitHub - asimovia/dnapps: DNApps: an open draft spec for sharing the DNA of an app instead of the app. One text file a coding agent grows into an app adapted to you. Open for comment until January 10, 2027. · GitHub

DNApps is an open draft specification for text-file capsules that coding agents use to build apps tailored to a person’s needs and setup. Its authors invite feedback on the format and seek more data on how capsule size compares with the resulting app.

Meta's 816 billion-parameter open model "Muse Spark Open" - AI Watch - YAYAFA

Translated title. Original: Meta、8,160億パラメータのオープンモデル「Muse Spark Open」 - AI Watch - YAYAFA

Meta AI Research announced Muse Spark Open, an 8.16B-parameter multi-modal model with a 1M token context window, available via multiple API platforms and downloadable from major cloud and ML stores. It targets software engineering and autonomous agents, with Apple Silicon optimizations enabling local multi-machine setups; on-device testing used Muse Code, including iOS changes for Mastodon and frame-accurate video review in DaVinci Resolve.

Hamza Tahir · human.json

An open-source Python platform for building AI pipelines and agents, featuring runnable agent traces with durable execution, checkpoints, and replay, gaining early popularity (≈310 GitHub stars).