Usable AI tools, platforms, agents, and workflow products
CloakHQ/CloakBrowser gained 7,769 stars in a week, reaching 18,544 stars. It is a stealth browser that modifies Chromium fingerprints at the C++ source code level, passes tests across 30+ anti-detection sites, and serves as a plug-and-play replacement for Playwright/Puppeteer.
colbymchenry/codegraph has surpassed 15,600 stars on GitHub, gaining over 10,700 stars this week alone. This tool provides AI coding agents with a pre-indexed code knowledge graph, proven to reduce token consumption by approximately 35% and tool calls by 70%, while running 100% locally.
Cognee has reached 17,300 stars on GitHub, and its slogan is quite bold: "Memory control plane for AI Agents in 6 lines of code." This infrastructure project, which provides memory capabilities to AI Agents, is rapidly becoming a new standard in AI application development.
Dograh is an open-source voice Agent platform that supports the OpenAI Realtime model and provides a complete voice interaction workflow. Version 1.30.1 was just released yesterday, with continuous iteration across 468 commits.
DreamServer is an all-in-one local AI platform covering LLM inference, chat interfaces, voice interaction, agents, workflows, RAG, and image generation. No cloud services, no subscriptions, and continuously iterated with 1,900+ commits.
tinyhumansai/openhuman gained 17,399 new stars on GitHub in a week, bringing its total to 25,425. This private AI Agent platform, written in Rust, integrates 118+ third-party services, features a desktop pet, can join Google Meet calls, retains cross-week memory—and runs entirely locally.
PentestGPT is an automated penetration testing framework based on large language models, boasting 13,200 stars and supporting agentic workflows. It enables AI to do more than just "write security reports" by truly participating in the complete penetration testing process.
ruvnet/RuView has reached 63,537 stars on GitHub, gaining over 7,600+ stars this week. This project uses standard WiFi signals for spatial awareness, capable of detecting people through walls, monitoring breathing and heart rate, and recognizing activities—all with just a $9 ESP32 development board.
Shadowbroker has rapidly surpassed 7,500 stars on GitHub. It integrates flight tracking, satellite orbits, and seismic monitoring into a single interface, and can be paired with an AI Agent to automatically uncover data correlations. This might be the most powerful open-source intelligence platform accessible to individuals.
HKUDS/ViMax gained 2,495 stars this week, reaching a total of 6,619. This open-source project from the HKU Data Science team enables AI to autonomously handle the entire workflow from concept to final cut—scriptwriting, storyboarding, character creation, and video generation, all in one place.
Datawhale's open-source Easy-Vibe project provides a complete learning path from zero basics to AI programming. "If you can speak, you can build apps" is more than just a slogan—it transforms abstract programming concepts into an intuitive learning experience through interactive tutorials, simulated IDE operations, and visualized AI principles. The tutorial now supports 10 languages, and its community-driven learning map clearly outlines what to learn at each stage.
Matt Pocock, one of the most influential educators in the TypeScript space, has open-sourced his collection of Skills from his `.claude` directory, rapidly reaching 98,200 stars. This isn't a tutorial; it's a practical skill pack that can be directly loaded and executed by Claude Code—covering core scenarios in a real engineer's daily workflow, from code reviews and architectural decisions to debugging and refactoring.
RuView has reached 62,595 stars on GitHub, gaining 8,028 new stars this week. Using a $9 ESP32 board to read WiFi signal reflections, it can detect human presence through walls and monitor breathing and heart rates without any cameras or wearables. All 1,463 tests pass, and its 4-bit quantized model is only 8KB. But is it genuinely useful or just overhyped? Let's break it down.
obra/superpowers has reached 98,200 stars on GitHub, making it one of the most popular projects on this week's trending list. It's not a coding tool, but a complete software development methodology for coding Agents—from requirement clarification to specification documents, from implementation plans to sub-agent-driven development. It enables Claude Code to work autonomously for hours without going off track.
South Korean audio tech company Supertone has open-sourced Supertonic 3, a text-to-speech system with only 99M parameters that supports 31 languages. It runs entirely locally without requiring a GPU and outputs high-quality 44.1kHz audio. Last week, it released the `serve` mode for its Python SDK, supporting OpenAI-compatible API endpoints.
CLI-Anything is an open-source project by the HKUDS team, aiming to enable AI Agents to directly control *all* traditional software. With 38K stars and its CLI-Hub platform (clianything.cc), it represents a new “Agent-first” paradigm for software interaction.
Forge is a Python framework that boosts an 8B-parameter small model’s success rate on agent tasks from 53% to 99% via Guardrails. A trending HN post (415 points), it demonstrates how systematic constraints enable inexpensive models to match the performance of costly ones.
OpenAI has announced the adoption of Google's SynthID technology to watermark its AI-generated images, alongside a verification tool. Scoring 259 points on a Hacker News hot post, this marks the arrival of a cross-platform standard solution for labeling AI-generated content.
OpenHuman is an open-source personal AI super-intelligence platform built with Rust, supporting local deployment, multi-model routing, and an Agent skill system. It was updated just 44 minutes ago, has 2,113 commits, and boasts an extremely active community.
rtk is a CLI proxy tool written in Rust that intelligently intercepts and rewrites common development commands, reducing LLM token consumption by 60-90%. With 51K+ stars, zero dependencies, and deployment as a single binary file.
NirDiamant/agents-towards-production has accumulated 19,874 stars on GitHub. It doesn't build frameworks or platforms—it does one thing: end-to-end, code-first tutorials that take you from GenAI Agent prototype to enterprise production. In an era of AI tool proliferation, how to deploy content is scarce.
HKUDS/CLI-Anything sits on GitHub Trending with 35,515 stars. Its radical concept: make all existing software "Agent-Native" — AI Agents can directly operate any software through CLI, without API adaptation. This is not just a tool, but an architectural paradigm.
The Darwin series proposes a new model optimization approach: no training, no fine-tuning—use evolutionary algorithms to "merge" multiple existing models into a stronger one. This week it hit GitHub Trending, providing developers with an actionable model enhancement toolkit.
Anil-matcha/Open-Generative-AI hit GitHub Trending this week with 15K stars. It integrates 200+ models including Flux, Midjourney, Kling, Sora, and Veo under MIT license with self-hosting support and no content filtering.
sleuth-io/sx proposes a simple but powerful concept: AI coding assistant skills, MCP services, and commands should have a unified package manager. Just like npm manages JavaScript packages and pip manages Python packages, sx manages AI Agent capability modules.
