heatdrop.ai

A browser-based command center that bundles Wikipedia, Khan Academy, offline maps, and local LLMs into one Debian install script.

OpenUI replaces bloated JSON schemas with a compact, streaming-first markup so models can render React components as they speak.

An open-source, community-maintained database that tracks AI model specs, pricing, and capabilities so you don't have to scrape provider docs.

A Next.js app that turns natural language into draw.io diagrams, complete with version history and an MCP server for Claude Desktop.

An Android client that replaces Zhihu's 110 MB ad machine with local recommendations, offline AI filtering, and the radical idea that you should choose what you read.

A Rust runtime that spawns 100 KVM-isolated microVM children in ~100 ms by copy-on-write forking a warmed parent snapshot.

CyberStrikeAI turns scattered security scanners into an orchestrated, chat-driven testing environment with a built-in C2 framework.

A GitHub repo that ranks and reviews third-party API gateways for Chinese developers who can't easily access OpenAI, Anthropic, or Google directly.

BestBlogs.dev ingests 1,000+ daily posts from 600+ feeds, then scores each on six dimensions before it ever reaches your eyes.

HelixDB fuses graph, vector, KV, document, and relational storage into a single Rust-built engine so you don't need five databases to build one RAG app.

A collection of prompt-injection techniques that use invisible Unicode trickery to jailbreak LLMs, with a focus on Claude.

A toolkit that turns an AI agent into an end-to-end video studio — voiceovers, motion graphics, and rendering included.

A Chinese open-source toolkit that automates the entire spam-to-riches pipeline: trend scraping, article generation, AI image creation, and direct publishing to WeChat Official Accounts.

KitOps treats models, datasets, and configs like container images so your existing registry becomes a versioned, auditable model store.

A VST3/AU wrapper that brings neural-network amp modeling from Python experiments to actual music production.

NCCL is NVIDIA's answer to the question "how do we make 256 GPUs talk to each other without the network becoming the bottleneck?"

Extracts, translates, and even re-injects text from Japanese visual novels in real time, with OCR fallback for stubborn games.

Most LLM tools break when you throw a book at them; this one chunks intelligently, preserves EPUB formatting, and resumes after a crash.

An opinionated reading list that tries to map every quantum algorithm that sounds like it could learn something.

A meticulously maintained icon pack that finally spares you from hunting down LLM brand assets one by one.