Accelerating heavyweights

AirLLM slices giant transformers into layer shards so they fit in consumer VRAM without quantization or distillation.

It exists to minimize token costs by keeping DeepSeek's prefix cache stable across long sessions, while treating models and plugins as configuration rather than code.

It encodes senior engineering workflows into structured Markdown prompts so AI agents stop winging it.

It exists to teach developers how to build generative AI applications through 21 modular lessons that mix conceptual explainers with working Python and TypeScript code.

A structured 12-week curriculum that teaches symbolic AI, neural nets, and even genetic algorithms—without pretending deep math or cloud ML don't exist.

Taste-Skill is a collection of portable prompt files that teach coding agents to generate layouts, type, and motion that don’t look like they came from a 2015 Bootstrap tutorial.

Because making AI agents read Twitter, Reddit, or Bilibili usually means wrestling with a dozen scrapers, API keys, and cookie jars.

Strix deploys autonomous LLM agents to attack your applications and validate findings with working proof-of-concepts, trading the false-positive noise of static scanners for actual exploits.

OfficeCLI treats Word, Excel, and PowerPoint as structured data you manipulate with single commands, not COM objects you wrestle into submission.

Firecrawl turns search, scraping, and browser interaction into a single API so your agents can read the web without wrestling with proxies, rate limits, or JavaScript rendering.

huashu-design is an agent skill that converts a brief sentence into a finished HTML prototype, MP4 animation, or editable PowerPoint.

It exists because clicking 'generate' isn't enough when you need to control every model, parameter, and preprocessing step.

It forces your AI coding agent to ask what you're building, write a plan, and get sign-off before it writes a single line of code.

Built in a fit of rage after a $16 “open source” web-to-Markdown tool gated features behind API tokens.

Voicebox is a local-first alternative to ElevenLabs and WisprFlow that clones voices, dictates anywhere, and gives AI agents a mouthpiece — all without shipping audio to the cloud.

It gives your terminal an AI pair programmer that can either build freely or explore cautiously without touching your files.

It teaches how LLMs work by implementing tokenization, attention, pretraining, and finetuning in pure PyTorch, one notebook at a time.

Most AI agents treat your editor like a stranger; this one wires the LSP, debugger, and filesystem into the same brain.

The repo that launched a thousand AI hype cycles now wants to be the WordPress of autonomous agents.

A local proxy that reroutes Claude Code's API calls to 17 other providers, letting you use the official CLI with free or local models instead.