Personality-driven AI agents โ Frontend Dev, Backend Architect, Growth Hacker & more. Designed to act like a real dev team.
Orchestrate role-playing, autonomous AI agents for collaborative tasks. 44k+ GitHub stars, used by 60% of Fortune 500.
Multi-agent conversation framework by Microsoft. Handles hundreds of concurrent agents with async communication and Docker sandboxing.
Graph-based agent framework from LangChain. State machines, cycles, conditional branching. Fastest benchmarks, lowest latency.
Open-source platform for building generative AI apps with visual workflows, built-in RAG, and multi-model support. 129k+ stars.
Systematically evaluate and red-team LLM apps. Tests OWASP LLM Top 10 & NIST AI RMF for prompt injection, PII leaks & more.
Design skill for Claude that fights "AI slop" โ 17 commands and curated anti-patterns to force unique frontend code instead of generic UIs.
Terminal-based AI pair programmer with deep Git integration. Auto-commits changes with descriptive messages. 100+ language support.
Let LLMs run code (Python, JS, Shell) locally. Natural-language terminal interface โ create files, control browsers, analyze data.
Daniel Miessler's modular AI framework โ 100+ curated prompt patterns for extracting wisdom, summarizing, analyzing, and creating.
Self-hosted AI coding assistant. Drop-in GitHub Copilot replacement with full privacy. VS Code, JetBrains, and Vim extensions.
Interactive CLI for Claude that can create files, execute commands, manage projects, and do multi-step coding tasks autonomously.
Data-centric RAG framework โ connect PDFs, databases, APIs to LLMs. Index, query, and retrieve from any data source.
Production-ready RAG engine with deep document understanding, built-in agent capabilities, and one of GitHub's fastest-growing projects.
Full-stack desktop RAG app โ chat with docs, multi-user, supports any LLM. Built-in vector DB, web scraping, and agent skills.
Unified context DB for AI agents with hierarchical retrieval (L0โL1โL2). File-system memory via viking:// protocol.
The standard reinforcement learning environment suite (OpenAI Gym successor) โ Atari, MuJoCo, CartPole, robotics & custom envs. Foundation for all RL research. 8k+ โญ.
Knowledge graphs + multi-agent swarm simulation to forecast real-world scenarios โ policy outcomes, market trends, and more.
Open Agent Social Interaction Simulations โ scales to 1M agents on Twitter & Reddit platforms. MiroFish's underlying engine by CAMEL-AI.
Stanford's landmark "Smallville" โ 25 AI agents with memory, reflection, and planning living believable lives in a virtual town. 18k+ โญ.
OpenBMB's multi-agent simulation platform โ test AI agents in shared interactive environments for task-solving and social dynamics.
NeurIPS 2025 โ simulates financial markets with LLM agents modeling individual investor behavior and social interactions.
Enterprise-grade production multi-agent orchestration โ universal swarm framework for complex project management and workflows.
Run LLMs locally with one command. Supports Llama, Mistral, Gemma & more. OpenAI-compatible API, cross-platform desktop apps.
The engine that made local LLMs possible โ run Llama, Mistral, Phi on CPU/GPU with quantization (GGUF). Powers Ollama, LM Studio, and every local AI tool. 78k+ โญ.
Run LLM inference across multiple everyday devices โ cluster laptops, phones, GPUs into one AI supercomputer. No master node, peer-to-peer. 25k+ โญ.
Microsoft's deep learning optimization library โ train 100B+ parameter models on limited GPUs with ZeRO optimizer, pipeline parallelism, and 10x speedups. 35k+ โญ.
2-5x faster LLM fine-tuning with 70-80% less VRAM. Custom CUDA kernels, no accuracy loss. Supports Llama, Mistral, DeepSeek.
Meta's official Llama model repository โ the most important open-weight LLMs in existence. Llama 3/4, model cards, usage guides, and reference implementations.
Beautiful desktop GUI for fine-tuning, evaluating, and chatting with open LLMs locally. LoRA, RLHF, quantization, and model comparison โ no code needed.
Karpathy's minimal framework โ train a GPT-2 level model for ~$100 in 4 hours. Full stack: tokenizer, pretraining, chat UI.
All-in-one fine-tuning framework with web UI. Supports 100+ models, LoRA/QLoRA/RLHF/DPO. 42k+ stars.
Automated directional ablation โ suppresses refusal vectors in local LLMs, removing safety alignment while preserving intelligence.
The most popular Stable Diffusion web interface โ txt2img, img2img, inpainting, extensions ecosystem, and one-click setup. 145k+ โญ.
HuggingFace's diffusion model library โ the standard for SDXL, FLUX, Sora-style video, and audio generation. 27k+ โญ.
Black Forest Labs' FLUX โ currently the best open-source text-to-image model. Outperforms Midjourney v6 on benchmarks. The new standard. 20k+ โญ.
100% open-source node-based workflow for AI image, video, 3D & audio generation. Supports Flux, SDXL, and custom pipelines.
82M parameter open-weight TTS model. Natural-sounding speech in English, French, Korean, Japanese & Mandarin. Apache licensed.
The breakout star of 2026 โ runs locally, browses web, fills forms, runs code, controls smart home. Writes its own new skills. 250k+ โญ.
Make websites accessible for AI agents โ automate browsing, form filling, data extraction, and any web task with LLM-powered automation.
Low-code visual platform for building AI agents & RAG workflows โ drag-and-drop interface on LangChain. 140k+ โญ.
Ultra-realistic voice cloning from 10-30s of audio. Emotion control, multilingual, real-time generation. 26k+ โญ.
3Blue1Brown's animation engine for explanatory math videos โ programmatically render equations, geometric transforms, and scenes in Python. 73k+ โญ.
DeepMind's landmark AI โ solved the 50-year protein folding problem, predicting 3D protein structures from sequence alone. 200M+ structures released. Nobel Prize-adjacent. 14k+ โญ.
IBM's open-source quantum computing SDK โ build & run quantum circuits on real IBM quantum hardware or simulators. The standard for quantum ML. 5k+ โญ.
Every LLM has a silent flaw: layer burial. Each layer adds equally to the stack โ researchers deleted 40%+ of LLaMA's layers and barely noticed. Early layers get buried and forgotten. Moonshot AI fixed this for nearly free: group layers into chunks โ summarize each chunk โ every layer attends over those summaries. No need to store all activations. Baked into training, not inference. Result: 48B params, 1.4T tokens, matches a baseline that used 25% MORE compute โ better math, coding, and reasoning. Deeper now means smarter, not just bigger.
MIT CSAIL's "Neural Thickets" discovery: around pretrained weights lies a dense thicket of task-expert solutions. RandOpt randomly perturbs weights N times, picks the top K, and ensembles by majority vote โ matching PPO/GRPO post-training in O(1) steps, fully parallel, no gradient descent.
Stanford's AI research writing system โ generates Wikipedia-quality long-form articles from scratch. Multi-perspective knowledge curation with source citations. 18k+ โญ.
The reasoning model that shook the industry โ matches o1 performance via RL-trained chain-of-thought. Open weights, distilled checkpoints, and full training methodology. 50k+ โญ.
3D Gaussian Splatting for real-time radiance field rendering โ reconstruct photorealistic 3D scenes from photos in minutes. The NeRF successor. 15k+ โญ.
Meta's Segment Anything Model โ zero-shot image segmentation that identifies and masks any object in any image with a single click or prompt. 47k+ โญ.
Meta's audio generation library โ MusicGen creates full music tracks from text prompts; AudioGen produces sound effects. State-of-the-art generative audio. 22k+ โญ.
High-performance parallel programming in Python โ write GPU-accelerated physics simulations, fluid dynamics, and graphics shaders with Python syntax. 26k+ โญ.
Rive's low-level C++ renderer & runtime โ state machine-driven interactive animations for AI simulations, math visuals, games & web. Cross-platform (Web/iOS/Android/Desktop).
High-throughput, memory-efficient LLM inference and serving. PagedAttention, continuous batching, OpenAI-compatible API.
Self-hosted ChatGPT-style interface for Ollama & OpenAI-compatible APIs. Rich UI, RAG, web search, multi-model support.
GPT-4-class open model with Mixture-of-Experts architecture. 671B total params, 37B active per token. 100k+ โญ.
The most powerful node-based Stable Diffusion GUI โ modular workflow editor for image/video generation. Supports SDXL, FLUX, ControlNet, LoRA, and custom nodes. 75k+ โญ.
Make AI agents that control web browsers autonomously โ navigate, click, fill forms, extract data. Built on Playwright with vision + DOM understanding. 60k+ โญ.
Google's open-source AI agent โ Gemini multimodal model right in your terminal. Code, debug, create. 87k+ โญ.
Anthropic's on-device AI coding assistant โ privacy-first, terminal-based, with deep codebase understanding. 46k+ โญ.
Open-source workflow automation with native AI capabilities. Visual canvas, 400+ integrations, LangChain built-in. 160k+ โญ.
Multi-agent framework that assigns GPT roles โ Product Manager, Architect, Engineer โ to collaboratively build software. 62k+ โญ.
The OG autonomous AI agent platform โ create, deploy, and manage agents that accomplish tasks independently. 175k+ โญ.
LAION's open-source ChatGPT alternative โ community-built conversational AI with 600k+ human feedback messages. RLHF training data and models. 37k+ โญ.
AI-powered software development agents (formerly OpenDevin) โ autonomous coding, debugging, and deployment. Full IDE, terminal, and browser access. 50k+ โญ.
Use ALL major AI models with NO API token โ Claude, ChatGPT, Gemini, DeepSeek, Grok, Qwen, Kimi, Manus. Web UI, HTTP API, and CLI/TUI. Tool calling and streaming supported.
The foundational framework for building LLM-powered apps โ chains, agents, tools, memory, and retrieval. 123k+ โญ.
Community-curated .cursorrules library โ best practices, coding patterns, and AI assistant rules for Cursor IDE.
Deepset's production-ready framework for building RAG, search, and NLP pipelines. Modular, extensible, enterprise-grade. 42k+ โญ.
Convert any URL to clean, LLM-friendly markdown โ just prefix with r.jina.ai. Handles JS rendering, PDFs, and complex layouts. The simplest RAG data pipeline. 20k+ โญ.
Build production-ready conversational AI in minutes โ chat UI, multi-step reasoning, file uploads, streaming. 8k+ โญ.
Self-improving memory layer for personalized AI experiences. Give agents long-term context that learns and grows. 25k+ โญ.
Build multi-modal agents with memory, knowledge, tools, and reasoning. Production-ready with function calling. 20k+ โญ.
Princeton's autonomous software engineer โ fixes GitHub issues, writes patches, and resolves bugs independently. 18k+ โญ.
Prompt, run, edit, and deploy full-stack web apps โ AI-powered development in the browser with StackBlitz. 28k+ โญ.
Chat with your database or DataFrame using natural language. SQL generation, data visualization, and smart analysis. 15k+ โญ.
Enterprise natural language to SQL engine โ connect databases and query them in plain English with high accuracy.
Fast, template-based vulnerability scanner. 8000+ community templates for CVEs, misconfigs, and exposed panels. 23k+ โญ.
Lightweight web security testing toolkit โ modern Burp Suite alternative. Fast, extensible, with built-in AI assistance.
The OSINT Swiss Army Knife for phone numbers โ scan numbers for carrier, location, line type, and social footprint. Web UI + CLI. 13k+ โญ.
Hunt down social media accounts by username across 400+ social networks simultaneously. The definitive username OSINT tool. 62k+ โญ.
Find a person's profile across 1000+ social media sites โ API, CLI, and Web App. Detection modules for names, usernames, emails, and phone numbers.
Interactive shell for Instagram OSINT โ analyze any account's followers, following, geotagged photos, stories, and linked contacts by nickname.
Username enumeration engine with a massive JSON dataset of 500+ sites โ the data backbone that powers Sherlock, SpiderFoot, and recon-ng.
Simplistic OSINT engine for breach database intelligence โ search Instagram, Twitter, phone numbers, websites, and compromised card data from public sources.
Truecaller phone number lookup library โ retrieve caller name, carrier, location, and spam score. Supports bulk searches for mass reconnaissance.
Open Multi-Agent Interactive Classroom from Tsinghua โ immersive multi-agent learning with lesson generation, classroom simulation & OpenClaw integration.
Won the AI Math Olympiad Progress Prize โ fine-tuned 7B model with tool-integrated reasoning. Open-source math AI breakthrough.
AI agent system for solving IMO 2025 problems โ uses Gemini, GPT, and XAI APIs with multi-model reasoning chains.
LangGraph-based self-verification pipeline for IMO-level problems โ agent reasons, verifies, and corrects its own solutions.
AIMO Kaggle competition solutions โ notebooks, fine-tuning recipes, and evaluation pipelines for mathematical reasoning.
Interactive theorem prover and programming language โ the backbone of AI-assisted mathematical proof (used by AlphaProof). 8k+ โญ.
The largest formalized mathematics library โ 1M+ lines covering algebra, analysis, topology, number theory. Foundation for AI math proving.
HuggingFace's open-source robotics platform โ real-world AI for manipulation, locomotion, and embodied learning. 8k+ โญ.
NVIDIA's robot learning framework built on Isaac Sim โ GPU-accelerated physics, RL training, sim-to-real transfer. 3k+ โญ.
Open vision-language-action model for robot manipulation โ 7B param generalist agent from Stanford & Google DeepMind. 2k+ โญ.
OpenAI's robust speech recognition โ multilingual transcription & translation. The gold standard for speech-to-text. 75k+ โญ.
Suno's transformer-based text-to-audio โ generates realistic speech, music, sound effects, and non-verbal cues. 37k+ โญ.
Fast local neural text-to-speech โ runs on Raspberry Pi, 30+ languages, ONNX runtime. Perfect for offline voice assistants. 8k+ โญ.
Open-source voice synthesis studio โ multi-engine cloning, emotions & paralinguistic tags, post-processing effects, unlimited generation length, async queue, voice profile management, and a stories editor.
Headless browser built from scratch for AI & automation โ full JS execution via CDP, 100x faster than Chrome headless. Written in Zig for raw performance. The browser AI agents deserve.
Hands-on AI engineering tutorials โ RAG pipelines, multi-agent systems, MCP servers, fine-tuning, podcast gen, OCR & more.
Collection of AI projects covering Machine Learning, Deep Learning & Generative AI โ practical implementations with advanced techniques & concepts.
500+ real-world projects across ML, Deep Learning, Computer Vision & NLP โ the largest curated collection of AI projects with source code.
Industry-grade end-to-end Generative AI projects โ LLMs, RAG pipelines, AI agents & multimodal apps with deployment guides.
Portfolio of Generative AI projects & AI agents โ image generation, code assistants, text processing & multimodal applications.
100 curated projects covering CNNs, RNNs, GANs, Computer Vision & NLP โ from beginner classification to advanced generative models.
Comprehensive collection of deep learning tutorials, projects, books & communities โ covering neural networks, CNNs, RNNs & RL.
The definitive curated list of ML frameworks, libraries & tools โ 66k+ stars. Covers every ML domain from CV to NLP to reinforcement learning.
Curated list of inspiring project ideas for ML, NLP, Computer Vision & recommender systems โ perfect for portfolio building.
Comprehensive guide to Generative AI โ tutorials, tools & libraries for text, image & multimodal AI. Covers LLMs, diffusion & more.
Garry Tan's Claude Code virtual engineering team โ orchestrates multiple AI agents as specialized engineers to build full projects collaboratively. 51k+ โญ.
NousResearch's autonomous agent with a self-improving learning loop โ adapts strategies, refines tool usage, and gets smarter over time. 14k+ โญ.
Comprehensive agentic skills framework & developer methodology โ curated prompt engineering patterns, workflows, and skills for AI-powered development. 117k+ โญ.
Open-source orchestration platform for building agent-powered companies โ workflow automation, task management, and multi-agent coordination. 35k+ โญ.
Anthropic's official agent skills specification & library โ standardized skill definitions, best practices, and reference implementations for Claude agents. 104k+ โญ.
100% open-source terminal AI coding agent โ supports 75+ LLM providers, TUI interface, LSP integration, and custom commands. Model-agnostic, privacy-first. 18k+ โญ.
DeepMind's AI that does frontier research โ autonomously solves open math problems (Erdลs problems, arithmetic geometry), writes publishable papers, and uses 100x less compute than prior systems. Powered by Gemini with novel self-verification.
Claude Code plugin that captures everything from coding sessions, compresses it with AI, and injects relevant context into future sessions. Persistent memory across conversations.
The definitive curated list of Claude Code skills, hooks, slash-commands, agent orchestrators, CLAUDE.md files, status lines, and plugins. The ecosystem directory.
AI skill that injects professional design intelligence โ 161 industry-specific reasoning rules, intelligent design system generation, multi-platform support. Turns Claude into a UI/UX expert.
Near-optimal KV cache compression for LLM inference โ 3-bit keys, 2-bit values with custom Triton kernels + vLLM integration. Benchmarked on RTX 5090 & 8x RTX 3090.
Traffic light for AI agents โ TypeScript/Node multi-agent orchestrator with shared state, guardrails, and adapters for 17 AI frameworks including LangGraph, CrewAI & AutoGen. MCP server built-in.