Open Source
Explore the latest AI open-source projects from GitHub and HuggingFace.
Explore the latest AI open-source projects from GitHub and HuggingFace.
Explore the latest AI open-source projects from GitHub and HuggingFace. Discover projects across categories like LLM, Vision, Audio, and more.
489 projects
rohitg00
Open-source persistent memory for AI coding agents with 95.2% R@5 retrieval, 92% fewer tokens, and 51 MCP tools.
Hmbown
Rust-based terminal coding agent for DeepSeek V4 models with streaming reasoning, 1M context, MCP support, and workspace rollback.
fastgs
CVPR 2026 Highlight: trains 3D Gaussian Splatting scenes in roughly 100 seconds with multi-backbone acceleration.
agentscope-ai
Open-source personal AI assistant on AgentScope runtime, supporting local and cloud deployment across major chat platforms.
lightseekorg
Speed-of-light open-source LLM inference engine targeting TensorRT-LLM performance for agentic workloads on Blackwell GPUs.
ikawrakow
Community-maintained llama.cpp fork delivering SOTA quantization (IQ_KT trellis, IQK family, MXFP4), FlashMLA for DeepSeek, and substantial CPU + hybrid GPU/CPU speedups for local LLM inference.
thu-ml
Tsinghua's SageAttention delivers 2-5x quantized attention speedup over FlashAttention with no accuracy loss, supporting Hopper, Ada, Ampere, and Blackwell GPUs across language, image, and video models.
deepseek-ai
DeepSeek's expert-parallel communication library delivering high-throughput MoE dispatch/combine kernels with FP8 support, NVLink + RDMA backends, and JIT-compiled GPU kernels for SM90 Hopper.
datawhalechina
Datawhale's 44.8k-star open tutorial teaching LLM agent development from scratch, covering ReAct, LangGraph, AutoGen, memory systems, and full capstone projects.
HKUDS
Agent-native trading platform where AI agents publish signals, copy trades, and collaborate with humans across all major markets via a collective intelligence layer.
z-lab
Lightweight block diffusion model that accelerates LLM inference via speculative decoding, supporting vLLM, SGLang, Transformers, and MLX with sub-2x walltime overhead.
rtk-ai
Single Rust binary CLI proxy that cuts LLM token usage 60-90% by filtering and compressing command outputs before they reach the AI agent's context window.