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.
625 projects
openai
Official Python framework for multi-agent workflows with handoffs, sandbox execution, guardrails, and tracing — supports 100+ LLM backends beyond OpenAI.
tensorzero
Open-source LLMOps platform unifying LLM gateway, observability, evaluation, optimization, and experimentation — built in Rust for sub-1ms p99 latency with 15+ provider support.
microsoft
npm-style dependency management for AI agents — declare skills, prompts, plugins, and MCP servers in a single apm.yml manifest for reproducible, shareable agent configurations.
aaif-goose
Open-source AI agent by Block/AAIF that installs, executes, edits, and tests with any LLM across macOS, Linux, and Windows with 70+ MCP extensions.
atilaahmettaner
MCP server bringing real-time market data, 30+ technical indicators, 6 backtesting strategies, Reddit sentiment analysis, and multi-exchange crypto screening directly into Claude Desktop and compatible AI clients.
huggingface
Hugging Face's barebones agent library where agents write Python code as actions — 30% fewer steps than JSON agents, ~1,000-line auditable core, 100+ LLMs via LiteLLM.
z-lab
DFlash accelerates LLM inference through block diffusion-based speculative decoding, supporting vLLM, SGLang, Hugging Face Transformers, and MLX backends across NVIDIA and Apple Silicon hardware.
HKUDS
Ultra-lightweight personal AI agent by HKUDS with 99% less code than comparable systems, supporting 13 messaging platforms and multi-provider LLMs via MIT license.
Polymarket
Open-source developer framework for building AI trading agents on Polymarket prediction markets, featuring RAG pipelines, multi-source data aggregation, sentiment analysis, and automated trade execution.
obra
Superpowers is a 154,000-star MIT-licensed agentic methodology framework available across all major AI coding platforms, enforcing a six-stage development pipeline with mandatory TDD, git-worktree isolation, and two-stage code review to produce reliable, production-ready code.
lsdefine
GenericAgent is a 3,300-line open-source autonomous agent that grows a personal skill tree from every solved task—each new capability reduces future token consumption by up to 6x, supporting Claude, Gemini, Kimi, and MiniMax.
Magika is a Google-built open-source deep learning tool that detects 200+ file content types with ~99% accuracy in ~5ms on CPU, trained on 100 million samples and deployed in Gmail, Drive, and Safe Browsing.