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.
760 projects
PKU-YuanGroup
PKU-YuanGroup's open-source 14B real-time long video generation model — minute-scale, coherent video at 19.5 FPS on a single H100 without anti-drifting tricks.
Sapient Intelligence
A 1B-parameter text model and full pretraining framework on the Hierarchical Reasoning Model architecture — pretrain a foundation model from scratch for about $1,000.
Tencent
Tencent's open-source, RustVMM/KVM-based secure sandbox for AI agents: hardware-isolated microVMs that boot in under 60ms with E2B SDK compatibility.
GeeeekExplorer
A from-scratch vLLM reimplementation in ~1,200 lines of readable Python with prefix caching, tensor parallelism, torch.compile, and CUDA graphs — vLLM-comparable offline speed.
FoundationAgents
An MIT-licensed, open-source general AI agent from the MetaGPT team — run an autonomous, browser- and code-using assistant with your own LLM keys, no invite code.
facebookresearch
Oxford VGG and Meta AI's CVPR 2025 Best Paper: a feed-forward transformer that infers camera poses, depth, point maps, and 3D tracks from images in seconds.
resemble-ai
Resemble AI's open-source TTS family with zero-shot voice cloning from seconds of audio, a low-latency Turbo model, and 23+ language multilingual synthesis (MIT).
OpenBMB
OpenBMB's efficient multimodal LLM series that runs image and video understanding on phones — MiniCPM-V 4.6 hits GPT-4V-class quality at just 1.3B params (Apache-2.0).
m-bain
A fast open-source ASR pipeline on top of Whisper: word-level timestamps via forced alignment, 70x realtime batched inference, and speaker diarization (BSD-2).
bradautomates
An MIT-licensed /watch command that lets an AI agent 'watch' any video — captions first, then yt-dlp + ffmpeg frame extraction and Whisper fallback — grounding answers in what's actually on screen.
stanfordnlp
Stanford NLP's framework for programming — not prompting — language models: declare tasks as typed Python modules and let optimizers auto-tune prompts, few-shot examples, and weights (MIT).
huggingface
Hugging Face's open-source voice-agent pipeline (VAD -> STT -> LLM -> TTS) exposed through an OpenAI Realtime-compatible WebSocket API, with every component swappable and self-hostable (Apache-2.0).