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.
886 projects
Together AI
Together AI's MIT-licensed design skill for Claude Code, Cursor, and Codex that refuses to look AI-generated — picking a macrostructure and theme, then running 57 slop-test gates and a self-critique to push output off the defaults models default to.
wonderwhy-er (Eduards Ruzga)
An MIT-licensed MCP server that gives AI assistants real terminal control, filesystem search, and diff-based file editing — plus in-memory code execution and native Excel/PDF/DOCX handling — using host client subscriptions instead of metered API tokens.
Open Interpreter
An Apache-2.0 coding agent — a Rust fork of OpenAI's Codex — optimized to get frontier-like performance from low-cost, open models like Kimi K3, Qwen, and DeepSeek, with swappable agent harnesses plus ACP and Codex compatibility.
nyra health
A verbatim variant of Whisper from nyra health that transcribes every word — fillers, stutters, and false starts included — with crisp word-level timestamps. 1st on the OpenASR verbatim leaderboard, INTERSPEECH 2024, CC BY-NC licensed.
Roblox
Roblox's open-source foundation model for 3D intelligence. Cube 3D v0.5 generates meshes from text by tokenizing shapes, and the May 2026 CubePart release adds open-vocabulary, part-controllable generation of engine-ready assets.
DeepSeek AI
DeepSeek AI's MIT-licensed vision-language model that treats text as pixels — compressing long contexts ~10x into vision tokens at ~97% decode precision. A DeepEncoder + 3B-MoE OCR pipeline that outputs Markdown, tables, and formulas at scale.
bytedance
ByteDance's Apache-2.0 unified video generation/editing framework pairing a 7B MLLM semantic planner with a 14B DiT renderer — first-tier against commercial editors in blind human-preference arena voting.
debpalash
The open-source ElevenLabs alternative: local zero-shot voice cloning, video dubbing, dictation, and audiobook production across 646 languages, orchestrating 14 TTS + 10 ASR engines on your own hardware (AGPL-3.0).
kyutai-labs
Kyutai's MIT-licensed 100M-parameter TTS that runs ~6x real-time on 2 CPU cores with ~200ms latency — streaming audio, voice cloning from a WAV, 6 languages, and zero GPU required.
unclecode
The most-starred open-source web crawler on GitHub (Apache-2.0): turns any site into clean, LLM-ready Markdown for RAG and agents, with async crawling, structured extraction, and a secure-by-default Docker server.
mozilla-ai
Mozilla.ai's Apache-2.0 project that collapses an LLM and its runtime into a single cross-platform executable — built on llama.cpp and Cosmopolitan Libc, running install-free on macOS, Linux, Windows, and BSD.
pytorch
PyTorch's official BSD-3 platform for large-scale LLM pretraining — a clean, minimal reference codebase composing FSDP2, tensor/pipeline/context parallelism, torch.compile, and Float8/MXFP8 low-precision training.