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
vercel-labs
Vercel Labs' Apache-2.0 coding agent written in Zig — a 7.8 MiB Unix-shell-style CLI that also embeds into editors via ACP or into JavaScript hosts as WebAssembly.
pwilkin
A standalone C++/GGML port of Microsoft's TRELLIS.2-4B image-to-3D pipeline — three flow DiTs, mesh extraction and UV-textured GLB export with no Python at runtime, on CUDA, ROCm, Vulkan or CPU.
mudler
The LocalAI team's from-scratch C++20 port of vLLM's serving core — a 66 MiB binary against a 9.1 GiB install, token-for-token identical output, GGUF native, on CUDA, CPU, Metal and Vulkan.
tiiuae
TII's Apache-2.0 inference stack for Falcon Perception (0.6B) and Falcon OCR (0.3B) — early-fusion single-Transformer VLMs that emit full-resolution masks in one shot and read documents to text, LaTeX or HTML.
QuentinFuxa
Apache-2.0 self-hosted real-time speech-to-text: SOTA simultaneous-ASR commit policies, Sortformer diarization, 200-language translation, and swappable Whisper/Voxtral/Canary/Qwen3 backends behind OpenAI- and Deepgram-compatible APIs.
OHF-Voice
The Open Home Foundation's fast, fully local neural TTS engine behind Home Assistant and NVDA — VITS voices on ONNX Runtime across 44 locales, with v1.7.0 adding a pitch-accent-aware Japanese phonemizer.
PatterAI
An MIT-licensed Python and TypeScript SDK that gives an AI agent a phone number in four lines, orchestrating the full voice stack across 27+ swappable LLM, STT, TTS, realtime and carrier providers.
Yuliang-Liu
A HUST-affiliated document-native vision encoder released as a reusable backbone in 21M-113M sizes, with a 113M-image multilingual pretraining corpus and a 0.7B parsing model scoring 83.3 on MDPBench.
NanoNets
NanoNets' MIT-licensed context layer that turns a repo into a linked markdown code graph coding agents can read, cutting tool calls 46% and tokens 42% with a tree-sitter tier that needs no API key.
yetone
A cross-platform team chat where AI agents sit on the same roster as humans, with each agent's brain running either on Cumora's cloud pods or on your own machine via a local Claude Code or Codex CLI.
perplexityai
Perplexity's Go tool for AI agent endpoint security: it normalizes hooks, OTLP logs and on-disk session artifacts into one CEL-evaluated event model for local detection, opt-in pre-action blocking and after-the-fact forensics.
drumih
A model-specific Swift and Metal runtime that runs Google's Gemma 4 26B-A4B in about 2 GB of RAM on any Apple Silicon Mac by keeping a 1.35 GB core resident and streaming routed experts from SSD per token.