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
agno-agi
Apache 2.0 SDK for building and running agent platforms (40k+ stars, formerly Phidata) — framework-agnostic runtime with 50+ API endpoints, pluggable storage in your DB, 100+ tool integrations, JWT/RBAC, OpenTelemetry tracing, cron scheduling, and a control-plane UI you can deploy in your own cloud.
Tencent-Hunyuan
Tencent's ICML 2026 universal feed-forward 3D reconstruction model — emits point clouds, multi-view depth, camera parameters, surface normals, and 3D Gaussian Splats in one pass, with Multi-Modal Prior Prompting for optional camera/depth/intrinsics conditioning.
OpenGVLab
Shanghai AI Lab's 4B unified multimodal model that handles understanding, reasoning, image generation, and editing in one framework — MIT-licensed, with coordinated release of the GenEditEvalKit evaluation toolkit and the TextEdit benchmark.
MemPalace
Local-first AI memory system with verbatim storage and pluggable retrieval (ChromaDB default, SQLite/Qdrant/pgvector supported) — 96.6% R@5 on LongMemEval raw, 99%+ with LLM rerank, zero API calls and zero cloud dependency.
mvanhorn
Claude Code agent skill that researches any topic across Reddit, X, YouTube transcripts, HN, Polymarket, TikTok, Instagram, Threads, Bluesky, GitHub, and the web — synthesizes engagement-weighted briefs as shareable HTML files with 1,012 passing tests.
RyanCodrai
Rust + Python vector index implementing Google's TurboQuant quantization — 16x memory compression, FAISS-beating recall (R@1 +0.4 to +3.4), SIMD-optimized for ARM and AVX-512, with drop-in LangChain, LlamaIndex, Haystack, and Agno adapters.
zai-org
Apache-2.0 vision-language model family from Zhipu AI with a Thinking-mode switch, native multimodal function calling, and a scalable RL pipeline aimed at complex visual reasoning — long documents, GUI agents, and long-video event recognition.
Apple
Apple's open-source FastViTHD hybrid vision encoder cuts VLM Time-to-First-Token by up to 85x vs LLaVA-OneVision while matching accuracy — 0.5B/1.5B/7B variants with iOS demo and Apple Silicon-optimized inference for on-device multimodal.
OpenBMB
Apache-2.0 pocket-sized multimodal LLM with mixed 4x/16x visual token compression, native iOS/Android/HarmonyOS deployment, and a 9B omnimodal MiniCPM-o variant that approaches Gemini 2.5 Flash on vision and speech.
QwenLM
Apache-2.0 vision-language model family from Alibaba's Qwen team spanning 2B dense to 235B MoE, with native 256K context, GUI agent capability, and 3D grounding — the cleanest open VLM choice for edge-to-cloud deployments in 2026.
OpenWhispr
MIT-licensed cross-platform voice dictation and meeting transcription app with local Whisper/Parakeet models, BYOK cloud LLMs (GPT-5/Claude/Gemini/Groq), on-device speaker ID, and an MCP-integrated voice agent.
KoljaB
MIT-licensed low-latency Python library for real-time speech-to-text with WebRTC/Silero VAD, Porcupine/OpenWakeWord wake words, FastAPI streaming server, and pluggable engines (faster_whisper, Moonshine, sherpa-onnx).