https://github.com/alvinreal/awesome-opensource-ai
Curated list of the best truly open-source AI projects, models, tools, and infrastructure.
https://github.com/alvinreal/awesome-opensource-ai
List: awesome-opensource-ai
agents ai artificial-intelligence awesome awesome-list generative-ai llm machine-learning mlops open-source open-source-ai rag
Last synced: about 2 months ago
JSON representation
Curated list of the best truly open-source AI projects, models, tools, and infrastructure.
- Host: GitHub
- URL: https://github.com/alvinreal/awesome-opensource-ai
- Owner: alvinreal
- License: other
- Created: 2026-03-24T15:58:22.000Z (2 months ago)
- Default Branch: main
- Last Pushed: 2026-04-17T10:06:43.000Z (about 2 months ago)
- Last Synced: 2026-04-17T10:09:58.023Z (about 2 months ago)
- Topics: agents, ai, artificial-intelligence, awesome, awesome-list, generative-ai, llm, machine-learning, mlops, open-source, open-source-ai, rag
- Language: Python
- Homepage: https://awesomeosai.com
- Size: 1.87 MB
- Stars: 2,632
- Watchers: 23
- Forks: 243
- Open Issues: 12
-
Metadata Files:
- Readme: README.md
- Contributing: CONTRIBUTING.md
- License: LICENSE
Awesome Lists containing this project
- ultimate-awesome - awesome-opensource-ai - Curated list of the best truly open-source AI projects, models, tools, and infrastructure. (Other Lists / Vue Lists)
- awesome-ai-tools - Awesome Open Source AI - source AI projects worth knowing. (📋 Related Awesome Lists / Newsletters)
- awesome-awesome-artificial-intelligence - Awesome Open Source AI - source AI models, libraries, infrastructure, and developer tools. |  | (Table of Contents)
README
# Awesome Open Source AI
*A curated list of **battle-tested, production-proven** open-source AI models, libraries, infrastructure, and developer tools. Only elite-tier projects make this list. Updated April 2026.*
[](https://awesome.re)
[](./CONTRIBUTING.md)
[](./LICENSE)
by **Boring Dystopia Development**
---
**[ 🌱 Emerging ](./EMERGING.md)** • **[ Explore the List ](#-contents)** • **[ Submission Guidelines ](#contributing)** • **[ License ](#license)**
## 📋 Contents
- [🧬 1. Core Frameworks & Libraries](#-1-core-frameworks--libraries)
- [🧠 2. Open Foundation Models](#-2-open-foundation-models)
- [⚡ 3. Inference Engines & Serving](#-3-inference-engines--serving)
- [🤖 4. Agentic AI & Multi-Agent Systems](#-4-agentic-ai--multi-agent-systems)
- [🔍 5. Retrieval-Augmented Generation (RAG) & Knowledge](#-5-retrieval-augmented-generation-rag--knowledge)
- [🎨 6. Generative Media Tools](#-6-generative-media-tools)
- [🛠️ 7. Training & Fine-tuning Ecosystem](#section-7)
- [📊 8. MLOps / LLMOps & Production](#-8-mlops--llmops--production)
- [📈 9. Evaluation, Benchmarks & Datasets](#-9-evaluation-benchmarks--datasets)
- [🛡️ 10. AI Safety, Alignment & Interpretability](#section-10)
- [🧩 11. Specialized Domains](#-11-specialized-domains)
- [🖥️ 12. User Interfaces & Self-hosted Platforms](#section-12)
- [🧪 13. Developer Tools & Integrations](#-13-developer-tools--integrations)
- [📚 14. Resources & Learning](#-14-resources--learning)
---
### 🧬 1. Core Frameworks & Libraries
> Core libraries and frameworks used to build, train, and run AI and machine learning systems.
#### Deep Learning Frameworks
- **[PyTorch](https://github.com/pytorch/pytorch)**  - Dynamic computation graphs, Pythonic API, dominant in research and production. The current standard for most frontier AI work.
- **[TensorFlow](https://github.com/tensorflow/tensorflow)**  - End-to-end platform with excellent production deployment, TPU support, and large-scale serving tools.
- **[JAX](https://github.com/jax-ml/jax)**  + **[Flax](https://github.com/google/flax)**  - High-performance numerical computing with composable transformations (JIT, vmap, grad). Rising favorite for research and scientific ML.
- **[dm-haiku](https://github.com/google-deepmind/dm-haiku)**  - JAX-based neural network library from Google DeepMind. Elegant functional API with state management, widely used in DeepMind's research. Apache 2.0 licensed.
- **[Equinox](https://github.com/patrick-kidger/equinox)**  - Elegant easy-to-use neural networks and scientific computing in JAX. Callable PyTrees with filtered transformations, seamless interoperability with the JAX ecosystem. Apache 2.0 licensed.
- **[NumPyro](https://github.com/pyro-ppl/numpyro)**  - Probabilistic programming with NumPy powered by JAX for autograd and JIT compilation. Bayesian modeling and inference at scale.
- **[Keras](https://github.com/keras-team/keras)**  - High-level, beginner-friendly API that now runs on multiple backends (TensorFlow, JAX, PyTorch). Perfect for rapid experimentation.
- **[tinygrad](https://github.com/tinygrad/tinygrad)**  - Minimalist deep learning framework with tiny code footprint. The "you like pytorch? you like micrograd? you love tinygrad!" philosophy - simple yet powerful.
- **[PaddlePaddle](https://github.com/PaddlePaddle/Paddle)**  - Industrial deep learning platform from Baidu serving 23+ million developers and 760,000+ companies. China's first independent R&D framework with advanced distributed training and deployment capabilities.
- **[PyTorch Geometric](https://github.com/pyg-team/pytorch_geometric)**  - Library for deep learning on irregular input data such as graphs, point clouds, and manifolds. Part of the PyTorch ecosystem.
- **[timm (PyTorch Image Models)](https://github.com/huggingface/pytorch-image-models)**  - The largest collection of PyTorch image encoders and backbones. 900+ pretrained models including ResNet, EfficientNet, Vision Transformer, ConvNeXt, and more with training and inference scripts. Apache 2.0 licensed.
- **[Triton](https://github.com/triton-lang/triton)**  - Language and compiler for writing highly efficient custom deep-learning primitives. Powers kernel optimizations in PyTorch, JAX, and other frameworks. MIT licensed.
- **[GGML](https://github.com/ggml-org/ggml)**  - Tensor library for machine learning. The foundational C/C++ library powering llama.cpp and many on-device inference engines. MIT licensed.
- **[MLX](https://github.com/ml-explore/mlx)**  - Array framework for machine learning on Apple silicon. Efficient unified memory design with NumPy-like API, automatic differentiation, and multi-device support. MIT licensed.
#### High-Performance Compute Libraries
- **[oneDNN](https://github.com/uxlfoundation/oneDNN)**  - oneAPI Deep Neural Network Library. Cross-platform performance library of basic building blocks for deep learning, optimized for Intel CPUs, GPUs, and Arm architectures. Apache 2.0 licensed.
- **[ONNX](https://github.com/onnx/onnx)**  - Open standard for machine learning interoperability. Open Neural Network Exchange provides an open ecosystem that empowers AI developers to choose the right tools as their project evolves. Apache 2.0 licensed.
- **[IREE](https://github.com/iree-org/iree)**  - Retargetable MLIR-based machine learning compiler and runtime toolkit. Lowers ML models to unified IR that scales from datacenter to mobile and edge deployments. Apache 2.0 licensed.
#### Rust ML Frameworks
- **[Burn](https://github.com/tracel-ai/burn)**  - Next-generation deep learning framework in Rust. Backend-agnostic with CPU, GPU, WebAssembly support.
- **[Candle (Hugging Face)](https://github.com/huggingface/candle)**  - Minimalist ML framework for Rust. PyTorch-like API with focus on performance and simplicity.
- **[linfa](https://github.com/rust-ml/linfa)**  - Comprehensive Rust ML toolkit with classical algorithms. scikit-learn equivalent for Rust with clustering, regression, and preprocessing.
#### Julia ML Frameworks
- **[Flux.jl](https://github.com/FluxML/Flux.jl)**  - 100% pure-Julia ML stack with lightweight abstractions on top of native GPU and AD support. Elegant, hackable, and fully integrated with Julia's scientific computing ecosystem.
- **[MLJ.jl](https://github.com/JuliaAI/MLJ.jl)**  - Comprehensive Julia machine learning framework providing a unified interface to 200+ models with meta-algorithms for selection, tuning, and evaluation. MIT licensed.
- **[ModelingToolkit.jl](https://github.com/SciML/ModelingToolkit.jl)**  - High-performance symbolic-numeric modeling framework for scientific machine learning. Automatically generates fast functions for model components like Jacobians and Hessians with automatic sparsification and parallelization. MIT licensed.
#### NLP & Transformers
- **[spaCy (Explosion AI)](https://github.com/explosion/spaCy)**  - Industrial-strength natural language processing with 75+ languages, transformer pipelines, and production-grade NER, parsing, and text classification.
- **[Transformers (Hugging Face)](https://github.com/huggingface/transformers)**  - The de facto standard library for pretrained NLP models. 1M+ models, 250,000+ downloads/day. BERT, GPT, Llama, Qwen, and hundreds more.
- **[sentence-transformers](https://github.com/UKPLab/sentence-transformers)**  - Classic library for sentence and image embeddings.
- **[tokenizers (Hugging Face)](https://github.com/huggingface/tokenizers)**  - Fast state-of-the-art tokenizers for training and inference.
- **[fairseq2](https://github.com/facebookresearch/fairseq2)**  - FAIR Sequence Modeling Toolkit 2. Complete rewrite of fairseq with modern PyTorch APIs, native support for LLM training (70B+ models), vLLM integration, and first-party recipes for instruction finetuning and preference optimization. MIT licensed.
#### Data Processing & Manipulation
- **[Pandas](https://github.com/pandas-dev/pandas)**  - The gold standard for data analysis and manipulation in Python.
- **[Polars](https://github.com/pola-rs/polars)**  - Blazing-fast DataFrame library (Rust backend) - modern alternative to pandas for large-scale workloads.
- **[cuDF](https://github.com/rapidsai/cudf)**  - GPU DataFrame library from RAPIDS. Accelerates pandas workflows on NVIDIA GPUs with zero code changes using cuDF.pandas accelerator mode.
- **[Modin](https://github.com/modin-project/modin)**  - Parallel pandas DataFrames. Scale pandas workflows by changing a single line of code - distributes data and computation automatically.
- **[Dask](https://github.com/dask/dask)**  - Parallel computing for big data - scales pandas/NumPy/scikit-learn to clusters.
- **[NumPy](https://github.com/numpy/numpy)**  - Fundamental array computing library that powers almost every AI stack.
- **[SciPy](https://github.com/scipy/scipy)**  - Scientific computing algorithms (optimization, linear algebra, statistics, signal processing).
- **[NetworkX](https://github.com/networkx/networkx)**  - Creation, manipulation, and study of complex networks. The foundational graph analysis library for Python data science.
- **[cuGraph](https://github.com/rapidsai/cugraph)**  - GPU graph analytics library with NetworkX-compatible API. 10-100x faster than CPU for large-scale graph algorithms. Apache 2.0 licensed.
- **[Vaex](https://github.com/vaexio/vaex)**  - Out-of-Core hybrid Apache Arrow/NumPy DataFrame for Python. Visualize and explore billion-row datasets at millions of rows per second. MIT licensed.
- **[Datashader](https://github.com/holoviz/datashader)**  - High-performance large data visualization. Renders billions of points interactively without aggregation artifacts. BSD-3-Clause licensed.
- **[Zarr](https://github.com/zarr-developers/zarr-python)**  - Chunked, compressed, N-dimensional array storage. Scalable tensor data format optimized for cloud and parallel computing. MIT licensed.
- **[NVIDIA DALI](https://github.com/NVIDIA/DALI)**  - GPU-accelerated data loading and augmentation library with highly optimized building blocks for deep learning applications. Apache 2.0 licensed.
- **[Narwhals](https://github.com/narwhals-dev/narwhals)**  - Lightweight compatibility layer between DataFrame libraries. Write Polars-like code that works seamlessly across pandas, Polars, cuDF, Modin, and more. MIT licensed.
- **[Ibis](https://github.com/ibis-project/ibis)**  - Portable Python dataframe library with 20+ backends. Write pandas-like code that runs locally with DuckDB or scales to production databases (BigQuery, Snowflake, PostgreSQL) by changing one line. Apache 2.0 licensed.
- **[skrub](https://github.com/skrub-data/skrub)**  - Machine learning with dataframes for dirty categorical data. Preprocessing and feature engineering for heterogeneous data with seamless pandas/Polars integration. BSD-3-Clause licensed.
- **[Oxen](https://github.com/Oxen-AI/Oxen)**  - Lightning fast data version control for machine learning. Optimized for large datasets with efficient diffing, branching, and collaboration. Apache 2.0 licensed.
- **[Pandera](https://github.com/unionai-oss/pandera)**  - Statistical data testing and validation for dataframes. Pydantic-like API for pandas, Polars, and other dataframe libraries with type hints and lazy validation. MIT licensed.
- **[Snorkel](https://github.com/snorkel-team/snorkel)**  - System for quickly generating training data with weak supervision. Programmatically label, build, and manage training data using labeling functions and probabilistic consensus models. Powers Snorkel Flow and used by Google, Apple, and Intel. Apache 2.0 licensed.
#### Classical ML & Gradient Boosting
- **[scikit-learn](https://github.com/scikit-learn/scikit-learn)**  - Industry-standard library for traditional machine learning (classification, regression, clustering, pipelines).
- **[XGBoost](https://github.com/dmlc/xgboost)**  - Scalable, high-performance gradient boosting library. Still dominates Kaggle and tabular competitions.
- **[LightGBM](https://github.com/microsoft/LightGBM)**  - Microsoft's ultra-fast gradient boosting framework, optimized for speed and memory.
- **[CatBoost](https://github.com/catboost/catboost)**  - Gradient boosting that handles categorical features natively with great out-of-the-box performance.
- **[sktime](https://github.com/sktime/sktime)**  - Unified framework for machine learning with time series. Scikit-learn compatible API for forecasting, classification, clustering, and anomaly detection.
- **[StatsForecast](https://github.com/Nixtla/statsforecast)**  - Lightning-fast statistical forecasting with ARIMA, ETS, CES, and Theta models. Optimized for high-performance time series workloads.
- **[MLForecast](https://github.com/Nixtla/mlforecast)**  - Scalable machine learning for time series forecasting. Train any sklearn-compatible model on millions of time series with efficient feature engineering. Apache 2.0 licensed.
- **[cuML](https://github.com/rapidsai/cuml)**  - GPU-accelerated machine learning algorithms with scikit-learn compatible API. 10-50x faster than CPU implementations for large datasets. Apache 2.0 licensed.
- **[SynapseML](https://github.com/microsoft/SynapseML)**  - Distributed machine learning on Apache Spark. Scalable, composable APIs for text analytics, vision, anomaly detection with seamless Python/Scala/R/.NET integration. MIT licensed.
- **[Darts](https://github.com/unit8co/darts)**  - User-friendly forecasting and anomaly detection for time series. Unifies classical statistical models (ARIMA, ETS) with modern neural networks (N-BEATS, TFT, DeepAR) in a single scikit-learn compatible API. Apache 2.0 licensed.
- **[PyTorch Forecasting](https://github.com/sktime/pytorch-forecasting)**  - Time series forecasting with PyTorch. Multiple neural architectures (N-BEATS, TFT, DeepAR) with in-built interpretation capabilities, built on PyTorch Lightning for distributed training. MIT licensed.
#### AutoML & Hyperparameter Optimization
- **[Optuna](https://github.com/optuna/optuna)**  - Modern, define-by-run hyperparameter optimization with pruning and visualizations. Extremely popular in 2026.
- **[AutoGluon](https://github.com/autogluon/autogluon)**  - AWS AutoML toolkit for tabular, image, text, and multimodal data - state-of-the-art with almost zero code.
- **[FLAML](https://github.com/microsoft/FLAML)**  - Microsoft's fast & lightweight AutoML focused on efficiency and low compute.
- **[Katib (Kubeflow)](https://github.com/kubeflow/katib)**  - Kubernetes-native AutoML for hyperparameter tuning, early stopping, and neural architecture search. Framework-agnostic with support for TensorFlow, PyTorch, XGBoost, and custom training operators. Apache 2.0 licensed.
- **[AutoKeras](https://github.com/keras-team/autokeras)**  - Neural architecture search on top of Keras.
#### Interactive ML Apps & Notebooks
- **[Streamlit](https://github.com/streamlit/streamlit)**  - The fastest way to build and share data apps. Transform Python scripts into beautiful web applications with minimal code. Widely used for ML model demos, data visualization, and internal tools.
- **[Gradio](https://github.com/gradio-app/gradio)**  - Build and share delightful machine learning apps, all in Python. The de facto standard for creating interactive ML demos with automatic UI generation from function signatures. Powers thousands of Hugging Face Spaces.
- **[Marimo](https://github.com/marimo-team/marimo)**  - A reactive notebook for Python — run reproducible experiments, query with SQL, execute as a script, deploy as an app, and version with git. Stored as pure Python. All in a modern, AI-native editor.
#### Model Training & Optimization Utilities
- **[Hugging Face Accelerate](https://github.com/huggingface/accelerate)**  - Simple API to make training scripts run on any hardware (multi-GPU, TPU, mixed precision) with minimal code changes.
- **[DeepSpeed](https://github.com/microsoft/DeepSpeed)**  - Microsoft's deep learning optimization library for extreme-scale training (ZeRO, offloading, MoE).
- **[Transformers](https://github.com/huggingface/transformers)**  - Library of pretrained transformer models and utilities for text, vision, audio, and multimodal training and inference.
- **[FlashAttention](https://github.com/Dao-AILab/flash-attention)**  - Fast exact attention kernels that reduce memory usage and accelerate transformer training and inference.
- **[xFormers](https://github.com/facebookresearch/xformers)**  - Optimized transformer building blocks and attention operators for PyTorch.
- **[PyTorch Lightning](https://github.com/Lightning-AI/lightning)**  - High-level wrapper for PyTorch that removes boilerplate and adds best practices.
- **[fastai](https://github.com/fastai/fastai)**  - Deep learning library providing practitioners with high-level components for state-of-the-art results. Built on PyTorch with a focus on usability and transfer learning. Apache 2.0 licensed.
- **[PyTorch Ignite](https://github.com/pytorch/ignite)**  - High-level library for training and evaluating neural networks in PyTorch with an engine, events & handlers system for maximum flexibility. BSD-3-Clause licensed.
- **[ONNX Runtime](https://github.com/microsoft/onnxruntime)**  - High-performance inference and training for ONNX models across hardware.
- **[einops](https://github.com/arogozhnikov/einops)**  - Flexible, powerful tensor operations for readable and reliable code. Supports PyTorch, JAX, TensorFlow, NumPy, MLX.
- **[safetensors](https://github.com/huggingface/safetensors)**  - Simple, safe way to store and distribute tensors. Fast, secure alternative to pickle for model serialization.
- **[torchmetrics](https://github.com/Lightning-AI/torchmetrics)**  - Machine learning metrics for distributed, scalable PyTorch applications. 80+ metrics with built-in distributed synchronization.
- **[torchao](https://github.com/pytorch/ao)**  - PyTorch native quantization and sparsity for training and inference. Drop-in optimizations for production deployment.
- **[SHAP](https://github.com/shap/shap)**  - Game theoretic approach to explain the output of any machine learning model. Industry standard for model interpretability.
- **[skorch](https://github.com/skorch-dev/skorch)**  - Scikit-learn compatible neural network library that wraps PyTorch. Seamlessly integrate PyTorch models with scikit-learn pipelines, grid search, and cross-validation.
- **[Composer](https://github.com/mosaicml/composer)**  - Supercharge your model training. MosaicML's PyTorch training library with built-in algorithms for efficient training (FSDP, gradient compression, progressive resizing) and seamless distributed training on large-scale clusters. Apache 2.0 licensed.
---
### 🧠 2. Open Foundation Models
> Pretrained language, multimodal, speech, and video models with publicly available weights.
#### Large Language Models (Base + Chat)
- **[RWKV-7 "Goose" (BlinkDL)](https://github.com/BlinkDL/RWKV-LM)**  - Novel RNN architecture with transformer-level LLM performance. 100% attention-free, linear-time, constant-space (no kv-cache), infinite ctx_len. Linux Foundation AI project with runtime already deployed in Windows & Office.
- **[Qwen3.6-Plus (Alibaba)](https://github.com/QwenLM/Qwen)**  - Latest flagship series released April 2026 with 1M context window, agentic coding performance competitive with Claude 4.5 Opus, and enhanced multimodal capabilities.
- **[Gemma 4 (Google)](https://github.com/google-deepmind/gemma)**  - Released April 2026 in four sizes (E2B, E4B, 26B MoE, 31B Dense). First major update in a year with Apache 2.0 license, complex logic, and agentic workflows.
- **[Kimi K2 (Moonshot AI)](https://github.com/MoonshotAI/Kimi-K2)**  - State-of-the-art 1T parameter MoE model with 32B activated parameters and 128K context. Trained with Muon optimizer for exceptional reasoning and coding performance.
- **[Kimi K2.5 (Moonshot AI)](https://github.com/MoonshotAI/Kimi-K2.5)**  - Frontier open-weight MoE model with 256K context, strong coding and reasoning performance, and native multimodal + tool-use support for agentic workflows.
- **[Phi-4 (Microsoft)](https://github.com/microsoft/PhiCookBook)**  - Small but highly capable models optimized for reasoning, edge devices, and on-device inference. Includes Phi-4-reasoning variants with thinking capabilities.
- **[GLM-5 (Zhipu AI)](https://github.com/zai-org/GLM-5)**  - Strong open model line with solid coding, reasoning, and agentic-task performance.
- **[OLMo 2 (Allen AI)](https://github.com/allenai/OLMo)**  - Fully open-source LLMs (1B–32B) with complete transparency: models, data, training code, and logs. Designed by scientists, for scientists.
- **[Llama 4 (Meta)](https://github.com/meta-llama/llama-models)**  - First native multimodal MoE open-source models (Scout: 10M context, Maverick: 400B+ params). Released April 2025 with enterprise-grade capabilities.
- **[GPT-OSS (OpenAI)](https://github.com/openai/gpt-oss)**  - OpenAI's first open-weight models since GPT-2 (120B and 20B MoE). Apache 2.0 licensed with state-of-the-art performance for their size class. Released August 2025.
- **[InternLM3 (Shanghai AI Lab)](https://github.com/InternLM/InternLM)**  - 8B parameter instruction model with state-of-the-art performance on reasoning and knowledge-intensive tasks. Trained on only 4 trillion tokens (75% cost savings). Supports deep thinking mode via long chain-of-thought. Apache 2.0 licensed.
#### Coding & Reasoning Models
- **[DeepSeek-Coder-V2 / R1-Coder](https://github.com/deepseek-ai/DeepSeek-Coder)**  - Best-in-class open coding model (236B MoE). Outperforms closed models on many code benchmarks.
- **[Qwen3-Coder-Next (Alibaba)](https://github.com/QwenLM/Qwen3-Coder)**  - Leading open coding model. Strong Pareto frontier for cost-effective agent deployment.
#### Multimodal Models (Vision + Language)
- **[MMaDA (Gen-Verse)](https://github.com/Gen-Verse/MMaDA)**  - Open-sourced multimodal large diffusion language model with unified architecture for text, image generation and multimodal reasoning. MIT licensed, NeurIPS 2025.
- **[Qwen3-VL (Alibaba)](https://github.com/QwenLM/Qwen3-VL)**  - Latest flagship VLM with native 256K context (expandable to 1M), visual agent capabilities, 3D grounding, and superior multimodal reasoning. Major leap over Qwen2.5-VL.
- **[GLM-4.5V / GLM-4.1V-Thinking (Zhipu AI)](https://github.com/zai-org/GLM-V)**  - Strong multimodal reasoning with scalable reinforcement learning. Compares favorably with Gemini-2.5-Flash on benchmarks.
- **[MiniCPM-V 2.6](https://github.com/OpenBMB/MiniCPM-V)**  - Handles images up to 1.8M pixels with top-tier OCR performance. Excellent for on-device deployment.
- **[Gemma 4 (Google)](https://github.com/google-deepmind/gemma)**  - Multimodal model supporting vision-language input, optimized for efficiency, complex logic, and on-device use.
- **[Magma (Microsoft)](https://github.com/microsoft/Magma)**  - Foundation model for multimodal AI agents that perceives the world and takes goal-driven actions across digital and physical environments. CVPR 2025.
#### Speech & Audio Models (TTS, STT, Music)
- **[FunASR](https://github.com/modelscope/FunASR)**  - Fundamental end-to-end speech recognition toolkit with SOTA pretrained models. Supports ASR, VAD, speaker verification, diarization, and multi-talker ASR. Industrial-grade with 31-language support and real-time transcription services. MIT licensed.
- **[Whisper (OpenAI → community forks)](https://github.com/openai/whisper)**  - The gold-standard open speech-to-text model. Massive community fine-tunes available.
- **[faster-whisper (SYSTRAN)](https://github.com/SYSTRAN/faster-whisper)**  - Reimplementation of Whisper using CTranslate2 for up to 4x faster inference with same accuracy. Supports batched processing and 8-bit quantization.
- **[OuteTTS / CosyVoice 2](https://github.com/edwko/OuteTTS)**  - High-quality open TTS with natural prosody and multilingual support.
- **[Fish Speech / StyleTTS 2](https://github.com/fishaudio/fish-speech)**  - Zero-shot TTS with excellent voice cloning. Extremely popular in 2026.
- **[MusicGen / AudioCraft (Meta)](https://github.com/facebookresearch/audiocraft)**  - Open music and audio generation models.
- **[VibeVoice (Microsoft)](https://github.com/microsoft/VibeVoice)**  - Open-source frontier voice AI with expressive, longform conversational speech synthesis. 7B parameter TTS with streaming support.
- **[Qwen3-TTS (Alibaba)](https://github.com/QwenLM/Qwen3-TTS)**  - Open TTS series supporting stable, expressive, and streaming speech generation with free-form voice design and vivid voice cloning. Natural language instruction-driven control over timbre, emotion, and prosody. Apache 2.0 licensed.
- **[Chatterbox (Resemble AI)](https://github.com/resemble-ai/chatterbox)**  - State-of-the-art open TTS family with 350M parameter Turbo variant. Single-step generation with native paralinguistic tags for realistic dialogue.
- **[Dia (Nari Labs)](https://github.com/nari-labs/dia)**  - 1.6B parameter TTS generating ultra-realistic dialogue in one pass with nonverbal communications (laughter, coughing). Emotion and tone control via audio conditioning.
- **[Step-Audio (StepFun)](https://github.com/stepfun-ai/Step-Audio)**  - 130B-parameter production-ready audio LLM for intelligent speech interaction. Supports multilingual conversations (Chinese, English, Japanese), emotional tones, regional dialects (Cantonese, Sichuanese), adjustable speech rates, and prosodic styles including rap. Apache 2.0 licensed.
- **[Voxtral TTS (Mistral)](https://github.com/mistralai/mistral-inference)**  - 4B parameter state-of-the-art TTS with zero-shot voice cloning, 9-language support, and ~90ms time-to-first-audio for voice agents.
- **[WhisperSpeech](https://github.com/WhisperSpeech/WhisperSpeech)**  - Open source text-to-speech system built by inverting Whisper. High-quality voice cloning with zero-shot capabilities. MIT licensed.
#### Video & Animation Models
- **[CogVideoX (Zhipu AI / community)](https://github.com/THUDM/CogVideo)**  - High-quality open text-to-video model (5B-12B).
- **[Mochi 1 (Genmo)](https://github.com/genmoai/mochi)**  - 10B open video model with impressive motion and consistency.
---
### ⚡ 3. Inference Engines & Serving
> Inference runtimes, serving systems, and optimization tools for running models locally or in production.
#### Local / On-device Inference
- **[llama.cpp](https://github.com/ggml-org/llama.cpp)**  - Pure C/C++ inference engine with GGUF format support. The gold standard for CPU/GPU/Apple Silicon on-device running. Includes llama-server for OpenAI-compatible API. Now at 100K+ stars.
- **[Ollama](https://github.com/ollama/ollama)**  - Dead-simple local LLM runner with a one-line install, model registry, and OpenAI-compatible API.
- **[MLX](https://github.com/ml-explore/mlx)**  (Apple) - High-performance array framework + LLM inference optimized for Apple Silicon.
- **[MLC-LLM](https://github.com/mlc-ai/mlc-llm)**  - Deployment engine that compiles and runs LLMs across browsers, mobile devices, and local hardware.
- **[WebLLM](https://github.com/mlc-ai/web-llm)**  - High-performance in-browser LLM inference engine. Runs models directly in the browser with WebGPU acceleration.
- **[llama-cpp-python](https://github.com/abetlen/llama-cpp-python)**  - Official Python bindings for llama.cpp.
- **[KoboldCpp](https://github.com/LostRuins/koboldcpp)**  - User-friendly llama.cpp fork focused on role-playing and creative writing.
- **[RamaLama](https://github.com/containers/ramalama)**  - Container-centric tool for simplifying local AI model serving. Automatically detects GPUs, pulls optimized container images, and runs models securely in rootless containers with enterprise-grade isolation.
- **[LiteRT-LM](https://github.com/google-ai-edge/LiteRT-LM)**  - Google's production-ready inference framework for deploying LLMs on edge devices. Cross-platform support for Android, iOS, Web, Desktop, and IoT with GPU/NPU acceleration. Powers on-device GenAI in Chrome and Chromebook Plus. Apache 2.0 licensed.
#### High-performance Serving & API Servers
- **[llm-d](https://github.com/llm-d/llm-d)**  - Kubernetes-native distributed LLM inference framework. Donated to CNCF by RedHat, Google, and IBM. Intelligent scheduling, KV-cache optimization, and state-of-the-art performance across accelerators.
- **[LMDeploy](https://github.com/InternLM/lmdeploy)**  - Toolkit for compressing, deploying, and serving LLMs from OpenMMLab. 4-bit inference with 2.4x higher performance than FP16, distributed multi-model serving across machines.
- **[vLLM](https://github.com/vllm-project/vllm)**  - State-of-the-art serving engine with PagedAttention and continuous batching. Currently the fastest production-grade LLM server.
- **[LMCache](https://github.com/LMCache/LMCache)**  - Supercharge LLM inference with the fastest KV Cache layer. 3-10x delay savings and GPU cycle reduction for multi-round QA and RAG. Integrates seamlessly with vLLM for distributed, high-throughput deployments. Apache 2.0 licensed.
- **[vLLM Production Stack](https://github.com/vllm-project/production-stack)**  - Kubernetes-native production stack for vLLM inference. Automated deployment, autoscaling, and monitoring for enterprise-grade LLM serving. Built by the vLLM team for seamless integration.
- **[nano-vLLM](https://github.com/GeeeekExplorer/nano-vllm)**  - Minimalist vLLM implementation in ~1,200 lines of Python. Educational yet performant with prefix caching, tensor parallelism, and CUDA graph acceleration. Comparable inference speeds to full vLLM. MIT licensed.
- **[SGLang](https://github.com/sgl-project/sglang)**  - Next-gen serving framework with RadixAttention. Powers xAI's production workloads at 100K+ GPUs scale.
- **[TensorRT-LLM](https://github.com/NVIDIA/TensorRT-LLM)**  - NVIDIA's official high-performance inference backend.
- **[Aphrodite Engine](https://github.com/PygmalionAI/aphrodite-engine)**  - vLLM fork optimized for role-play and creative writing.
- **[AIBrix](https://github.com/vllm-project/aibrix)**  - Cost-efficient and pluggable infrastructure components for GenAI inference. Kubernetes-native control plane for vLLM with distributed KV cache, heterogeneous GPU serving, and intelligent routing. Apache 2.0 licensed.
- **[Triton Inference Server](https://github.com/triton-inference-server/server)**  - NVIDIA's production-grade open-source inference serving software. Supports multiple frameworks (TensorRT, PyTorch, ONNX) with optimized cloud and edge deployment.
- **[mistral.rs](https://github.com/EricLBuehler/mistral.rs)**  - Fast, flexible Rust-native LLM inference engine built on Candle. Supports text, vision, audio, image generation, and embeddings with hardware-aware auto-tuning.
- **[KTransformers](https://github.com/kvcache-ai/ktransformers)**  - Flexible framework for heterogeneous CPU-GPU LLM inference and fine-tuning. Enables running large MoE models by offloading experts to CPU with BF16/FP8 precision support.
- **[llamafile](https://github.com/mozilla-ai/llamafile)**  - Mozilla's single-file distributable LLM solution. Bundle model weights, inference engine, and runtime into one portable executable that runs on six OSes without installation.
- **[Xinference](https://github.com/xorbitsai/inference)**  - Unified, production-ready inference API for LLMs, speech, and multimodal models. Drop-in GPT replacement with single-line code changes. Supports thousands of models with auto-batching and distributed inference.
- **[LightLLM](https://github.com/ModelTC/LightLLM)**  - Pure Python-based LLM inference and serving framework with lightweight design, easy extensibility, and high-speed performance. Integrates optimizations from FasterTransformer, TGI, vLLM, and SGLang.
- **[TabbyAPI](https://github.com/theroyallab/tabbyAPI)**  - FastAPI-based API server for ExLlamaV2/V3 backends. OpenAI-compatible API with support for model loading/unloading, embeddings, speculative decoding, multi-LoRA, and streaming.
- **[GPUStack](https://github.com/gpustack/gpustack)**  - GPU cluster manager that orchestrates inference engines like vLLM and SGLang. Automated engine selection, parameter optimization, and distributed multi-GPU deployment for high-performance AI workloads.
- **[One-API](https://github.com/songquanpeng/one-api)**  - LLM API management and key redistribution system. Unifies multiple providers (OpenAI, Anthropic, Azure, etc.) under a single OpenAI-compatible API with built-in rate limiting, quota management, and cost tracking. MIT licensed.
- **[OpenLLM (BentoML)](https://github.com/bentoml/OpenLLM)**  - Production-grade platform for running any open-source LLMs as OpenAI-compatible API endpoints. Supports 50+ models with built-in streaming, batching, and auto-acceleration. Apache 2.0 licensed.
- **[Higress (Alibaba)](https://github.com/alibaba/higress)**  - AI-native API gateway born from Alibaba's internal infrastructure with 2+ years of production validation. Provides unified LLM API and MCP (Model Context Protocol) management with enterprise-grade 99.99% availability. Apache 2.0 licensed.
#### Additional Inference Engines
- **[CTranslate2](https://github.com/OpenNMT/CTranslate2)**  - Fast inference engine for Transformer models supporting OpenNMT and Hugging Face models. Optimized for CPU and GPU with batching, quantization (INT8/FP16), and dynamic memory management. Powers faster-whisper and other production deployments. MIT licensed.
#### Quantization, Distillation & Optimization
- **[GGUF](https://github.com/ggml-org/llama.cpp)**  (part of llama.cpp) - Modern quantized format that powers most local inference.
- **[bitsandbytes](https://github.com/bitsandbytes-foundation/bitsandbytes)**  - 8-bit and 4-bit optimizers + quantization.
- **[ExLlamaV2](https://github.com/turboderp/exllamav2)**  - Highly optimized CUDA kernels for 4-bit/8-bit inference.
- **[Optimum](https://github.com/huggingface/optimum)**  - Hardware-specific acceleration and quantization.
---
### 🤖 4. Agentic AI & Multi-Agent Systems
> Frameworks and platforms for building agent-based systems and multi-agent workflows.
#### Single-Agent Frameworks
- **[LangGraph](https://github.com/langchain-ai/langgraph)**  - Stateful, controllable agent orchestration.
- **[CrewAI](https://github.com/crewAIInc/crewAI)**  - Role-based agent framework.
- **[AutoGen (AG2)](https://github.com/microsoft/autogen)**  - Flexible multi-agent conversation framework.
- **[DSPy](https://github.com/stanfordnlp/dspy)**  - Framework for programming language model pipelines with modules, optimizers, and evaluation loops.
- **[Semantic Kernel](https://github.com/microsoft/semantic-kernel)**  - SDK for building and orchestrating AI agents and workflows across multiple programming languages.
- **[smolagents](https://github.com/huggingface/smolagents)**  - Lightweight agent framework centered on tool use and code-executing workflows.
- **[LangChain](https://github.com/langchain-ai/langchain)**  - Foundational library for agents, chains, and memory.
- **[Hermes Agent (NousResearch)](https://github.com/NousResearch/hermes-agent)**  - The agent that grows with you. Autonomous server-side agent with persistent memory that learns and improves over time.
- **[Agno](https://github.com/agno-agi/agno)**  - Build, run, and manage agentic software at scale. High-performance framework for multi-agent systems with memory, knowledge, and tools.
- **[Upsonic](https://github.com/Upsonic/Upsonic)**  - Agent framework for fintech and banking with built-in MCP support, guardrails, and tool server architecture.
- **[VoltAgent](https://github.com/VoltAgent/voltagent)**  - TypeScript-first AI agent engineering platform with memory, RAG, workflows, MCP integration, and voice support.
- **[PocketFlow](https://github.com/The-Pocket/PocketFlow)**  - 100-line minimalist LLM framework for building agent workflows. Lightweight, extensible architecture for tool use and autonomous task execution.
- **[Agent Development Kit (Google)](https://github.com/google/adk-python)**  - Code-first Python toolkit for building sophisticated AI agents with multi-agent orchestration, built-in evaluation, and flexible deployment. Model-agnostic with tight Google ecosystem integration. Apache 2.0 licensed.
- **[PydanticAI](https://github.com/pydantic/pydantic-ai)**  - Type-safe AI agent framework from the creators of Pydantic. Model-agnostic with 20+ providers, built-in observability via Logfire, MCP/A2A protocol support, and YAML/JSON agent definitions. MIT licensed.
- **[Qwen-Agent](https://github.com/QwenLM/Qwen-Agent)**  - Agent framework built on Qwen models featuring function calling, MCP support, code interpreter, RAG, and Chrome extension. Powers Qwen Chat with advanced tool use and planning capabilities. Apache 2.0 licensed.
#### Multi-Agent Orchestration
- **[MetaGPT](https://github.com/FoundationAgents/MetaGPT)**  - Simulates an entire "AI software company".
- **[CAMEL](https://github.com/camel-ai/camel)**  - First and best multi-agent framework for building scalable agent systems. Apache 2.0 licensed with extensive tooling for agent communication and task automation.
- **[Swarms](https://github.com/kyegomez/swarms)**  - Bleeding-edge enterprise multi-agent orchestration.
- **[Mastra](https://github.com/mastra-ai/mastra)**  - TypeScript-first agent framework with built-in RAG, workflows, tool integrations, observability and observational memory.
- **[Deer-Flow (ByteDance)](https://github.com/bytedance/deer-flow)**  - Open-source long-horizon SuperAgent harness that researches, codes, and creates. Handles tasks from minutes to hours with sandboxes, memories, tools, skills, subagents, and message gateway.
- **[OpenAI Agents SDK](https://github.com/openai/openai-agents-python)**  - Production-ready lightweight framework for multi-agent workflows. The evolution of Swarm with enhanced orchestration capabilities and enterprise-grade features.
- **[AgentScope](https://github.com/agentscope-ai/agentscope)**  - Alibaba's production-ready multi-agent framework with 23K+ stars. Features built-in MCP and A2A support, message hub for flexible orchestration, and AgentScope Runtime for production deployment.
- **[Microsoft Agent Framework](https://github.com/microsoft/agent-framework)**  - Microsoft's official framework combining AutoGen's agent abstractions with Semantic Kernel's enterprise features. Supports Python and .NET with graph-based workflows.
- **[Agency Swarm](https://github.com/VRSEN/agency-swarm)**  - Reliable multi-agent orchestration framework built on top of the OpenAI Assistants API with organizational structure modeling.
- **[elizaOS](https://github.com/elizaOS/eliza)**  - Autonomous multi-agent framework for building and deploying AI-powered applications. Features Discord/Telegram/Farcaster connectors, RAG support, and a modern web dashboard.
- **[Agent Squad (AWS Labs)](https://github.com/awslabs/agent-squad)**  - Flexible multi-agent orchestration framework with intelligent intent classification and context management. Supports Python and TypeScript with pre-built agents for Bedrock, Lex, and custom integrations. Apache 2.0 licensed.
- **[DeepResearchAgent](https://github.com/SkyworkAI/DeepResearchAgent)**  - Hierarchical multi-agent system for deep research tasks with automated task decomposition and execution across complex domains.
- **[BeeAI Framework (IBM)](https://github.com/i-am-bee/bee-agent-framework)**  - Production-ready multi-agent framework in Python and TypeScript. Features workflow orchestration, ACP/MCP protocol support, and deep watsonx integration. Part of Linux Foundation AI & Data program.
- **[AI Town](https://github.com/a16z-infra/ai-town)**  - Deployable starter kit for building virtual towns where AI characters live, chat and socialize. Inspired by Stanford's Generative Agents research with persistent agent memory and social interactions. MIT licensed.
#### Autonomous Coding Agents
- **[OpenHands (ex-OpenDevin)](https://github.com/All-Hands-AI/OpenHands)**  - Full-featured open-source AI software engineer.
- **[Goose](https://github.com/block/goose)**  - Extensible on-machine AI agent for development tasks.
- **[OpenCode](https://github.com/anomalyco/opencode)**  - Terminal-native autonomous coding agent.
- **[Aider](https://github.com/paul-gauthier/aider)**  - Command-line pair-programming agent.
- **[Pi (badlogic)](https://github.com/badlogic/pi-mono)**  - Terminal coding agent with hash-anchored edits, LSP integration, subagents, MCP support, and package ecosystem.
- **[Mistral-Vibe (Mistral)](https://github.com/mistralai/mistral-vibe)**  - Minimal CLI coding agent by Mistral. Lightweight, fast, and designed for local development workflows.
- **[Nanocoder (Nano-Collective)](https://github.com/Nano-Collective/nanocoder)**  - Beautiful local-first coding agent running in your terminal. Built for privacy and control with support for multiple AI providers via OpenRouter.
- **[Gemini CLI (Google)](https://github.com/google-gemini/gemini-cli)**  - Open-source AI agent that brings Gemini's power directly into your terminal. Supports code generation, shell execution, and file editing with full Apache 2.0 licensing.
- **[Archon](https://github.com/coleam00/Archon)**  - Workflow engine for deterministic AI coding agents. Define development processes as YAML workflows (planning → implementation → validation → review → PR) with isolated git worktrees for parallel execution. MIT licensed.
#### Domain-Specific Agents
- **[Composio](https://github.com/ComposioHQ/composio)**  - Tool integration layer for AI agents with 1000+ toolkits, authentication management, and sandboxed workbench. Powers tool use across major frameworks.
- **[Langflow](https://github.com/langflow-ai/langflow)**  - Visual low-code platform for agentic workflows.
- **[Dify](https://github.com/langgenius/dify)**  - Production-ready agentic workflow platform.
- **[OWL (camel-ai/owl)](https://github.com/camel-ai/owl)**  - Advanced multi-agent collaboration system.
- **[AI-Scientist-v2 (SakanaAI)](https://github.com/SakanaAI/AI-Scientist-v2)**  - Workshop-level automated scientific discovery via agentic tree search. Generates novel research ideas, runs experiments, and writes papers.
- **[PraisonAI](https://github.com/MervinPraison/PraisonAI)**  - 24/7 AI employee team for automating complex challenges. Low-code multi-agent framework with handoffs, guardrails, memory, RAG, and 100+ LLM providers.
- **[Agent-S (Simular AI)](https://github.com/simular-ai/Agent-S)**  - Open agentic framework that uses computers like a human. SOTA on OSWorld benchmark (72.6%) for GUI automation and computer control.
- **[UI-TARS Desktop (ByteDance)](https://github.com/bytedance/UI-TARS-desktop)**  - Open-source multimodal AI agent stack with native GUI agent capabilities. Desktop application bringing GUI agent and vision power to your computer, browser, and terminal. Apache 2.0 licensed.
- **[Browser Use](https://github.com/browser-use/browser-use)**  - Makes websites accessible for AI agents. Enables autonomous web automation, data extraction, and task completion with natural language instructions. MIT licensed.
- **[Steel Browser](https://github.com/steel-dev/steel-browser)**  - Open-source browser API for AI agents and apps. Batteries-included browser sandbox for web automation without infrastructure worries. Apache 2.0 licensed.
- **[TradingAgents](https://github.com/TauricResearch/TradingAgents)**  - Multi-agent framework for financial trading. Simulates professional trading firm operations with 6+ specialized agent roles, backtesting, risk management, and portfolio optimization. Built with LangGraph, supports multiple LLM providers.
- **[Parlant](https://github.com/emcie-co/parlant)**  - Conversational control layer for customer-facing AI agents. Enterprise-grade context engineering framework optimized for consistent, compliant, and on-brand B2C and sensitive B2B interactions. Apache 2.0 licensed.
#### Agent Memory & State
- **[Letta (ex-MemGPT)](https://github.com/letta-ai/letta)**  - Platform for building stateful agents with advanced memory that learn and self-improve over time.
- **[Mem0](https://github.com/mem0ai/mem0)**  - Universal memory layer for AI agents. Persistent, multi-session memory across models and environments.
- **[Hindsight](https://github.com/vectorize-io/hindsight)**  - State-of-the-art long-term memory for AI agents by Vectorize. Fully self-hosted, MIT-licensed, with integrations for LangChain, CrewAI, LlamaIndex, Vercel AI SDK, and more.
---
### 🔍 5. Retrieval-Augmented Generation (RAG) & Knowledge
> Retrieval systems, vector databases, embedding models, and related tooling for RAG pipelines.
#### Vector Databases & Search Engines
- **[Chroma](https://github.com/chroma-core/chroma)**  - Most popular open-source embedding database.
- **[Qdrant](https://github.com/qdrant/qdrant)**  - High-performance vector search engine in Rust.
- **[Weaviate](https://github.com/weaviate/weaviate)**  - GraphQL-native vector search engine.
- **[Milvus](https://github.com/milvus-io/milvus)**  - Scalable cloud-native vector database.
- **[Faiss](https://github.com/facebookresearch/faiss)**  - Similarity search and clustering library for dense vectors with CPU and GPU implementations.
- **[LanceDB](https://github.com/lancedb/lancedb)**  - Serverless vector DB optimized for multimodal data.
- **[Vespa](https://github.com/vespa-engine/vespa)**  - AI + Data platform with hybrid search (vector + keyword) and real-time indexing at scale. Battle-tested serving billions of queries daily.
- **[pgvector](https://github.com/pgvector/pgvector)**  - PostgreSQL extension for vector similarity search.
- **[Quickwit](https://github.com/quickwit-oss/quickwit)**  - Cloud-native search engine for observability. Open-source alternative to Datadog, Elasticsearch, Loki, and Tempo with native vector search support.
- **[Tantivy](https://github.com/quickwit-oss/tantivy)**  - Full-text search engine library inspired by Apache Lucene and written in Rust. Powers Quickwit and other production search systems.
- **[Manticore Search](https://github.com/manticoresoftware/manticoresearch)**  - Easy to use open source fast database for search. Good alternative to Elasticsearch with SQL-like interface and vector search capabilities.
- **[OpenSearch](https://github.com/opensearch-project/OpenSearch)**  - Open-source distributed and RESTful search and analytics suite with native vector search. Enterprise-grade fork of Elasticsearch with k-NN plugin for semantic search at scale.
- **[Marqo](https://github.com/marqo-ai/marqo)**  - Multimodal vector search for text, image, and structured data. End-to-end indexing and search with built-in embedding models. Apache 2.0 licensed.
- **[Vald](https://github.com/vdaas/vald)**  - Highly scalable distributed vector search engine. Cloud-native architecture with automatic indexing, horizontal scaling, and multiple ANN algorithm support. Apache 2.0 licensed.
- **[Annoy](https://github.com/spotify/annoy)**  - Approximate nearest neighbors library optimized for memory usage and fast loading. Powers Spotify's music recommendation with C++/Python bindings. Apache 2.0 licensed.
#### Embedding Models
- **[BGE (FlagEmbedding)](https://github.com/FlagOpen/FlagEmbedding)**  - BAAI's best-in-class embedding family.
- **[E5 (Microsoft)](https://github.com/microsoft/unilm)**  - High-performance text embeddings for retrieval.
- **[FastEmbed (Qdrant)](https://github.com/qdrant/fastembed)**  - Lightweight, fast Python library for embedding generation with ONNX Runtime. Supports text, sparse (SPLADE), and late-interaction (ColBERT) embeddings without GPU dependencies. Apache 2.0 licensed.
- **[EmbedAnything](https://github.com/StarlightSearch/EmbedAnything)**  - Minimalist, highly performant multimodal embedding pipeline built in Rust. Memory-safe, modular, and production-ready for text, image, and audio embeddings with seamless vector DB integration. Apache 2.0 licensed.
#### Embedding Benchmarks
- **[MTEB](https://github.com/embeddings-benchmark/mteb)**  - Massive Text Embedding Benchmark covering 1000+ languages and diverse tasks. The industry standard for evaluating and comparing embedding models.
#### RAG Frameworks & Advanced Retrieval Tools
- **[LlamaIndex](https://github.com/run-llama/llama_index)**  - Full-featured RAG pipeline with advanced indexing.
- **[Haystack](https://github.com/deepset-ai/haystack)**  - End-to-end NLP and RAG framework.
- **[RAGFlow](https://github.com/infiniflow/ragflow)**  - Deep-document-understanding RAG engine.
- **[GraphRAG (Microsoft)](https://github.com/microsoft/graphrag)**  - Knowledge-graph-based RAG.
- **[Docling](https://github.com/docling-project/docling)**  - Document processing toolkit for turning PDFs and other files into structured data for GenAI workflows.
- **[Unstructured](https://github.com/Unstructured-IO/unstructured)**  - Best-in-class document preprocessing.
- **[MinerU](https://github.com/opendatalab/MinerU)**  - High-accuracy document parsing for LLM and RAG workflows. Converts PDFs, Word, PPTs, and images into structured Markdown/JSON with VLM+OCR dual engine.
- **[Marker](https://github.com/datalab-to/marker)**  - Fast, accurate PDF-to-markdown converter with table extraction, equation handling, and optional LLM enhancement for RAG pipelines.
- **[ColPali / ColQwen](https://github.com/illuin-tech/colpali)**  - Vision-language models for document retrieval.
- **[LightRAG](https://github.com/HKUDS/LightRAG)**  - Graph-based RAG with dual-level retrieval system. Simple and fast with comprehensive knowledge discovery (EMNLP 2025).
- **[RAG-Anything](https://github.com/HKUDS/RAG-Anything)**  - All-in-One Multimodal RAG system for seamless processing of text, images, tables, and equations. Built on LightRAG.
- **[LangChain4j](https://github.com/langchain4j/langchain4j)**  - Java library for integrating LLMs into Java applications. Implements RAG, tool calling (including MCP support), and agents with seamless integration into enterprise Java frameworks like Spring Boot. Apache 2.0 licensed.
- **[Kernel Memory (Microsoft)](https://github.com/microsoft/kernel-memory)**  - Memory solution for users, teams, and applications. RAG pipelines with document ingestion, vector indexing, and natural language querying with citations. Supports multiple LLM providers and vector stores. MIT licensed.
- **[txtai](https://github.com/neuml/txtai)**  - All-in-one AI framework for semantic search, LLM orchestration and language model workflows. Embeddings database with customizable pipelines.
- **[Infinity](https://github.com/michaelfeil/infinity)**  - High-throughput, low-latency serving engine for text-embeddings, reranking, CLIP, and ColPali. OpenAI-compatible API.
- **[FlashRAG](https://github.com/RUC-NLPIR/FlashRAG)**  - Efficient toolkit for RAG research with 40+ retrieval and reranking models, 20+ benchmark datasets, and optimized evaluation pipelines (WWW 2025 Resource). MIT licensed.
- **[DocsGPT](https://github.com/arc53/DocsGPT)**  - Private AI platform for building intelligent agents and assistants with enterprise search. Features Agent Builder, deep research tools, multi-format document analysis, and multi-model support. MIT licensed.
- **[llmware](https://github.com/llmware-ai/llmware)**  - Unified framework for building enterprise RAG pipelines with small, specialized models. Optimized for AI PC and local deployment with 300+ models in catalog. Apache 2.0 licensed.
- **[AutoFlow](https://github.com/pingcap/autoflow)**  - Graph RAG-based conversational knowledge base tool built on TiDB Vector and LlamaIndex. Features Perplexity-style search with built-in website crawler. Apache 2.0 licensed.
- **[rerankers (Answer.AI)](https://github.com/AnswerDotAI/rerankers)**  - Lightweight unified API for all common reranking and cross-encoder models. Supports RankGPT, ColBERT, FlashRank, and API-based rerankers with a dependency-free core. Apache 2.0 licensed.
- **[KAG (OpenSPG)](https://github.com/OpenSPG/KAG)**  - Knowledge Augmented Generation framework for logical reasoning and factual Q&A in professional domains. Builds on OpenSPG knowledge graph engine to overcome traditional RAG vector similarity limitations. Supports multi-hop reasoning with schema-constrained knowledge construction. Apache 2.0 licensed.
- **[Chonkie](https://github.com/chonkie-inc/chonkie)**  - Lightweight document chunking library for fast, efficient RAG pipelines. Memory-safe with multiple chunking strategies (semantic, token, recursive) and direct vector DB integration. MIT licensed.
- **[PageIndex (VectifyAI)](https://github.com/VectifyAI/PageIndex)**  - Vectorless, reasoning-based RAG framework using document index structure. Achieves high accuracy without vector databases through intelligent context engineering and reasoning-based retrieval. MIT licensed.
#### Knowledge Graphs for RAG
- **[Graphiti](https://github.com/getzep/graphiti)**  - Build real-time temporal knowledge graphs for AI agents. Tracks how facts change over time with provenance to source data. Supports prescribed and learned ontology for evolving real-world data. Apache 2.0 licensed.
#### Web Data Ingestion
- **[Crawl4AI](https://github.com/unclecode/crawl4ai)**  - LLM-friendly web crawler that turns websites into clean Markdown for RAG and agentic workflows.
- **[Lightpanda](https://github.com/lightpanda-io/browser)**  - Machine-first headless browser in Zig; rendering-free and ultra-lightweight for AI agent browsing.
- **[Paperless-AI](https://github.com/clusterzx/paperless-ai)**  - Automated document analyzer for Paperless-ngx with RAG-powered semantic search across your document archive.
- **[Firecrawl](https://github.com/firecrawl/firecrawl)**  - Web Data API for AI - search, scrape, and interact with the web at scale. Clean markdown/JSON output with proxy rotation and JS-blocking handled automatically.
#### Document Conversion & Preprocessing
- **[MarkItDown (Microsoft)](https://github.com/microsoft/markitdown)**  - Python tool for converting files and office documents to Markdown. Supports PDF, PowerPoint, Word, Excel, images, audio, HTML, and more with OCR and transcription capabilities. MIT licensed.
- **[OmniParse](https://github.com/adithya-s-k/omniparse)**  - Ingest and parse any unstructured data into structured, actionable data optimized for GenAI applications. Supports documents, tables, images, videos, audio, and web pages with local deployment on T4 GPU. GPL-3.0 licensed.
- **[DocETL (UC Berkeley)](https://github.com/ucbepic/docetl)**  - Agentic LLM-powered data processing and ETL system for complex document processing. Query rewriting and evaluation for unstructured data analysis with 80% higher accuracy than baselines. MIT licensed.
---
### 🎨 6. Generative Media Tools
> Open-source models and applications for image, video, audio, and 3D generation and editing.
#### Image Generation & Editing
- **[ComfyUI](https://github.com/Comfy-Org/ComfyUI)**  - Node-based visual workflow editor for Stable Diffusion, FLUX, etc.
- **[Stable Diffusion WebUI Forge - Neo](https://github.com/Haoming02/sd-webui-forge-classic)**  - Actively maintained Forge-based Stable Diffusion web UI with the familiar extension-driven workflow.
- **[Fooocus](https://github.com/lllyasviel/Fooocus)**  - Midjourney-style UI with beautiful out-of-the-box results.
- **[Diffusers](https://github.com/huggingface/diffusers)**  - PyTorch library for diffusion pipelines spanning image, video, and audio generation.
- **[InvokeAI](https://github.com/invoke-ai/InvokeAI)**  - Full-featured creative studio.
- **[PowerPaint (OpenMMLab)](https://github.com/open-mmlab/PowerPaint)**  - Versatile image inpainting model supporting text-guided inpainting, object removal, and outpainting (ECCV 2024).
- **[SD.Next](https://github.com/vladmandic/sdnext)**  - All-in-one WebUI for AI generative image and video creation with multi-platform support, SDNQ quantization, and balanced CPU/GPU memory offload.
- **[Qwen-Image (Alibaba)](https://github.com/QwenLM/Qwen-Image)**  - 20B MMDiT image foundation model with state-of-the-art complex text rendering and precise image editing. Strong performance in Chinese text generation. Apache 2.0 licensed.
- **[Upscayl](https://github.com/upscayl/upscayl)**  - Free and open-source AI image upscaler for Linux, macOS, and Windows. Uses Real-ESRGAN and Vulkan architecture to enhance images by reconstructing high-resolution details. Cross-platform desktop app with batch processing. AGPL-3.0 licensed.
- **[Z-Image (Tongyi)](https://github.com/Tongyi-MAI/Z-Image)**  - Powerful and efficient image generation model family with 6B parameters. Includes Z-Image-Turbo for sub-second inference and Z-Image-Omni-Base for both generation and editing. Strong bilingual text rendering and instruction adherence. Apache 2.0 licensed.
#### Face Swap & Deepfake
- **[Deep-Live-Cam](https://github.com/hacksider/Deep-Live-Cam)**  - Real-time face swap and one-click video deepfake with only a single image. High-quality face swapping for live video streaming and content creation. AGPL-3.0 licensed.
#### Portrait Animation
- **[EchoMimic (Ant Group)](https://github.com/antgroup/echomimic)**  - Lifelike audio-driven portrait animations through editable landmark conditioning. High-quality talking head generation with precise lip synchronization and natural head movements. AAAI 2025. Apache 2.0 licensed.
#### Video Generation
- **[Wan2.2 (Alibaba)](https://github.com/Wan-Video/Wan2.1)**  - Leading open Mixture-of-Experts text-to-video model.
- **[HunyuanVideo (Tencent)](https://github.com/Tencent-Hunyuan/HunyuanVideo)**  - 13B-parameter systematic video generation framework. Leading quality among open models.
- **[SkyReels V2/V3 (Skywork)](https://github.com/SkyworkAI/SkyReels-V2)**  - First open-source infinite-length film generative model using AutoRegressive Diffusion-Forcing.
- **[Mochi 1 (Genmo)](https://github.com/genmoai/mochi)**  - 10B-parameter open video model.
- **[LTX-Video (Lightricks)](https://github.com/Lightricks/LTX-Video)**  - Fast native 4K video generation.
- **[Stable Video Diffusion (Stability AI)](https://github.com/Stability-AI/generative-models)**  - Official image-to-video and text-to-video implementation within Stability AI's generative models repository.
- **[Latte (Vchitect)](https://github.com/Vchitect/Latte)**  - Latent Diffusion Transformer for video generation with state-of-the-art quality (TMLR 2025). Apache 2.0 licensed.
- **[Open-Sora-Plan (PKU-YuanGroup)](https://github.com/PKU-YuanGroup/Open-Sora-Plan)**  - Reproduction of Sora with full open-source pipeline for text-to-video generation. MIT licensed.
- **[Open-Sora (HPC-AI Tech)](https://github.com/hpcaitech/Open-Sora)**  - Fully open-source video generation with 11B model achieving on-par performance with HunyuanVideo. Complete training pipeline for $200K. Apache 2.0 licensed.
- **[Helios (PKU-YuanGroup)](https://github.com/PKU-YuanGroup/Helios)**  - Efficient long-video generation framework with 24GB VRAM support for up to 10,000 frames (5+ minutes) and 1280×768 resolution. Apache 2.0 licensed.
#### Audio / Music / Voice Generation
- **[AudioCraft / MusicGen (Meta)](https://github.com/facebookresearch/audiocraft)**  - Controllable text-to-music and audio models.
- **[ACE-Step 1.5](https://github.com/ace-step/ACE-Step-1.5)**  - Local-first music generation model with broad hardware support across Mac, AMD, Intel, and CUDA devices.
- **[Fish Speech](https://github.com/fishaudio/fish-speech)**  - Zero-shot TTS and voice cloning.
- **[CosyVoice 2](https://github.com/FunAudioLLM/CosyVoice)**  - Natural multilingual TTS with emotional control.
- **[OuteTTS](https://github.com/edwko/OuteTTS)**  - High-quality open TTS.
- **[Amphion](https://github.com/open-mmlab/Amphion)**  - Comprehensive toolkit for Audio, Music, and Speech Generation (9.7K stars).
- **[Stable Audio Tools](https://github.com/Stability-AI/stable-audio-tools)**  - Stability AI's open-source audio and music generative models. Latent diffusion model for generating audio conditioned on metadata and timing, providing faster inference times and creative control for sound effects and music production. MIT licensed.
#### 3D & Creative Tools
- **[Hunyuan3D-2 (Tencent)](https://github.com/Tencent-Hunyuan/Hunyuan3D-2)**  - State-of-the-art open image-to-3D and text-to-3D.
- **[Trellis (Microsoft)](https://github.com/microsoft/TRELLIS)**  - Structured 3D latents for high-quality generation.
- **[gsplat (3D Gaussian Splatting tools)](https://github.com/nerfstudio-project/gsplat)**  - High-performance 3D Gaussian Splatting library.
- **[LichtFeld-Studio](https://github.com/MrNeRF/LichtFeld-Studio)**  - Native application for training, editing, and exporting 3D Gaussian Splatting scenes with MCMC optimization and timelapse generation. GPL-3.0 licensed.
- **[OpenSplat](https://github.com/pierotofy/OpenSplat)**  - Production-grade, portable implementation of 3D Gaussian Splatting with CPU/GPU support for Windows, Mac, and Linux. Creates 3D scenes from camera poses and sparse points. AGPL-3.0 licensed.
---
### 🛠️ 7. Training & Fine-tuning Ecosystem
> Tools for model training, fine-tuning, synthetic data generation, and distributed training.
#### Full Training Frameworks
- **[LLaMA-Factory](https://github.com/hiyouga/LLaMA-Factory)**  - One-stop unified framework for SFT, DPO, ORPO, KTO with web UI.
- **[Axolotl](https://github.com/axolotl-ai-cloud/axolotl)**  - YAML-driven full pipeline for SFT, DPO, GRPO.
- **[ms-swift](https://github.com/modelscope/ms-swift)**  - Unified training framework for 600+ LLMs and 300+ MLLMs with CPT/SFT/DPO/GRPO (AAAI 2025).
- **[Unsloth](https://github.com/unslothai/unsloth)**  - 2× faster, 70% less memory fine-tuning.
- **[LitGPT](https://github.com/Lightning-AI/litgpt)**  - Clean from-scratch implementations of 20+ LLMs.
- **[LLM Foundry](https://github.com/mosaicml/llm-foundry)**  - Databricks' training framework for composable LLM training with StreamingDataset and Composer.
- **[torchtune](https://github.com/pytorch/torchtune)**  - PyTorch-native library for post-training, fine-tuning, and experimentation with LLMs.
- **[kohya_ss](https://github.com/bmaltais/kohya_ss)**  - Gradio-based GUI and CLI for training Stable Diffusion models (LoRA, Dreambooth, fine-tuning, SDXL). Provides accessible interface to Kohya's powerful training scripts.
- **[TRL (Transformers Reinforcement Learning)](https://github.com/huggingface/trl)**  - Official library for RLHF, SFT, DPO, ORPO.
- **[verl](https://github.com/volcengine/verl)**  - Volcano Engine Reinforcement Learning for LLMs with PPO, GRPO, REINFORCE++, DAPO (EuroSys 2025).
- **[NeMo-RL](https://github.com/NVIDIA-NeMo/RL)**  - Scalable toolkit for efficient model reinforcement with DTensor and Megatron backends.
- **[OpenRLHF](https://github.com/OpenRLHF/OpenRLHF)**  - Easy-to-use, scalable RLHF framework based on Ray. Supports PPO, GRPO, REINFORCE++, DAPO with vLLM integration and async training. Apache 2.0 licensed.
- **[LMFlow](https://github.com/OptimalScale/LMFlow)**  - Extensible toolkit for finetuning and inference of large foundation models. Features RAFT alignment algorithm and comprehensive model support. Apache 2.0 licensed.
- **[XTuner](https://github.com/InternLM/xtuner)**  - A next-generation training engine built for ultra-large MoE models with efficient QLoRA and full-parameter fine-tuning. Apache 2.0 licensed.
- **[Ludwig](https://github.com/ludwig-ai/ludwig)**  - Low-code framework for building custom LLMs and deep neural networks. Declarative YAML configuration for training state-of-the-art models with PEFT/LoRA, 4-bit quantization, distributed training via HuggingFace Accelerate, and native Kubernetes support. Linux Foundation AI project. Apache 2.0 licensed.
- **[nanoGPT (Andrej Karpathy)](https://github.com/karpathy/nanoGPT)**  - The simplest, fastest repository for training/finetuning medium-sized GPTs. Clean, minimal, and hackable codebase for understanding transformer training from scratch. MIT licensed.
- **[TorchTitan (PyTorch)](https://github.com/pytorch/torchtitan)**  - PyTorch native platform for training generative AI models at scale. Showcases 4D parallelism (FSDP, tensor, pipeline, context) for LLM pretraining with 65%+ speedups over optimized baselines. BSD-3-Clause licensed.
- **[VeOmni (ByteDance)](https://github.com/ByteDance-Seed/VeOmni)**  - Versatile framework for both single- and multi-modal pre-training and post-training. Model-centric distributed recipe zoo supporting text, vision, audio, and video models with unified training interface. Apache 2.0 licensed.
- **[H2O LLM Studio](https://github.com/h2oai/h2o-llmstudio)**  - No-code GUI framework for fine-tuning LLMs. Streamlined interface for SFT, reward modeling, and model deployment. Apache 2.0 licensed.
- **[TinyZero](https://github.com/Jiayi-Pan/TinyZero)**  - Minimal reproduction of DeepSeek R1-Zero for countdown and multiplication tasks. Clean, accessible implementation for understanding RL-based reasoning training. Apache 2.0 licensed.
- **[PRIME-RL](https://github.com/PrimeIntellect-ai/prime-rl)**  - Agentic RL Training at Scale from Prime Intellect. Framework for large-scale reinforcement learning capable of scaling to 1000+ GPUs with fully asynchronous RL, FSDP2 training, and vLLM inference. Apache 2.0 licensed.
- **[slime](https://github.com/THUDM/slime)**  - LLM post-training framework for RL Scaling from THUDM. Supports SFT and RL training with multi-turn compilation feedback, powering projects like TritonForge for automated GPU kernel generation. Apache 2.0 licensed.
- **[rLLM](https://github.com/rllm-org/rllm)**  - Democratizing Reinforcement Learning for LLMs. Framework for training AI agents with RL featuring near-zero code changes, CLI-first workflow, and 50+ built-in benchmarks. Supports GRPO, REINFORCE, RLOO with verl and tinker backends. Apache 2.0 licensed.
- **[EasyR1](https://github.com/hiyouga/EasyR1)**  - Efficient, scalable, multi-modality RL training framework based on veRL. Extends veRL to support vision-language models with GRPO algorithm for efficient RL training. Apache 2.0 licensed.
- **[simpleRL-reason](https://github.com/hkust-nlp/simpleRL-reason)**  - Simple reinforcement learning recipe to improve models' reasoning abilities. Rule-based reward with GSM8K/Math datasets, extending from OpenRLHF. MIT licensed.
#### LoRA / PEFT Tools
- **[PEFT (Parameter-Efficient Fine-Tuning)](https://github.com/huggingface/peft)**  - Official library with LoRA, QLoRA, DoRA, etc.
- **[Liger Kernel](https://github.com/linkedin/Liger-Kernel)**  - Ultra-fast custom kernels for training speedup.
- **[MergeKit](https://github.com/arcee-ai/mergekit)**  - Advanced model merging tools.
#### Synthetic Data Generation
- **[distilabel](https://github.com/argilla-io/distilabel)**  - End-to-end pipeline for synthetic instruction data.
- **[Data-Juicer](https://github.com/alibaba/data-juicer)**  - High-performance data processing for LLM training.
- **[Argilla](https://github.com/argilla-io/argilla)**  - Open-source data labeling + synthetic data platform.
- **[SDV (Synthetic Data Vault)](https://github.com/sdv-dev/SDV)**  - High-fidelity tabular and relational synthetic data.
- **[DataTrove (Hugging Face)](https://github.com/huggingface/datatrove)**  - Platform-agnostic data processing pipelines for LLM training at scale. Handles filtering, deduplication, and tokenization on local machines or SLURM clusters.
- **[Bespoke Curator](https://github.com/bespokelabsai/curator)**  - Synthetic data curation for post-training and structured data extraction. Makes it easy to build pipelines around LLMs with batching and progress tracking. Apache 2.0 licensed.
- **[SDG (Harbin Institute)](https://github.com/hitsz-ids/synthetic-data-generator)**  - Specialized framework for generating high-quality structured tabular synthetic data with CTGAN models supporting billion-level data processing. Apache 2.0 licensed.
#### Distributed Training
- **[DeepSpeed](https://github.com/deepspeedai/DeepSpeed)**  - Extreme-scale training optimizations.
- **[Colossal-AI](https://github.com/hpcaitech/ColossalAI)**  - Unified system for 100B+ models.
- **[Megatron-LM](https://github.com/NVIDIA/Megatron-LM)**  - Distributed training framework and reference codebase for large transformer models at scale.
- **[Composer](https://github.com/mosaicml/composer)**  - MosaicML's PyTorch library for scalable, efficient neural network training with algorithmic speedups.
- **[Ray Train](https://github.com/ray-project/ray)**  - Scalable distributed training.
- **[Nanotron (Hugging Face)](https://github.com/huggingface/nanotron)**  - Minimalistic 3D-parallelism LLM pretraining with tensor, pipeline, and data parallelism. Designed for simplicity and speed.
- **[veScale (ByteDance)](https://github.com/volcengine/veScale)**  - Hyperscale PyTorch distributed training with flexible FSDP implementation for LLMs and RL training at scale.
- **[GPT-NeoX (EleutherAI)](https://github.com/EleutherAI/gpt-neox)**  - Production-grade distributed training framework for large autoregressive transformers, powering models like GPT-J and GPT-NeoX-20B.
- **[RLinf](https://github.com/RLinf/RLinf)**  - Scalable open-source RL infrastructure for post-training foundation models via reinforcement learning. Features M2Flow paradigm for embodied AI and agentic workflows with real-world robotics integrations. Apache 2.0 licensed.
- **[Streaming (MosaicML)](https://github.com/mosaicml/streaming)**  - High-performance data streaming library for efficient neural network training. Streams training data from cloud storage (S3, GCS, Azure) with local caching and deterministic shuffling. Apache 2.0 licensed.
#### Model Quantization & Optimization
- **[LLM Compressor (vLLM)](https://github.com/vllm-project/llm-compressor)**  - Transformers-compatible library for applying various compression algorithms to LLMs for optimized deployment with vLLM. Supports GPTQ, AWQ, SmoothQuant, AutoRound, and FP8/INT8 quantization with seamless Hugging Face integration.
- **[NVIDIA Model Optimizer](https://github.com/NVID