RunAnywhere SDK

Run LLMs locally with one command

Tool for running large language models locally on your machine with a simple CLI interface. Download and run Llama 3, Mistral, Gemma, Phi, Code Llama, and dozens of other open-source models with a single command. Features model management, GPU acceleration (NVIDIA/AMD/Apple Silicon), OpenAI-compatible API server, Modelfile for customization, and multi-model switching. Ideal for offline AI development, privacy-sensitive use cases, and local testing. 120K+ GitHub stars.

High-performance local LLM inference in C/C++

llama.cpp is the foundational C/C++ library with 75K+ GitHub stars powering local LLM inference on consumer hardware. Provides optimized CPU and GPU inference for quantized models in GGUF format. Supports LLaMA, Mistral, Phi, Gemma, and most open-weight families. Features 2-8 bit quantization for reduced memory, multi-GPU support, context extension, grammar-constrained output, and an OpenAI-compatible API server. The engine behind Ollama and LM Studio.

High-throughput LLM serving engine

vLLM is an Apache-2.0 LLM inference and serving engine focused on high-throughput self-hosted model APIs. It combines PagedAttention, continuous batching, prefix caching, quantization options, OpenAI-compatible serving, structured outputs, metrics, Docker/Kubernetes deployment guidance and integrations with agent and LLM frameworks.

Cross-platform on-device AI model runtime

Nexa SDK enables running frontier LLMs and multimodal models locally across PC, mobile, IoT, and wearables with automatic hardware acceleration for GPU, NPU, and CPU. It supports Qwen, Gemma, Llama, DeepSeek models with Python/C++ desktop SDKs, Android/iOS mobile SDKs, and Docker for edge deployment. Includes an OpenAI-compatible API server with chat and function calling support.

High-performance mobile neural network inference

NCNN is Tencent's high-performance neural network inference framework optimized for mobile and embedded platforms. It features pure C++ with zero dependencies, ARM NEON assembly optimization, Vulkan GPU acceleration, and sophisticated memory management for minimal footprint. Supports importing models from PyTorch, ONNX, Caffe, TensorFlow, and Keras with 8-bit quantization and half-precision storage for efficient on-device deployment across Android, iOS, and Linux.

Distributed AI compute engine for scaling Python and ML workloads

Ray is an open-source distributed computing framework built for scaling AI and Python applications from a laptop to thousands of GPUs. It provides libraries for distributed training, hyperparameter tuning, model serving, reinforcement learning, and data processing under a single unified API. Ray's public site highlights OpenAI and other enterprise users. Maintained by Anyscale with Apache-2.0 open-source licensing.

Unified framework for fine-tuning 100+ large language models

LLaMA-Factory is an open-source toolkit providing a unified interface for fine-tuning over 100 LLMs and vision-language models. It supports SFT, RLHF with PPO and DPO, LoRA and QLoRA for memory-efficient training, and continuous pre-training. The LLaMA Board web UI enables no-code configuration, while CLI and YAML workflows serve advanced users. Integrates with Hugging Face, ModelScope, vLLM, and SGLang for model deployment.

2x faster LLM fine-tuning with 70% less VRAM on a single GPU

Unsloth is an open-source framework for fine-tuning large language models up to 2x faster while using 70% less VRAM. Built with custom Triton kernels, it supports 500+ model architectures including Llama 4, Qwen 3, and DeepSeek on consumer NVIDIA GPUs. Unsloth Studio adds a no-code web UI for dataset creation, training observability, model comparison, and GGUF export for Ollama and vLLM deployment.

Microsoft's open-source frontier voice AI for long-form multi-speaker audio

VibeVoice is Microsoft's open-source voice AI family with both TTS and speech recognition models. The TTS model generates up to 90 minutes of expressive multi-speaker audio with 4 distinct voices. VibeVoice-ASR transcribes 60-minute recordings in a single pass with speaker identification and timestamps. Built on continuous speech tokenizers at 7.5 Hz and next-token diffusion, it compresses audio 80x more efficiently than Encodec while preserving fidelity.

High-performance OLTP graph-vector database in Rust built on object storage for AI memory

HelixDB is an open-source, unified graph-vector database engineered in Rust that merges relational, graph, and vector workloads into a single OLTP engine, using LMDB local caching and S3 object storage for scalable agent memory.

Modular graph-based RAG pipeline using hierarchical knowledge graph community summaries

Microsoft GraphRAG is an open-source retrieval framework that transforms unstructured text into structured knowledge graphs, clusters entities hierarchically using the Leiden algorithm, and generates dataset-wide summaries alongside entity-level local search for multi-hop reasoning.