{"meta":{"title":"D-Central \u2014 Local AI Runtime Comparison","description":"Comparison of 16 local AI inference runtimes for running open-weight LLMs on your own hardware: category, interface, model formats, OpenAI-compatible API, GPU support, license and OS.","generated":"2026-09-09T21:59:31+00:00","version":"1.0","license":"https:\/\/creativecommons.org\/licenses\/by\/4.0\/","license_name":"CC BY 4.0","source":"https:\/\/d-central.tech\/local-ai-runtime-comparison\/","record_count":16,"provenance":"Compiled from the projects' own repositories and documentation; D-Central is one node in the open-source AI ecosystem, not its maintainer.","disclaimer":"Local-AI tooling moves fast: confirm the current license, formats and API against the project repo before deciding. Many tools wrap llama.cpp under the hood."},"rows":[{"id":"ollama","name":"Ollama","category":"engine","interface":"CLI + server","model_formats":"GGUF (can import safetensors)","openai_api":"yes","gpu_support":"NVIDIA (CUDA), AMD (ROCm), Apple (Metal)","license":"MIT","os":"Linux \/ macOS \/ Windows","best_for":"simplest local model runner","source":"github.com\/ollama\/ollama"},{"id":"llama-cpp","name":"llama.cpp","category":"engine","interface":"CLI + server (also a C\/C++ library)","model_formats":"GGUF","openai_api":"yes","gpu_support":"NVIDIA (CUDA), AMD (ROCm), Apple (Metal), Vulkan, CPU","license":"MIT","os":"Linux \/ macOS \/ Windows","best_for":"underlying GGUF inference engine many tools wrap","source":"github.com\/ggml-org\/llama.cpp"},{"id":"lm-studio","name":"LM Studio","category":"app","interface":"Desktop GUI (+ local server, CLI 'lms')","model_formats":"GGUF, MLX","openai_api":"yes","gpu_support":"NVIDIA (CUDA), AMD (ROCm\/Vulkan), Apple (Metal)","license":"Closed (proprietary); free for personal use","os":"Linux \/ macOS \/ Windows","best_for":"non-technical desktop users","source":"lmstudio.ai"},{"id":"vllm","name":"vLLM","category":"server","interface":"Server (API) + Python library","model_formats":"HF safetensors, GPTQ, AWQ","openai_api":"yes","gpu_support":"NVIDIA (CUDA) first, AMD (ROCm)","license":"Apache-2.0","os":"Linux (Windows via WSL\/Docker)","best_for":"max-throughput GPU serving","source":"github.com\/vllm-project\/vllm"},{"id":"text-generation-webui","name":"text-generation-webui (oobabooga)","category":"app","interface":"Web UI (+ API)","model_formats":"GGUF, EXL2, GPTQ, AWQ, HF safetensors","openai_api":"yes","gpu_support":"NVIDIA (CUDA), AMD (ROCm), Apple (Metal), CPU","license":"AGPL-3.0","os":"Linux \/ macOS \/ Windows","best_for":"power users wanting many backends and quant formats","source":"github.com\/oobabooga\/text-generation-webui"},{"id":"jan","name":"Jan","category":"app","interface":"Desktop GUI (+ local server)","model_formats":"GGUF (llama.cpp engine)","openai_api":"yes","gpu_support":"NVIDIA (CUDA), AMD (Vulkan), Apple (Metal)","license":"Apache-2.0","os":"Linux \/ macOS \/ Windows","best_for":"open-source offline desktop ChatGPT alternative","source":"github.com\/menloresearch\/jan"},{"id":"gpt4all","name":"GPT4All","category":"app","interface":"Desktop GUI (+ local API server)","model_formats":"GGUF","openai_api":"yes","gpu_support":"NVIDIA\/AMD (Vulkan), Apple (Metal), CPU","license":"MIT","os":"Linux \/ macOS \/ Windows","best_for":"privacy-focused desktop chat with local documents","source":"github.com\/nomic-ai\/gpt4all"},{"id":"koboldcpp","name":"KoboldCpp","category":"app","interface":"GUI launcher + Web UI + Server (API)","model_formats":"GGUF","openai_api":"yes","gpu_support":"NVIDIA (CUDA), AMD (Vulkan), Apple (Metal), CPU","license":"AGPL-3.0","os":"Linux \/ macOS \/ Windows","best_for":"single-binary story writing and roleplay","source":"github.com\/LostRuins\/koboldcpp"},{"id":"localai","name":"LocalAI","category":"server","interface":"Server (API)","model_formats":"GGUF (plus multimodal\/whisper\/diffusers backends)","openai_api":"yes","gpu_support":"NVIDIA (CUDA), AMD (ROCm), Apple (Metal), CPU","license":"MIT","os":"Linux \/ macOS \/ Windows (Docker-first)","best_for":"self-hosted drop-in OpenAI API replacement","source":"github.com\/mudler\/LocalAI"},{"id":"llamafile","name":"Llamafile","category":"engine","interface":"CLI + server (single-file executable)","model_formats":"GGUF","openai_api":"yes","gpu_support":"NVIDIA (CUDA), AMD (ROCm), Apple (Metal), CPU","license":"Apache-2.0","os":"Linux \/ macOS \/ Windows \/ BSD","best_for":"single portable file that runs across OSes","source":"github.com\/mozilla-ai\/llamafile"},{"id":"open-webui","name":"Open WebUI","category":"frontend","interface":"Web UI","model_formats":"N\/A (uses a backend)","openai_api":"no","gpu_support":"N\/A (backend-dependent)","license":"BSD-3-Clause (v0.6.6+ adds a branding clause + CLA)","os":"Linux \/ macOS \/ Windows (Docker)","best_for":"self-hosted chat UI in front of Ollama or OpenAI-compatible APIs","source":"github.com\/open-webui\/open-webui"},{"id":"mlx-lm","name":"MLX \/ mlx-lm","category":"library","interface":"Python library + CLI","model_formats":"MLX (converts from HF safetensors)","openai_api":"yes","gpu_support":"Apple (Metal) only","license":"MIT","os":"macOS (Apple Silicon) only","best_for":"Apple Silicon native inference","source":"github.com\/ml-explore\/mlx-lm"},{"id":"hugging-face-tgi","name":"Hugging Face TGI","category":"server","interface":"Server (API)","model_formats":"HF safetensors, GPTQ, AWQ","openai_api":"yes","gpu_support":"NVIDIA (CUDA) first, AMD (ROCm), Intel Gaudi","license":"Apache-2.0","os":"Linux (Docker-first)","best_for":"production serving of Hugging Face models","source":"github.com\/huggingface\/text-generation-inference"},{"id":"exllamav2-tabbyapi","name":"ExLlamaV2 \/ TabbyAPI","category":"engine","interface":"Python library + Server (API via TabbyAPI)","model_formats":"EXL2, GPTQ","openai_api":"yes","gpu_support":"NVIDIA (CUDA)","license":"MIT (ExLlamaV2); AGPL-3.0 (TabbyAPI)","os":"Linux \/ Windows","best_for":"fast EXL2 quantized inference on NVIDIA GPUs","source":"github.com\/turboderp-org\/exllamav2"},{"id":"sglang","name":"SGLang","category":"server","interface":"Server (API) + Python library","model_formats":"HF safetensors, GPTQ, AWQ, FP8","openai_api":"yes","gpu_support":"NVIDIA (CUDA) first, AMD (ROCm)","license":"Apache-2.0","os":"Linux","best_for":"high-throughput structured\/agentic GPU serving","source":"github.com\/sgl-project\/sglang"},{"id":"anythingllm","name":"AnythingLLM","category":"app","interface":"Desktop GUI + Web UI (Docker)","model_formats":"GGUF (built-in engine; also connects to external providers)","openai_api":"no","gpu_support":"NVIDIA (CUDA), Apple (Metal), CPU","license":"MIT","os":"Linux \/ macOS \/ Windows","best_for":"all-in-one local RAG desktop app","source":"github.com\/Mintplex-Labs\/anything-llm"}]}