llamafile

llamafile is a Mozilla Builders project that collapses the complexity of running large language models into a single-file executable. It combines llama.cpp with Cosmopolitan Libc so that one file runs locally on most operating systems and CPU architectures without any installation. The project also includes whisperfile, a single-file speech-to-text tool built on whisper.cpp using the same packaging approach. llamafile is fully open source under the Apache 2.0 license and is actively maintained by Mozilla.ai.

• Single-file distribution — Download one .llamafile executable and run it directly; no Python environment, Docker, or package manager needed.
• Cross-platform support — The same file runs on macOS, Linux, Windows, BSD, and multiple CPU architectures thanks to Cosmopolitan Libc.
• Built on llama.cpp — Inherits broad model compatibility and GPU acceleration support from the widely-used llama.cpp inference engine.
• whisperfile included — A companion single-file speech-to-text tool built on whisper.cpp for audio transcription and translation, requiring no installation.
• Local inference — All computation runs on your own hardware; no data is sent to external servers.
• Pre-built model files — Ready-to-run llamafiles for popular models (e.g., Qwen, LLaVA) are hosted on Hugging Face for immediate download.
• Quick start — Download a .llamafile, mark it executable (chmod +x), and run it; Windows users rename with .exe extension.
• Versioned releases — Stable and legacy releases are available on GitHub; pre-built llamafiles indicate which server version they bundle.
• Open source — Apache 2.0 licensed core; llama.cpp and whisper.cpp modifications are MIT licensed for upstream compatibility.