llamafile, a Mozilla project, collapses an entire model and its runtime into one executable you can download and run. There is no installer, no dependencies, and no network connection required, which makes it ideal for air-gapped or fully offline use. Because everything runs locally, your prompts never leave the device. It ships the model, server, and web UI in a single Apache-2.0-licensed file that runs across Windows, macOS, and Linux, plus several BSDs. It is the most self-contained way to try local AI, though each file is large because the weights are baked in.
llamafile
A whole LLM packed into one offline executable file.
github.comFreeOpen SourceWindowsmacOSLinuxUnited States
PricingCompletely free and open source; no accounts and no subscriptions.
Strengths
- One file holds the model, runtime, and server
- Runs offline with zero installation
- Apache-2.0 and backed by Mozilla
- Cross-platform, even on air-gapped machines
Considerations
- Files are large because weights are bundled
- Fewer conveniences than a full app
Listed inPrivate AI
