Anvil Forges Transparent AI Fleets On Your Hardware With Plain Files

Anvil is a new open-source model runner that wraps llama.cpp, providing transparent controls and fleet management for local AI on private hardware. It stores models as plain GGUF files you can browse directly, avoiding opaque blob stores. You can manage multi-node fleets, preview all inference flags before launch, and pull models from Hugging Face.

Sovereignty Labs built anvil as a clean, local-first answer to tools that became opaque or shifted to cloud. The project gives consumer GPU owners and homelab users full transparency over their AI stack. It wraps unmodified llama-server, so there’s no hidden engine changes or performance loss.

Plain files and full visibility

Anvil suits people running local AI on consumer GPUs, home servers, or small office fleets who want control without raw llama.cpp’s complexity. It removes hidden behavior and shows exact model commands beforehand. Privacy-conscious users get a tool that never phones home and stores models openly.

Built on llama.cpp, not a fork

Sovereignty Labs kept anvil a wrapper, not a fork, relying on unmodified llama.cpp for inference. The solo developer, not a professional coder, seeks community testing and feedback. The project stays open-source without a cloud pivot, with GitHub sponsorships as a future option.

"The goal is not to replace llama.cpp. The goal is to wrap the rough edges while keeping attribution, transparency, and control exactly where they belong." — Source: Reddit