0.90.0 — Encrypt a fine-tuned LLM at rest

Encrypt a fine-tuned GGUF at rest.

arkavo model protect wraps local weights as OpenTDF (.gguf.tdf). Load still goes through llama.cpp. Nothing plaintext is written back to disk. Identity is arkavo login (CWT). KAS is consulted once at load. Inference never leaves the machine.

Watch the explainer: https://youtu.be/vCAw7GECydQ

arkavo model protect path/to/model.gguf
arkavo login
arkavo chat --model <name>

Wrapping is additive: a sibling archive never displaces a loadable plaintext, so an existing setup keeps working until you delete the source (--delete-source). Gate the wrap with --attribute <fqn> for your team and your agents.

This is at-rest protection and an access gate. Once the model is in GPU memory it behaves like any local model. A wrap with no attributes is loadable by anyone the KAS admits. Not DRM.

Shipped in #664.

macOS Arkavo Creator required for Arkavo account.

What's Changed

• Add GGUF-TDF wrapping with plaintext GGUF still loadable by @arkavo-com in #664

Full Changelog: 0.89.4...0.90.0