Agentinosaurus Local Models extension

Runtime: llama.cpp b11960, MIT license. See LLAMA-LICENSE.txt and licenses
contained in the upstream release archives (including LLVM OpenMP).
Source: https://github.com/ggml-org/llama.cpp/tree/b11960
Station installers include the unmodified CPU release archives; GPU archives
are downloaded from the same pinned release only when chosen.
Optional CUDA binaries retain NVIDIA's applicable terms:
https://docs.nvidia.com/cuda/eula/index.html

LiquidAI d1-2B Q8_0 and matching vision projector: LFM Open License v1.0.
See LICENSE-LiquidAI.txt. The model is not licensed under Apache 2.0.
https://huggingface.co/LiquidAI/d1-2B-GGUF/tree/5ebcce3bab1c0521f2cab68742a647c7d038adbe
Weights are fetched directly from the pinned upstream repository, unmodified.
Station verifies exact artifact size and SHA-256 before use.
