Yourmodel.Yourmachine.Yourproof.

Pure Rust inference runtime. Merkle-verified models. Sandboxed execution. No cloud, no trust assumptions.

$cargo install yule

The gap.

Parsing

GGUF parsers in C++ have had multiple memory safety CVEs. Yule’s is pure Rust.

Integrity

No local runtime can prove which model produced a response. Yule signs every inference.

Isolation

Model files run with full system access. Yule sandboxes the process at the kernel level.

blake3 hash tree over every tensor. Know your model hasn’t been tampered with.

Job Objects. No child processes. No clipboard. No escape.

Ed25519-signed, hash-chained records. Every inference is tamper-evident.

Hand-written compute shaders. 8.5× over CPU. No CUDA dependency.

yule

$

yule-coreyule-inferyule-gpuyule-verifyyule-attestyule-sandboxyule-apiyule-registryyule-cliyule-bench

CPU (AVX2)0.00 tok/s

Vulkan (RTX 4090)0.00 tok/s

8.5×speedup

TinyLlama 1.1B Q4_K_M, 2048 context. Single-user decode.

$cargo install yule

Run your first inference in 30 seconds.