{"slug":"trymirai-uzu","kind":"smoke","status":"testing","tier":"smoke","title":"uzu: the Rust inference engine that 'absolutely rips'","category":"local-models/runtimes","topic":"speed-claims","repos":[{"repo":"trymirai/uzu","stars":2385,"license":"MIT"}],"claims":[{"claim":"This local Rust-based inference engine absolutely rips","source":"https://x.com/Dan_Jeffries1/status/2109639251079950344"}],"plan":"Smoke first: build the engine from source with cargo at a pinned tag and confirm the CLI runs. A tokens-per-second claim run against a downloaded model is queued.","who_its_for":"People running local models who care about raw speed.","explainer":"A high-performance local inference engine written in Rust. Rocky first checks it builds and runs, then will measure real decode speed so 'rips' becomes a number.","measure":["Does it build from source","Does the CLI run","Decode tokens/second on a real model (claim run, queued)"],"notes_so_far":["Method: build at tag 0.6.9 with the stock Rust toolchain in a container; smoke only for now."],"sneak_peek":["Smoke test: uzu 0.6.9 built with cargo in about 87 seconds, but the build produced only a library and no uzu command-line tool.","uzu targets Apple Silicon, so this Linux GPU box can't run its speed claim. A fair test needs a Mac with Apple Silicon."],"charts":[],"gallery":[],"results":{"jobs":[{"id":21,"title":"uzu: cargo build --release the Rust inference engine, CLI --help (smoke)","kind":"smoke","status":"done","started":"2026-10-18T05:10:26Z","finished":"2026-10-18T05:12:19Z","wall_seconds":113,"install_friction":null,"artifacts":[]}],"smoke_only":true,"ingested":"2026-10-18T15:02:33Z"},"last_test_finished":"2026-10-18T05:12:19Z","results_in":true}