@abram2k1@mcgeedan42 Lol, no. On most jobs we are cheaper. For enterprise we generally prefer flat rates well below what people pay on GH Runners / Blacksmith so our profit margins are based on how efficient we make the CI. CI runners being paid by the minute means they want your runs to be long.
@hey_itsyash@mcgeedan42 Yup! Blacksmith focuses on raw hardware, which isn't enough to keep up with AI CO/CD demands. We re-build everything w/ no bars held: Linux Kernel, Compilers, Testing Libraries, and a users flakey tests / code.
@techpoto@mcgeedan42@ycombinator For our optimizations that can get 10x+ RN we support Go, Rust, Python, JS/TS, and Java. Everything else usually gets about 2x optimizations, though we can go figure out how to get other frameworks setup for our systems if need be!
@Hussein__khidr@mcgeedan42 Generally, we've been building out as much infra to get latency as low as we possibly can. On our systems cold start is not very representative. We have hooks for agents so we can warm our runners in anticipation of a PR or CI job.
@crescentforeal God, I remember back in middle school if I didn't have my laptop I'd write python in my notepad so I could type it into a file when I got home.
@RISHilift@mcgeedan42 Exactly, our goal is honestly to get every run to sub 10s so CI functions as real time feedback for agents instead of a checkmark at the end of a job.
@Hussein__khidr@mcgeedan42 This one was actually a cold run! Compilation cost us about 50 seconds on our 2 min 53 second job. Speedups came from our splitting system and increased parallelization.