gremlinlabs

the entire AI dev tools space is just bloated VSCode forks wearing different hats

what if instead of forking your way to product-market fit, you built something native from the ground up?

you'd have to start with inference.

multitasking level: extreme

Initial implementation of the LFM3 family of models in Vibe-MLX. Solid numbers on an M3 Max.

Omni and optimizations coming soon.

vibe-jinja 1.2 is released!

3.9×–24.6× faster than Python Jinja2 across our render benchmarks. ⚡️

Better chat-template compatibility, parser hardening, and a reusable render arena—with a new regression gate to keep it fast.

Built in Zig & banned by the Zig community. 🧐