LibLayaX: run the Laya decision model in your app at 670 decisions/sec, no server needed
felipedaragon · reddit · 2026-10-03
Developer felipedaragon released four MIT-licensed open-source projects — LibLayaX (C/C++), RLaya (Rust), LLaya (Lua), and DLaya (Delphi/Free Pascal) — letting apps embed the Laya decision model directly, with no server or Python.
- Laya doesn't chat: given text plus a question (e.g. "is this customer asking for a refund?"), it returns a yes/no probability in a fraction of a second — built for agent routing, ticket triage, and quick classifications
- Roughly 670 decisions per second on a laptop GPU; Windows/Linux/macOS on CPU and GPU
- Built on Nandakishor Mukkunnoth's laya model and Lars Karlslund's laya.cpp port
The build story is notable too: the author wrote no code himself — Claude wrote it all, while he planned, tested on real machines, and reported back. Total time from first prompt to release: about 32 hours, including a nasty Windows-on-ARM freeze bug only reproducible on his actual hardware.
More from coding & agent
- Claude-coded video zooms 42 powers of ten from quarks to the universe in 94s, no video model — imjustnewatai · 2026-10-03
- User tasks Muse agent with filing a small claims court lawsuit — Scobleizer · 2026-10-03
- Grok Bot, Meta Muse, Dot, Hermes: which AI agent to use for which task — HarperSCarroll · 2026-10-03
- Stanford CS224V Agentic AI course opens: a bottom-up anti-hallucination stack for trustworthy agents — stanfordnlp · 2026-10-03
- Solo dev builds FTL-like airship game in a week with Claude, shipping 7 languages and 100 achievements — MosskeepForest · 2026-10-03
- Cloudflare's Clef decision models (27B and 9B) now available on Ollama — ollama · 2026-10-03