Sep 14 to Sep 20, 2026

VLLM patch and guide for running Flash-Next on a CMP 170HX with streamed embeddings.

Serving the model through ExLlamaV3 quantisation.

Cheap dual-card build that also scales to 800 tok/s at 23 concurrent users.

Jev plays Tetris: a games & real time build posted by Alan Daitch (@AlanDaitch) that puts Jev, TypeSafe AI's System One decision model, to work, with a demo video on the post.

Convert → quantize → deploy path that turns Laya (and Kev, NanoJev, PlayJev) into an INT8 ONNX service on the official wire protocol. Runtime is onnxruntime, tokenizers, numpy.

An open-source router that asks Jev or a self-hosted Laya how hard each prompt is, then picks the model and reasoning effort so simple turns stay cheap.

Elixir library that downloads the official checkpoint on first load and answers typed choice and noul questions; pick EMLX on Apple Silicon or EXLA for CPU/NVIDIA.

No Python, no torch, no ONNX export. Loads the ungated safetensors directly and runs any of the three checkpoints on CPU, Metal or CUDA.

Two LLM routers behind one endpoint compared on 180 labelled requests: Laya picks small, medium or powerful in one forward pass versus a GPT-5 nano reasoning router. Local and free for the Laya side.

Independent Jev-vs-Laya measurements plus a local-first cascade: Laya decides first, high-confidence answers execute, low-confidence ones escalate to Jev. Reported as matching Jev's accuracy at lower cost.

Allen Porter's native Assist conversation agent scores intents and entities in one Laya pass, runs device commands locally and escalates open-ended queries to a fallback LLM.