Aug 31 to Sep 6, 2026

Documents a tuning ladder with measured throughput across context lengths.

Pure-CPU C implementation aimed at laptops with no discrete GPU.

Brian Chew built a single-file venue screen where a terminal-style ASCII bot greets each Luma check-in by name and emails them a Cursor credit code.

Optimised serving stack for the Blackwell workstation card.

Targets old V100 server cards, supporting Flash-Next and GLM-5.3-Flash.

Experimental one-box DGX Spark deployment.

Community measurements of memory needs at different quants and contexts.

Showcases long-context concurrent serving on used cards.

Shivay Lamba had Astra turn his photo into a Blender character and place it in an editorial WebGL site with a rotating Speaking Orbit globe that maps his talks worldwide.

SGLang patch and kernels for a single mid-range card.

Antonio Leiva had Astra produce six browser games and an arcade website to launch them, published on a chatgpt.site address, with a short video tour of each.

A reproducible recipe with attributed oMLX and MLX patches, setup scripts and raw benchmarks for the 6-bit quant on an M3 Ultra. Written by an Austin AI consultant for people running big models on Macs.

Hybrid MoE serving on a 128GB Spark.

Llama.cpp tree for hybrid integrated and external GPU machines.

Connects DeepSeek Harness to Busabase, a database that can host apps and skills for the agent. It was shared on Hacker News as a way to give agents durable app state.

Android port of the OpenSim biomechanics toolkit written with Astra, with a demo video in the repo.

Simon Willison's pelican-riding-a-bicycle test run through Blender.

ROCm patches speeding Flash-Next decode on AMD AI Pro cards.