Articles

Everything published here, newest first.

Cutting an agent's turn from 150k to 10k tokens

Where the tokens were hiding, and why it came before the tuning

September 12, 2026 · 15 mins read

Tuning a 26B MoE agent on a DGX Spark to over 100 t/s

Six configuration changes, each measured the same way

September 10, 2026 · 60 mins read