04/09/2026
We did a fun local AI experiment with the Ornith 9b open-weights model. π¦
Before heading off for a flight, Paris Kasidiaris pulled Ornith 9b into Ollama on an M1 MacBook Air (16GB, 6 years old, fanless) to see if it could produce anything useful completely offline. The task was to build a Python load-testing CLI for an internal service that needs to stay resilient under heavy traffic.
A few minutes (and a good chunk of battery) later, it delivered a full CLI: typed arguments, sensible defaults, a token-bucket rate limiter, threadpools, etc.
It wasn't flawless. There was some redundant code and a couple of internal libraries it couldn't reach. But getting roughly 90% of a genuinely useful tool from a coding model that is just 9B (~6GB) in size, running entirely offline on a 6-year-old fanless laptop, is seriously impressive. π³
Kudos to the Ornith team! Super excited to see where this goes.
ποΈ Full writeup (with code): https://withlogic.co/var/log/2026/09/01/ornith-experiment-load-testing