Mosaic: a pre-focus layer for local vision models
Every modern vision model already chunks images to understand them. Local models need that chunking made explicit, because they can't paper over a missed detail the way a frontier model can.
Catching the 7,000-character write
When your local model's tool call drops a required parameter and a long file almost gets thrown away.
The bottom-up edit rule
When a model queues five edits against one file, working top-down is a bug. Here's the order that fixed it.
The 12,000-token message I didn't know I was sending
My agent's context window kept jumping from 22% to 60% in a single turn. The leak wasn't where I was looking.
How I'd set up LM Studio today
A recent Llama.cpp update pushed me from 60 tokens per second to 80-plus on the same machine. Here's what I'd run and what I'd turn on.