Local models on robserver: what is worth running now?

Field notes on when local models make sense, where hosted models still win, and why the answer changes by task.

Local models are not a universal replacement for hosted systems, but they are becoming strategically useful.

They shine when latency, privacy, cost control or offline capability matter. They struggle when frontier reasoning, multimodal breadth or deep tool integration is required.

The working rule: run local where the task is bounded and repetitive; use frontier hosted models where judgment and synthesis matter most.