Objective
Assess whether a compact local model can summarize user-provided notes and suggest a useful next action within the constraints of consumer hardware.
What would be tested
- Use synthetic notes with a fixed set of expected facts to check whether summaries preserve meaning.
- Measure cold-start time, response latency, peak memory, and behavior on low-power hardware.
- Compare answers with and without source text, and check whether uncertainty is made visible.
- Inspect network traffic to confirm which data, if any, leaves the device.
Current observations
This is a conceptual experiment. No model has been benchmarked and no performance or privacy result is claimed. The proposed evaluation treats local execution and reliable answers as separate questions.
Next step
Define a small synthetic task set and a reproducible local environment before choosing a model.
How the sandbox works