Artificial intelligence

Local AI assistant

Investigating small language models for useful everyday tasks, with personal data kept on the device.

Research conceptConceptual experiment · No measured results

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