GLOSSARY
On-Device AI
AI that runs entirely on your phone, laptop or appliance — no cloud round-trip — trading peak capability for privacy, latency and offline freedom.
Three enablers converged: small models got good (SLMs), phones got NPUs, and quantization shrank memory needs. The result is transcription, translation, image editing and even language models running fully offline — Apple Intelligence's local tier, Gemini Nano in Android, Whisper on a laptop, Ollama on a desktop.
The trade is a capability ceiling: an on-device model will never match a data-center giant on hard reasoning. But for the common cases — summarize this note, transcribe this meeting, fix this photo — local wins on privacy (data never leaves), latency (no network) and cost (no per-query bill). Watch for hybrid designs: local model first, cloud escalation only when the task demands it.