Ollama, Continue top local M3 Air stack

On a 24GB M3 Air, the smartest move is not the biggest model, it's the cleanest workflow. You'll usually get more mileage from a lightweight runtime and editor integration than from trying to squeeze a giant model into a fanless laptop.

• –Ollama is the easiest free backend and has the broadest ecosystem for local apps, agents, and terminal-first workflows.
• –LM Studio is the friendliest standalone app, though some Apple Silicon users prefer MLX-native wrappers if they want maximum speed.
• –Continue is the best no-cost IDE layer if the real goal is coding help without paying for a full AI editor subscription.
• –The thread's model advice stays in the 7B-14B range, with quantized coder models like Qwen2.5-Coder-14B in Q5 feeling like the sweet spot.
• –A minority view says the fanless Air is still the wrong machine for serious local inference, so free cloud LLMs may win if latency matters more than privacy.