What lightweight local AI setup would work on an old Chromebook?

Start by checking the hardware before choosing anything. Run `lscpu` and `free -h` to see the CPU and available RAM. On an old Chromebook, those limits will matter more than whether you’re using Debian. A model that feels usable on one machine may be painfully slow on another.

This combination may be difficult on very weak hardware. Local models often need several gigabytes of RAM, and faster performance usually requires a capable GPU or neural-processing hardware. Ollama can make installation easier, and tools such as AirLLM can reduce GPU requirements by using system RAM, but they won’t make an underpowered Chromebook fast. An Ubuntu-based setup may offer more convenient hardware and AI/ML tooling, while Debian may require more manual configuration.

Instead of searching for one program that does everything, build the assistant in separate parts. You could use a small model through llama.cpp or Ollama, whisper.cpp for speech-to-text, and the Spotify API as a tool the model can call. A simple flow would be speech-to-text → local model → Spotify or another tool → spoken or visual response. Start with a tiny model responding in the terminal, then add voice, one API action, and finally the visual interface.