An M2 16GB Mac can run a local LLM server, but slowly
A MacBook Air with an M2 chip and 16GB of memory was tested as a home LLM server after stopping a cloud plan. Another mini PC at home connects to that Mac and receives answers from the hosted LLM.
The setup works, but the are very slow and tool use is heavily limited. The practical question is whether it is worth continuing with local hardware or paying $20 per month for a cloud plan again.
Key points
- An M2 Mac with 16GB of memory was used as a local LLM server.
- A separate mini PC connected to the Mac to use the .
- The setup worked, but response speed was very slow.
- Tool use was limited, making agent work less practical.
- The comparison is local versus a $20 monthly cloud plan.