Tuning

Optional adjustments to squeeze more performance out of the agent — starting with the Podman machine memory budget on macOS and Windows.

Increase memory limit (macOS and Windows)

On macOS and Windows, Podman runs inside a virtual machine with a fixed memory budget. Larger models need a larger budget. To raise it (example: 16 GB):

$ podman machine stop
$ podman machine set --memory 16384   # MB, so this = 16 GB
$ podman machine start

On native Linux there is no Podman machine — containers use host memory directly and this step does not apply.

Related

• Configuration Reference — CTX_SIZE and other parameters that affect memory use inside the pod.
• CUDA Setup — enable GPU acceleration to run larger local LLM models without increasing RAM.
• Benchmark — repeatable benchmark prompt and results across hardware configurations.