Local AI headroom

What it does

• You pick a model family and size, 4, 8 or 16-bit, and a context length from 4K to 128K.
• Estimates what it needs (weights, the context cache and runtime), then compares with what the GPU can use and what is free right now.
• Answers Fits, Tight, Make room or Won't fit, with the line that decided it.
• Make room suggests relaunching or quitting apps that hold memory, through the normal Hedroom actions.
• Notices Ollama, LM Studio, Jan and GPT4All when they are running, and says what they use.
• Can raise the GPU memory limit until you restart, after a confirmation that says how little is left for macOS.

What it never does

• Download, load or run a model, or talk to a model app.
• Promise speed. It estimates memory only.
• Offer a limit that leaves macOS less than 8 GB.
• Raise the limit without your confirmation. When Hedroom quits it puts the limit back to what it was, unless you changed it yourself.

Free or Pro

The check and the reading of the GPU memory limit are free. Raising the limit is part of Pro. Resetting it is never locked.

The 7-day trial includes everything in Pro. Full pricing table.

Honest limits

• The estimate can be off by about 15%. Layer and head numbers are typical for a model family, not read from your file.
• macOS does not publish its default GPU limit. Hedroom shows its own reading: about two-thirds of memory up to 36 GB, three-quarters above.
• On Macs with 8, 16, 18 or 24 GB, Hedroom does not offer a higher limit, because the most it would allow is no more than macOS already gives the GPU.
• Raising the limit needs the same helper as fan control, which requires approval in System Settings. A higher limit leaves less for macOS, and your Mac can slow down or become unstable if memory runs short. It lasts until you restart.