Weights
Roughly the parameter count times bytes per parameter: about 0.55 bytes at 4-bit, 1 byte at 8-bit, 2 bytes at full 16-bit precision.
Personal Agent · Mac configurator
Check the Mac you already own, or plan the one you'll buy, for a Personal Agent that runs its model on your own hardware. No cloud model required.
Memory decides which models fit. Storage decides how many you can keep. This tool works out both in seconds.
Two minutes, no sign-up
Choose your Mac and how you'll use your agent. Results update as you go. Figures are planning estimates; Brian confirms them before any install.
Why memory decides everything
On an Apple silicon Mac, the processor and graphics share one pool of memory. A local model loads all of its weights into that pool, so the memory you bought sets the ceiling on the model you can run.
Speed comes from memory bandwidth. A Mac Studio moves data several times faster than a base Mac mini, so the same model answers faster.
Roughly the parameter count times bytes per parameter: about 0.55 bytes at 4-bit, 1 byte at 8-bit, 2 bytes at full 16-bit precision.
A model like Gemma 4 26B A4B only uses about 4B parameters per word, so it's fast. It still needs memory for all 26B.
The model's context grows with conversation length, roughly 0.1–0.5 MB per token. We plan with a conservative 0.25 MB.
macOS lets the GPU use about two-thirds to three-quarters of memory by default. Brian can raise that limit safely during setup.
Upgrading an older Mac
Upgrades done outside Apple are not covered by Apple's warranty or AppleCare. We explain the options honestly; you decide.
Memory
On every M-series Mac, memory is built into the chip package. No slot, adapter or software unlock adds more. A one-off chip-level rework was demonstrated on an M1 MacBook Air in 2021, but it isn't a reliable service and we don't recommend it.
Your options: run a smaller local model, shorten the context window, or trade up. Brian can recommend the smallest Mac that fits.
Storage · every Mac
A fast Thunderbolt or USB4 SSD is the simplest, warranty-safe way to add room. Your model library can live there while macOS and your agent stay on the internal drive.
Best for: 256 GB and 512 GB Macs, and anyone who wants several models on hand.
Storage · selected models
The 2024 Mac mini with M4 or M4 Pro uses a removable storage module. Third-party modules up to 4 TB exist; installing one requires a full DFU restore from a second Mac. Mac Studio modules are also removable, but larger third-party modules are not yet proven. The 2026 M5 Pro Mac mini's storage is soldered.
Out of warranty: back up first. A mistake can damage the Mac.
Intel Macs
Many Intel Macs accept more memory, but the local-model tools we use require Apple silicon. Extra memory won't make an Intel Mac a practical home for a local agent.
Instead: a Mac mini with Apple silicon is the most affordable route.
Included with the Signature install
This tool gives you a starting point. With the $1,490 Signature install, Brian personally reviews your Mac or your planned purchase on a live call: memory and storage plan, model choice and quantization, macOS GPU memory tuning, and any upgrade or buying decision. Then he installs and tests it with you. It's done by hand, not automated.
Sources · checked October 2026
Speeds are rough estimates from memory bandwidth, not benchmarks. Prices are US list prices where Apple published them and change often; confirm at apple.com before buying.