AI Stays Local

Hardware

What can my computer run?

Answer six questions about the machine you already own. The result tells you which profile it falls into, what class of model that supports, where the evidence for it stands, and what it will not do.

Apple unified memory, system RAM and discrete GPU memory behave differently, so we ask rather than assume.

Optional. Leave blank if you do not know.

Optional. Model weights need room.

Worked out in your browser. Nothing you type here is sent anywhere, stored, or logged.

Compatibility is decided by the operating system, the architecture, the storage, the acceleration and the model profile together, not by memory capacity alone.

What each class of machine is for.

Choose a profile. Every figure is worked out on this page from the operating system reserve, and nothing here is a speed: a speed is only published where it was measured.

Choose a hardware profile

Essential

Not measured yet

A compatible system with 16 GB of memory and adequate SSD storage.

Memory
16 GB
Memory type
Unified memory, System memory
Usable after the operating system
A 16 GB machine keeps 5 GB for the operating system, leaving about 11 GB
Model class
Efficient local assistants
Workloads
Local chat, an efficient model profile, light coding and a limited document collection.
Limitations
Frontier-class models do not run here, and we will not suggest otherwise. Sixteen gigabytes is an entry point, not a guarantee: the operating system, architecture, storage and acceleration all take part in the decision.

Personal

Not measured yet

24 to 48 GB of unified or system memory, depending on the platform.

Memory
24–48 GB
Memory type
Unified memory, System memory
Usable after the operating system
A 32 GB machine keeps 6 GB for the operating system, leaving about 26 GB
Model class
Efficient and mid-size open models
Workloads
A full private assistant, cited documents and more capable model profiles.
Limitations
Large mixture-of-experts profiles remain out of reach at this tier.

Professional

Measured on one machine

64 to 128 GB of memory, or a discrete GPU configuration with enough system RAM and fast NVMe beside it.

Memory
64–128 GB
Memory type
Unified memory, System memory, GPU memory
Usable after the operating system
A 128 GB machine keeps 8 GB for the operating system, leaving about 120 GB
Model class
Professional open models
Workloads
Larger models, extensive document collections, the local API and professional automation.
Limitations
Evidence exists for one machine of this class, with 130.7 GB of unified memory. A machine with less memory is not covered by it, and the largest profiles need the machine to themselves.

Frontier Workstation

Not measured yet

192 to 512 GB of unified memory, or an equivalent high-memory workstation.

Memory
192–512 GB
Memory type
Unified memory, System memory
Usable after the operating system
A 512 GB machine keeps 12 GB for the operating system, leaving about 500 GB
Model class
Frontier-class open models
Workloads
Frontier-class open-weight models and advanced local workloads.
Limitations
Model availability at this tier is decided by validation, not by memory alone. No machine of this class has been measured, so no model is claimed for it.

Private AI Rack

Planned

Multiple trusted computers connected inside a private network.

Memory
Several machines
Memory type
Across private nodes
Usable after the operating system
Per machine. Each node keeps its own memory.
Model class
Very large profiles across machines
Workloads
Designed for distributed execution, team access and very large model profiles.
Limitations
Multi-node execution is designed and not implemented. Each machine keeps its own memory. Nodes would provide aggregate resources through explicit distributed execution; they do not form one shared memory pool.
Available through a controlled private beta. Larger model profiles are added as they complete hardware validation.