AI Stays Local

DocumentationGetting started

Requirements

What you can run depends on memory, storage speed, operating system, model architecture and the size of the document collection. No GPU farm required for a validated profile.

ProfileMemoryWhat it costs you
Everyday systemsWhatever the machine came withSpeed depends on the machine. No machine of this kind has been measured yet, so nothing is claimed for one.
High-memory desktop or laptopMore RAM is the thing that matters mostProfiles are added to this tier as they complete hardware validation.
AI workstationHigh system or unified memory; the 122B profile peaked at 96.9 GB on the recorded machineMeasured on one recorded machine with 130.7 GB of unified memory. Qwen3.5 122B-A10B is Interactive there, in private validation only and not available to obtain. A machine with less memory is not covered by that evidence.
Enterprise-owned infrastructureSized against the profile you intend to runModel profiles are validated against your hardware rather than promised in advance. The 744B run streams its weights from NVMe at 0.17 tokens per second: a research demonstration, not a product capability.

Measured figures, with the hardware that produced them, are on benchmarks. Consumer macOS and Windows builds are planned and no timings for them are published.

Last updated 2026-09-22T10:05:46-04:00 · Request access

These pages describe the contract: ports, endpoints, volumes, the permission model and the behaviour you can expect. The command reference for the CLI is published with the build each tester receives, so that it matches the binary in their hands rather than a page written ahead of it.

Available through a controlled private beta. Larger model profiles are added as they complete hardware validation.