AI Stays Local

DocumentationGetting started

Overview

AI Stays Local runs an open-weight model on hardware you control. You grant access to folders you choose, ask questions in ordinary language, and check each answer against the file and page it cites.

What it is

A local runtime and the product around it: a desktop application, a container image, a command line and an OpenAI-compatible HTTP endpoint that listens on loopback. One runtime, four ways to reach it.

What it is not

  • Not a hosted service. There is no account, and no request leaves your machine during local inference.
  • Not a wrapper around someone else’s chat interface.
  • Not a claim to match a frontier hosted model. Which models run, and how fast, depends on your hardware.

Where things stand

SurfaceState
RuntimeOperational for approved testers. Local inference on hardware you control.
Cited document answersOperational for approved testers. Answers that name the file and page, or decline.
Desktop appPlanned. The guided experience. Not built yet; no installer exists.
DockerPlanned. Not published yet; no image exists.
CLI and local APIPrivate beta. Distributed to selected testers.
Consumer macOS buildPlanned. Not built, and not measured.
Consumer Windows buildPlanned. Not built, and not measured.
Enterprise deploymentPlanned. On infrastructure the customer owns. Not offered yet.
Qwen3.5 122B-A10BPrivate validation only. Interactive, validated on the recorded professional hardware profile. Not available to obtain.
GLM-5.2 744BResearch demonstration. Executed by AI Stays Local at 0.17 tokens per second. Not usable interactively, and not available.

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.