LM-Kit One · Download

Available now.

Signed packages for Windows, Linux and macOS, and a container image, version 2026.9.6. Pick your platform below.

Follow the quickstart Release notes
Five signed packages Free to build and evaluate No activation, no time limit

What ships in every release.

Every package carries the full server: inference, documents, search, agents and the admin console, with nothing else to install.

Windows

Windows x64

An Authenticode-signed installer, or a portable archive that runs from any folder. Runs as a Windows service or for a desktop session.

Linux

Linux x64

A self-contained archive: unpack it and run the server, with nothing else to install.

Linux

Linux Arm64

The same self-contained archive, built natively for Arm64 servers and boards.

macOS

macOS Apple Silicon

A Developer ID-signed package, notarized and stapled, verified per release.

Container

Container, amd64 and arm64

One multi-architecture image on GitHub Container Registry: docker pull ghcr.io/lm-kit/lm-kit-one:2026.9.6. CUDA runtimes included; add --gpus all for an NVIDIA GPU.

Windows installers are Authenticode-signed and the macOS package is notarized. Every release publishes SHA-256 checksums and release notes; the Trust Center covers signing and verification.

What it runs on.

A fresh install listens on the local machine only, and the console prices each model against your hardware before you download it.

Operating systems
Windows x64 / Arm64 · Linux x64 / Arm64 · macOS Apple Silicon
Run modes
Windows service · desktop session · Linux and macOS daemons
Acceleration
CUDA · Vulkan · Metal · CPU AVX/AVX2 fallback

Ready in the first hour.

The install takes minutes. These three steps are what make the first hour count.

Also available

The same engine, on NuGet

LM-Kit.NET ships the identical document intelligence engine as an in-process SDK, free to build and evaluate.

dotnet add package LM-Kit.NET Download LM-Kit.NET

Prove

Walk the first hour

The quickstart shows the three outcomes worth proving first: extraction, cited answers, a client repointed.

Follow the quickstart

Verify

Check your clients now

The coverage matrix states every OpenAI, Anthropic and Ollama endpoint the server answers, and every one it does not.

Read the coverage matrix

Assess

Start the security review

Seven separately controlled boundaries, stated for the people who sign off, so approval work can start on day one.

Read the security page

LM-Kit One

Be running in the first hour.

Follow the quickstart Read the coverage matrix