# Tenzro downloads

> Compiled Tenzro node and CLI binaries, version 0.2.0, for Linux x86_64,
> Linux aarch64 and Windows x86_64. macOS is not yet published. This host
> serves the files; https://tenzro.com/downloads is the page that describes
> them and is the canonical entry point.

Every build carries the llama.cpp CUDA backend for language models and the ONNX
Runtime CUDA execution provider for forecasting, vision, embeddings, segmentation,
detection and audio. The same binary runs on the CPU where there is no GPU.

## Node

- [Linux x86_64](https://downloads.tenzro.com/tenzro-node-linux-x86_64-cuda.tar.gz) — 742 MB, CUDA 13, glibc 2.39 or newer
- [Linux aarch64](https://downloads.tenzro.com/tenzro-node-linux-aarch64-cuda.tar.gz) — 718 MB, CUDA 13, Grace-Blackwell / GH200 / Jetson Thor
- [Windows x86_64](https://downloads.tenzro.com/tenzro-node-windows-x86_64-cuda.zip) — 236 MB, needs the CUDA 13 redistributables and the VC++ runtime

The node archive also contains `rpc-server`, the ggml RPC sidecar the node spawns
on loopback for LAN cluster serving. It is not a separate program to run.

## CLI

- [Linux x86_64](https://downloads.tenzro.com/tenzro-linux-x86_64.tar.gz) — 515 MB
- [Linux aarch64](https://downloads.tenzro.com/tenzro-linux-aarch64.tar.gz) — 629 MB
- [Windows x86_64](https://downloads.tenzro.com/tenzro-windows-x86_64.zip) — 139 MB

## Integrity

- [SHA256SUMS](https://downloads.tenzro.com/SHA256SUMS)
- [MANIFEST.txt](https://downloads.tenzro.com/MANIFEST.txt) — how each artifact was built

These builds are not signed and are not reproducible. SHA256SUMS is the only
integrity statement published with them. Source is Apache-2.0 licensed.

## Elsewhere

- [Downloads page](https://tenzro.com/downloads)
- [Tenzro](https://tenzro.com/)
- [Architecture](https://tenzro.com/architecture)
- [Site content map](https://tenzro.com/llms.txt)