Install the launcher

The launcher, tensorless, is one static binary with no runtime dependencies. Install it with the install script, or download a release archive and check it yourself.

curl -fsSL https://tensorless.dev/install.sh | sh

Where it runs

Each release has an archive for three platforms:

Platform	Archive	What it can do
Linux x86_64	`linux-x86_64`	Everything: pull, serve and benchmark models.
Linux aarch64	`linux-aarch64`	The launcher runs, but the 0.1.0 engine image is published for `linux/amd64` only, so it cannot serve a model on an Arm host.
macOS, Apple silicon	`macos-arm64`	Read recipes and see exactly what would run: `models`, `info`, `capabilities` and `--dry-run`. A Mac has no NVIDIA GPU, so it serves no model.

Serving a model also needs an NVIDIA driver, podman or docker, and NVIDIA container support. Prepare a Linux host for a P100 sets those up.

Install with the script

You need curl or wget, and sha256sum or shasum.

Run the script:
Terminal window
```
curl -fsSL https://tensorless.dev/install.sh | sh
```
It finds your platform, downloads that archive and its .sha256 file from the GitHub release, checks the archive’s sha256, and copies the one tensorless binary to ~/.local/bin. It changes nothing else. You see:
```
Downloading tensorless 0.1.0 for linux-x86_64
Checked sha256 cb0c796fd83a334eb67f163fffb30016f51513f8f6ca4a30e7144875487415eb
Installed /home/you/.local/bin/tensorless
```
If the directory is not on your PATH, the script says so and prints the line to add, for example:
```
/home/you/.local/bin is not on your PATH. Add it, for example:
  export PATH="/home/you/.local/bin:$PATH"
```
Run that line, and add it to your shell’s startup file (such as ~/.bashrc) to keep it.

Two environment variables change what the script does:

Variable	Default	Meaning
`TENSORLESS_VERSION`	`0.1.0`	The release to install.
`TENSORLESS_INSTALL_DIR`	`$HOME/.local/bin`	Where the binary goes. The script creates the directory if needed.

Set them for the sh that runs the script:

curl -fsSL https://tensorless.dev/install.sh | TENSORLESS_INSTALL_DIR="$HOME/bin" sh

On any other platform, the script stops with no release archive for your system, and suggests building from source. See Build the launcher from source.

Install from a release archive

Download the archive for your platform and its .sha256 file. Set P to linux-x86_64, linux-aarch64 or macos-arm64:

V=0.1.0
P=linux-x86_64
curl -fLO https://github.com/iammrduncan/tensorless/releases/download/v$V/tensorless-$V-$P.tar.gz
curl -fLO https://github.com/iammrduncan/tensorless/releases/download/v$V/tensorless-$V-$P.tar.gz.sha256

Check the archive, then unpack it. On Linux:
Terminal window
```
sha256sum -c tensorless-$V-$P.tar.gz.sha256 && tar -xzf tensorless-$V-$P.tar.gz
```
On macOS, use shasum -a 256 -c in place of sha256sum -c. The check prints the archive’s name followed by OK.
The archive’s folder, tensorless-0.1.0-linux-x86_64/, holds the tensorless binary with LICENSE, LICENSES/, NOTICE, THIRD_PARTY_NOTICES.md and README.md. Put the binary on your PATH:
Terminal window
```
mkdir -p ~/.local/bin
install -m 755 tensorless-$V-$P/tensorless ~/.local/bin/
```

Check that it worked

tensorless version

tensorless 0.1.0 (commit d104660be24427080948eafb0522265e1450d642), linux/amd64

tensorless models lists the launch targets this launcher knows. On a machine without an NVIDIA GPU, it cannot tell whether they fit:

MODEL                                            CARD  STATUS    ENGINE         FITS     CACHED
iamMrDuncan/Qwen3.8-27B-PN3-GGUF                 p100  released  engine-v0.1.0  unknown  no
iamMrDuncan/Qwen3.8-27B-UD-IQ3_XXS-Q3Floor-GGUF  p100  released  engine-v0.1.0  unknown  no
fits is unknown: nvidia-smi not found

Next: Prepare a Linux host for a P100, or go straight to Run your first model on a P100 if the host is ready.