OpenJet installation documentation

From the upstream OpenJet documentation. For a guided overview, see Install OpenJet.

Install with your agent

Ask Codex or Claude Code to follow the agent installation instructions. It can inspect your hardware, choose a model, configure OpenJet without the wizard, verify inference, and give you usage commands.

Once installed, the noninteractive path is:

openjet setup --non-interactive --dry-run --json
openjet setup --non-interactive --yes --json

Clone and install

git clone https://github.com/l-forster/open-jet.git
cd open-jet
./install.sh

On macOS, use:

./install-macos.sh

On Windows, run:

curl -L https://www.openjet.dev/install.bat -o install.bat && install.bat

This creates a local virtualenv, installs the repo, creates open-jet / openjet command launchers, and exits. Run openjet setup --yes to accept the hardware recommendation or openjet setup for the wizard. It is an editable Python install, not a compiled release build.

If stale in-place compiled extension artifacts exist under src/, the installer removes them so imports continue to use the source tree.

Optional compiled wheel

If you want a compiled package artifact, build it separately:

./scripts/build-compiled.sh

This produces a wheel in dist/ and keeps the normal editable install path separate.

Runtime

OpenJet is local-only. It uses llama-server from llama.cpp with GGUF models.

Recommended path

Start with:

openjet setup

OpenJet will first try to reuse an existing local setup. If pieces are missing, setup can provision them for you.

Setup can:

reuse llama-server if it is already installed
clone and build llama.cpp to produce llama-server
use a local .gguf
download a recommended GGUF directly

Automatic provisioning still needs:

network access
git
cmake
a working build toolchain

Manual build guides are still available if you want to control the runtime yourself:

Public SDK import

For scripts and integrations, use:

from openjet.sdk import OpenJetSession