Install
ParetoOps is a single native program. Install it whichever way suits your stack; every channel installs the same CLI. Prebuilt for macOS (Apple silicon and Intel), Linux (x64 and arm64, any distribution) and Windows (x64 and arm64).
Requires Node.js 18 or newer (only to launch the native binary). Run it without installing:
npx pareto-ops --helpOr install it globally:
npm install -g pareto-opspareto-ops --helpnpm install skips optional dependencies with --omit=optional; ParetoOps ships its binary as an
optional, platform-specific dependency, so don’t use that flag for it.
Requires Python 3.9 or newer. This installs both the pareto-ops command and the Python SDK:
pip install pareto-opspareto-ops --helpfrom pareto_ops import ParetoOptimizer, load_trials
result = ParetoOptimizer(load_trials("results.json")).compute(min_success_rate=0.9)print(result.sweet_spot.config_name)GitHub Actions
Section titled “GitHub Actions”- uses: certiqa/paretoops-action@v1 with: results: results.json config: claude-haiku-4-5 min-acc: 0.9The ParetoOps CI Gate action installs the CLI from npm and runs the gate. See Add a CI gate.
Container
Section titled “Container”docker run --rm -v "$PWD:/work" -w /work ghcr.io/certiqa/pareto-ops:v1 analyze results.jsonThe image also works as a docker:// step on Linux GitHub runners.
License key
Section titled “License key”Local analysis (pareto-ops analyze) needs no key and runs 100% offline out of the box. To run the automated CI gate with PR comments and step summaries, generate a free key at paretoops.dev/free (it asks for your email).
Both Free and Pro features read a license key from the PARETO_LICENSE_KEY environment variable:
export PARETO_LICENSE_KEY="POL2-..."Keys are checked offline: the CLI never calls home. For paired regression gates against baselines, explore Pro or join the interest list.