Install the Gauntlet CLI

Linux and macOS (amd64, arm64); on Windows, WSL.

curl -fsSL https://app.usegauntlet.dev/api/cli/install.sh | sh

Gauntlet Cloud needs an invitation (request access). Self-hosted: the same command with your server's URL.

Try Gauntlet on this machine

gauntlet demo

Needs Docker on x86-64 Linux or WSL2. Starts Gauntlet in one Docker container (Kubernetes included; first start downloads about 3 GB) and opens it signed in. gauntlet demo stop keeps your projects; gauntlet demo rm removes it.

Where	`~/.local/bin/gauntlet`; elsewhere: `GAUNTLET_INSTALL_DIR=/usr/local/bin`
Pin a version	`GAUNTLET_VERSION=<version>`: installs only if the server serves that version
Check	`gauntlet version` shows the CLI's and the server's version

Quick start: test your branch

gauntlet login --server https://app.usegauntlet.dev
# asks for an API token: account menu → Account & API tokens
cd your-app
gauntlet init
# writes gauntlet.yaml (from your chart, compose file or Dockerfiles) and .gauntlet.json
gauntlet validate gauntlet.yaml
gauntlet env up
# builds, pushes and starts this branch's environment
gauntlet run --test test.yaml --wait

test.yaml is an AI test (kind: TestTemplate): user journeys in plain words, with success criteria. The console's test editor writes one.

init picks your only project, or the one named like the directory, or asks (--project <slug>). Check both files it writes: the registry must be one your cluster pulls from (--registry). Scripts and CI: gauntlet login --token-stdin, or set GAUNTLET_SERVER and GAUNTLET_TOKEN.

Test from Claude Code

claude mcp add -s user gauntlet -- gauntlet mcp

Claude Code can then build its branch, start the environment, run a test and read the findings.

Every command, .gauntlet.json and CI: Docs → CLI and Claude Code (on a self-hosted Gauntlet: Docs in its console).