Verify Kindo
Run this check after you first install Kindo, and after any upgrade you want to confirm. Run it from any machine that reaches the public addresses of the Kindo web app and the API. It doesn’t need to reach the cluster. It needs:
- A working sandbox.
- Models that can call tools. The chat model must call the sandbox shell tool, and in Actions mode the model must also call the file-write tool. Both must accept plain text in the message and follow the organization’s DLP defaults.
- Helm, Python 3.11 or later, uv (recommended), and Node 24.22.1 or a later 24 version on that machine. Installing the check also needs access to the Kindo registry, to PyPI or your PyPI mirror, and to Playwright’s browser downloads.
- A superadmin API key, saved in a file on that machine. If you lost it, see Install Kindo.
- Your CA trusted on that machine, if your certificate comes from a private CA: in the system certificate store and in Chromium’s own store on Linux, in the keychain on macOS.
It leaves test data in the system organization: chats, agents, uploaded files, a Knowledge Store document, a secret, a user group and audit events.
-
Install the CLI with the check on that machine. Install it as the user who runs it, not as root. Replace the version placeholder with your instance’s version, such as 2026.11.0.
Log in with the registry username and password Kindo provided:
Terminal window helm registry login registry.kindo.ai --username '<registry-username>'Download the CLI:
Terminal window download="$(mktemp -d)"helm pull oci://registry.kindo.ai/kindo-cli/kindo-cli --version <version> --untar --untardir "$download"Check the download. On Linux:
Terminal window (cd "$download/kindo-cli" && sha256sum -c SHA256SUMS)On macOS:
Terminal window (cd "$download/kindo-cli" && shasum -a 256 -c SHA256SUMS)With uv (recommended), install the CLI and the check:
Terminal window uv tool install --force "$download"/kindo-cli/kindo_cli-*.whl \--with "$download"/kindo-cli/kindo_e2e-*.whlInstall the browser the check uses:
Terminal window "$(NO_COLOR=1 uv tool dir)/kindo-cli/bin/python" -m kindo_e2e.node_runtime install chromiumCheck its version. It should report the version you downloaded.
Terminal window kindo --versionIf it reports another version, or the command isn’t found, put uv’s tool folder first on your path:
Terminal window export PATH="$HOME/.local/bin:$PATH"Without uv, do the rest of this step instead. Check that Python is 3.11 or newer:
Terminal window python3 --versionInstall with uv instead if it’s older.
Create a Python environment for the CLI. On Debian or Ubuntu, install the Python venv package first:
Terminal window sudo apt-get install python3-venvTerminal window python3 -m venv "$HOME/.kindo-cli"Install the CLI and the check into it:
Terminal window "$HOME/.kindo-cli/bin/pip" install "$download"/kindo-cli/kindo_cli-*.whl "$download"/kindo-cli/kindo_e2e-*.whlInstall the browser the check uses:
Terminal window "$HOME/.kindo-cli/bin/python" -m kindo_e2e.node_runtime install chromiumPut the CLI on your path by adding its folder to the front:
Terminal window export PATH="$HOME/.kindo-cli/bin:$PATH"Check its version. It should report the version you downloaded.
Terminal window kindo --version -
Run the check. Use the addresses you configured for the API and the app. An alias such as an internal hostname or a port-forward is refused.
Terminal window kindo validate \--api-base-url https://api.<domain> \--base-url https://app.<domain> \--api-key-file ./superadmin-api-key -
Read the result. A passing run prints one line and where the full log is:
[pass] <n> of <n> checks passed on https://app.<domain>Full log: ~/.kindo/logs/<time>-validate.logA check that fails or is skipped is listed under that line with its reason. A skipped check doesn’t fail the run. If a check fails, fix the cause and run the command again. Browser screenshots and traces are in
kindo-validate-artifacts/. -
Confirm the cleanup. If the run fails, check that the system organization’s DLP defaults are back to your settings, and remove anything the output says it couldn’t.
To list every check, add --verbose. To watch the browser, add --headed. To save the results as JUnit XML, add --junit <path>. To test one model through the public API, add --api-model <model-id>.
