Get started

Install the cluster CLI

Install sentinel, the CLI for a cluster, and run the first commands.

The sentinel command does the tasks that the console does, for a cluster. It also runs a local Cortex proxy around any command.

sentinel is not agentop. agentop is the viewer for Cortex on your computer, and Quickstart on a laptop installs it. The two programs are separate.

For each command and each option, see the CLI reference.

Step 1: install the program#

curl -fsSL https://raw.githubusercontent.com/sentinel/sentinel-cli/main/downloadSentinel | sh
export PATH="$PATH:$HOME/.config/sentinel"

Add the export line to the profile of your shell to make the change permanent. You can also move the program to a directory that is already on your path:

sudo mv "$HOME/.config/sentinel/sentinel" /usr/local/bin/

Confirm the installation:

sentinel version

Step 2: sign in#

For a Kind cluster that you installed from this repository:

sentinel login
sentinel agents list

For a shared server, give the address one time:

sentinel --server https://<your-host>/api/v1 login
sentinel agents list

The command stores each server, namespace and token as a named context in ~/.config/sentinel/config.toml. The arrangement is the same arrangement that kubectl uses:

sentinel config get-contexts
sentinel config use-context dev
sentinel config set-context --namespace team1

To examine the permissions in your token:

sentinel auth status

Step 3: deploy an agent#

sentinel agents import from-image \
  --name orders \
  --containerImage ghcr.io/x/y:latest \
  --envVar LOG_LEVEL=debug

sentinel agents wait orders --timeout 5m
sentinel agents get orders

The agents wait command exits when the agent is ready, so you can put it in a sequence of commands. It also exits with an error when the agent cannot become ready. The causes are a failed build, a rollout that passed its time limit, or a name that the server does not have.

The sentinel tools commands operate on tools in the same way.

Run a command behind Cortex#

You can put a local Cortex proxy around any command, with no cluster. This method is the method that Quickstart on a laptop uses, with explicit configuration:

sentinel authbridge exec \
  --config ./authbridge.yaml \
  -- claude "explain this repo"

The --config option accepts a local YAML file or an address that serves YAML. The command passes each argument after -- to your program without a change, and exits with the exit status of your program.

The command sets HTTP_PROXY. It sets HTTPS_PROXY and the certificate variables also when the TLS bridge runs. It does not change a variable that you already set. It stops each part when your program exits.

The log of Cortex goes to the file that --logfile names. The default file is /tmp/authbridge.log. A separate file keeps the log out of the output of your program.

Next#

Edit this page on GitHub