Skip to main content

Get started with the CLI

clika-cli is the platform from a terminal. Everything you can do in the web application, from listing devices to starting a benchmark and reading its results, has a command, and the commands are generated from your deployment's own API, so they always match what the platform can do. This page takes you from nothing to your first useful commands. The reference then covers every command.

What you need​

  • An account on a CLIKA Platform deployment, and its address, the one you type into the browser.
  • A terminal on Linux, macOS or Windows. The CLI takes about 15 MB of disk and about 70 MB of RAM (system requirements).

1. Install it​

The CLI is downloaded from your own deployment, so the version you install always matches the platform you talk to.

  1. Sign in to the web application and open Settings, then Developer access.
  2. On the Install the CLI card, pick the tab for your computer and copy the install command shown there. The command needs no token: the CLI download is open to anyone who can reach the platform, and the installer checks the file against the published checksum. (clika-cli auth cli-download-token still mints an attribution token for scripted installs; it is optional.)
  3. Paste it into your terminal and run it.

On Linux and macOS the command runs the shell installer, which installs clika-cli into /usr/local/bin when that is writable and ~/.local/bin otherwise:

curl -fsSL "https://platform.clika.io/files/installation/cli/install.sh" | CLIKA_BASE_URL="https://platform.clika.io" sh

On Windows it is a PowerShell command. It runs in Windows PowerShell 5.1 and PowerShell 7, needs no administrator rights, installs clika-cli.exe into %LOCALAPPDATA%\Programs\clika\bin, and adds that directory to your user PATH:

$env:CLIKA_BASE_URL='https://platform.clika.io'; irm 'https://platform.clika.io/files/installation/cli/install.ps1' | iex

Both installers detect your CPU architecture, check the download against the published checksum, and upgrade an installed CLI in place when you run them again.

Check that it runs:

clika-cli version

2. Log in​

Create an API key on the same Developer access tab, then save it in a profile together with your deployment's address:

clika-cli login --base-url https://platform.clika.io --api-key

The command prompts for the key without echoing it and writes the profile to ~/.config/clika-cli/default.json, readable by you alone. Every later command reads it, so you type the address and the key once. If you work with more than one deployment, give each its own profile with --profile <name> and pick it the same way on every command. On an on-premise deployment with a self-signed certificate, add --insecure-tls to the login and to later commands, or set CLIKA_INSECURE_TLS=1 in the environment. The profile does not record it.

Register your AI clients in the same step​

The installers can also connect Claude Code, Claude Desktop and Codex to the platform once the CLI is in place. Set CLIKA_MCP to the clients you use, separated by commas, and the installer runs clika-cli mcp install for each:

curl -fsSL "https://platform.clika.io/files/installation/cli/install.sh" | CLIKA_BASE_URL="https://platform.clika.io" CLIKA_MCP=claude-code,claude-desktop,codex sh
$env:CLIKA_MCP='claude-code,claude-desktop,codex'; $env:CLIKA_BASE_URL='https://platform.clika.io'; irm 'https://platform.clika.io/files/installation/cli/install.ps1' | iex

Each registration needs a logged-in profile. At a terminal, mcp install offers the login and prompts for the API key without echoing it, so step 2 happens on the way. For an unattended install, set CLIKA_API_KEY as well: the installer first runs login, which saves that key to the default profile, then registers the clients. If a registration fails, the CLI stays installed, the installer names the clients that failed, and clika-cli mcp install <client> retries one. Using with AI (MCP) covers each client.

3. Look around​

clika-cli devices list
clika-cli models list
clika-cli benchmark-groups list

Each command prints a table of what your current project holds, the same rows the web application shows. Add -o json to any command to get the raw answer for a script, and --help to any command or group to see its flags.

4. Read a benchmark from the terminal​

Start a benchmark in the web application, or pick one that already finished, then follow and read it from the shell:

clika-cli benchmarks watch "my-first-run"
clika-cli benchmarks results "my-first-run"
clika-cli benchmarks results "my-first-run" -o json

watch prints the run's progress until every leg has finished and exits non-zero if one did not, which is what makes a benchmark a step in a CI pipeline. results prints the headline metrics, and with -o json the full result for further processing.

Where next​

  • Benchmarks: start a run from the terminal, choose its models, tests and devices, and cancel or share it.
  • Devices: register, inspect and control devices, run a command on one, push a file to it.
  • apply and export: keep job and service definitions as YAML files in git.
  • Using with AI (MCP): give Claude Desktop, Claude Code, Codex or another assistant the platform as tools.
  • Cheat sheet: the commands people reach for most, on one page.