Quickstart

Install ArmorClaude or ArmorCodex, approve your machine, and get your first recorded run in about five minutes.

You need an account at tools.armoriq.ai and either Claude Code or OpenAI Codex already working on your machine.

Once you have an account, the console walks you through the last two steps below at tools.armoriq.ai/start, and tracks your progress as you go.

Create your account

Go to tools.armoriq.ai and sign up with Google, GitHub, or an email address and password. Email sign-up sends a six-digit code, valid for five minutes.

A workspace is created for you automatically on first sign-in. You do not have to name or configure anything.

Connect your terminal

Run one command. It installs the plugin into your assistant, plus the small armoriq CLI used for login and status.

curl -fsSL https://armoriq.ai/install_armorclaude.sh | bash
curl -fsSL https://armoriq.ai/install_armorcodex.sh | bash

The command first asks Connect your ArmorIQ account now? (Y/n). Answer yes and it prints a short code and opens a browser tab asking you to approve this machine. Check the code in the browser matches the one in your terminal, then approve. If you decline, or if the terminal cannot prompt at all, the install still finishes and tells you to run armoriq login when you are ready.

Nothing is created until you approve. A code is valid for 15 minutes; if yours expires, run the command again for a fresh one.

On approval, one key is written to ~/.armoriq/credentials.json with owner-only permissions. It reads nothing else on your machine and sends no source code.

If you would rather not run shell commands yourself, the console offers a Copy as a prompt button that hands the whole install to your agent.

Run your first task

Ask your assistant to do one small thing, so there is something to look at. The console suggests:

List the files in this directory and tell me what this project is.

The prompt, the tools it used, the tokens, and the verdict arrive in the console within a few seconds. The setup page watches for it and marks itself complete the moment the first run lands.

What you should see

Open tools.armoriq.ai. The home page replaces the setup timeline with your Workspace overview, and your run appears under Recent runs.

From there, three places are worth a look on day one:

If nothing arrives

Work down this list.

  1. Check the machine is connected. In the console, open Settings → API Keys. An approved machine shows a key with a device name attached. No key means the approval did not complete, so run the install command again.
  2. Check the install actually reached the login step. The installer asks Connect your ArmorIQ account now? (Y/n) before anything is created, and in a terminal it cannot prompt in (CI, and some IDE terminals) it skips login entirely and prints Run armoriq login to connect later. Running the install command again hits the same skip, so run armoriq login directly instead.
  3. Start a new assistant session. The plugin attaches when a session starts. A session that was already running when you installed will not be recorded.
  4. Give it a moment. The setup page polls every four seconds, then slows to every fifteen after two minutes. It never stops, and it re-checks whenever you return to the tab, so you do not need to reload.

A session that reports token usage but no tool traces is a known shape, not a bug: historical token imports cannot reconstruct policy decisions, errors, or retries. The console labels it rather than guessing.

Next

Rules are the other half of the product, and the part worth doing deliberately. Start with how a decision is made, then write your first rule.

On this page