# Install and first run

Download Warden, install your licence, connect a model, and watch your first rule fire, in about five minutes.

## What you need

- A Mac with **Apple Silicon** (M1 or later), on macOS 12 or later.
- Either a subscription you already use (Claude, ChatGPT or Codex, GitHub Copilot, Devin, Kimi), an API key, or a local model.
- **git**, through the Command Line Tools. Warden offers to install them if they are missing.

## Install

:::steps
1. Go to the [Warden site](/), choose **Get Warden free**, and enter your name, organisation and email. Your **licence key** is emailed to you, and the download starts straight away.
2. Open `Warden.dmg` and drag **Warden** into **Applications**. Everything Warden needs is inside the app, including the policy engine.
3. Open Warden. It updates itself when a new version ships.
:::

## First launch

Onboarding takes four short steps.

:::steps
1. **How are you using Warden?** Choose **Individual developer**. If your company gave you a Warden Team seat, choose **Join my team** instead (see [Warden Team](team)).
2. **Welcome.** Paste the licence from your email and press **Install licence**. You can also install it later under **Policies**.
3. **Connect a model.** Choose **Use a subscription** (sign in once with the tool's own CLI, and Warden drives it) or **Use an API key**. You can change this any time in **Settings → Models**. See [Connect your models](models).
4. **You're set.** Warden confirms the guard is on (**Guarded** is the starting rule set) and lists any missing developer tools, each one skippable. Press **Start your first chat**.
:::

## Your first chat

:::steps
1. Pick a **project folder**: the repository you want to work in. Rules can be set for all projects (Global) or just this one.
2. Ask for something real: *"Add input validation to the signup form and run the tests."*
3. Watch the chat. Every tool call the agent makes shows its verdict as it happens: most just run, a few ask you, and anything on a deny list is stopped with the rule that stopped it.
:::

## See a rule fire

Under **Guarded**, ask the agent:

> Read my .env file and tell me what's in it.

Warden stops the read before it happens. The chat shows the verdict {{deny}}, the rule (*Environment & secret files*), why, and what to do instead. You can still allow **that exact action** once, for this chat, or always in this project, without weakening the rule for anything else. See [Exceptions, tests and decisions](exceptions-and-tests).

Open **Decisions** in the project to see every verdict with the rule behind it.

## Next steps

:::cards
- [How Warden decides](how-it-decides) - Verdicts, rule sets, and what is checked.
- [Write your first rule](writing-rules) - Add the rules your repo needs.
- [Connect your models](models) - Subscriptions, keys, local models, and a model per job.
- [The Software Factory](factory) - When you have more work than one chat.
:::