---
title: "Quickstart"
description: "Connect your inbox, create a Shell with rules you choose, and connect your AI agent to Gmail. About 10 minutes, no credit card."
route: /docs/get-started/quickstart
---

# Quickstart

Your AI agent never gets your Gmail. It gets a Shell: rules you write, which Clam checks before every request. This page takes you from invite to a connected agent in about 10 minutes.

You need a Gmail account, the AI app you want to connect, and an invite. We're in early access, so [request one](https://console.useclam.ai/get-early-access) if you don't have it yet. The 14-day trial needs no credit card.

## 1. Connect your inbox

Sign in at [console.useclam.ai](https://console.useclam.ai) with the Google account your invite went to, then click **Connect your inbox**. Clam uses this access to label your mail and to carry out what your rules allow. Your agent never sees it. [What Clam can see](https://useclam.ai/docs/trust/privacy-and-security)

## 2. Check Clam's labels

Clam reads up to 100 of your latest emails and labels them: work, receipts, travel, banking and so on. Your rules point at these labels, so check a few emails you know. If they're right, click **Looks good**. If not, click **Customize** to fix them, now or later. The labels also show up in Gmail, under `clam/`.

Customize is also where you add labels of your own, up to ten, for what only you know about your mail: `kids-school`, `squash-club`. Say what counts, and your rules can use them like any other label.

## 3. Create a Shell

Pick the template closest to the job you have in mind, from **Inbox Janitor** to **Travel Concierge** to **Work Assistant**, or start from scratch. Change its rules in your own words.

Not sure where to start? Copy this one.

Read and search my work email and receipts. Draft replies, but never send. Keep banking, medical, family and login codes off-limits.

Before saving, Clam shows you how it read your rules. Anything they don't allow is blocked. Then choose the Shell's **Transparency**, which sets what your agent learns about mail it can't open: nothing (Invisible), a count (Counted, the default), or the sender and time (Outlined). Try the Shell in the built-in chat before a real agent touches it.

Creating the Shell starts your trial, and Clam labels your newest 1,000 emails in the background. Want your agent to work only at certain times? Set **Office hours** on the Shell's Settings tab: outside them, its requests are refused and logged. For help with rules, read [Writing effective policies](https://useclam.ai/docs/policies/writing-effective-policies).

## 4. Connect your agent

Pick yours and follow its page.

Pick your agent and follow its page: Claude, Claude Code, ChatGPT, Codex, Hermes, Grok, OpenClaw, Gemini CLI or any other MCP client.

If it asks which Shell to use, pick the one you just made. Then click **Check connection** to confirm it reached Clam.

## 5. Tell your agent to use Clam

Your agent may have other ways into your mail, so tell it which one to use. Paste this in:

From now on, use Clam for all email. Don't use any other connection to my email. If Clam refuses something, tell me instead of trying another way.

If your agent keeps standing instructions, put it there so it outlasts this chat.

An instruction isn't a lock, and Clam can only check what goes through Clam. If your agent has another way into your mail, such as a Gmail connector, remove it.

Every request your agent makes now shows up in your mailbox's **Audit log**, with Clam's decision on each.

## Next steps

- [Write your own rules](https://useclam.ai/docs/policies/writing-effective-policies)
- [Have your agent ask before it acts](https://useclam.ai/docs/trust/approvals)
- [Cut an agent off](https://useclam.ai/docs/trust/remove-access)