---
title: "What your agent can do"
description: "Your agent gets 12 Gmail tools plus one for approvals. Clam checks every call against your Shell's rules before it runs, so available never means allowed."
route: /docs/trust/what-your-agent-can-do
---

# What your agent can do

Your agent reaches Gmail through a Shell, which offers 12 email tools plus one for approvals. Clam checks every call against the Shell's rules before it runs, so a tool being available never means a request is allowed.

## One Shell per connection

Each agent you connect gets its own Shell, with its own address, rules and activity. A rule on one Shell does nothing on another. Pausing or deleting a Shell cuts its agent off: see [Remove access](https://useclam.ai/docs/trust/remove-access).

## The tools

These are all of them. There is no tool to change a Shell's rules, manage labels, empty the trash or reach anything outside Gmail.

| Tool | What it does |
|------|--------------|
| Search | Finds messages matching a query. Messages your rules block are left out. |
| Read | Opens one message. |
| Create draft | Saves a new email, or a reply, as a draft. Nothing is sent. |
| Send | Sends a new email. |
| Reply | Replies to an existing thread. |
| Reply all | Replies to everyone on the original To and Cc lines. |
| Forward | Forwards a message, with its attachments and inline images, as a new thread. |
| Archive | Takes a message out of the inbox without deleting it. |
| Delete | Moves a message to the trash. Gmail empties the trash after 30 days. |
| Restore | Puts a message back in the inbox, undoing an archive or a delete. |
| Mark read | Marks a message as read. |
| Mark unread | Marks a message as unread. |
| Redeem approval | Collects the result of an action you approved. The agent calls it on its own; see [Approvals](https://useclam.ai/docs/trust/approvals). |

Only Send, Reply, Reply all and Forward put mail in someone else's inbox. Delete is the only tool that can lose a message, and only once Gmail empties the trash.

## Available is not allowed

Your agent sees every tool. Your Shell's rules decide what it may do with them: each rule says Allow, Ask me before, or Deny.

- Anything no rule allows is blocked. A new Shell with no rules blocks everything.
- When rules conflict, a block beats ask me before, and ask me before beats allow.
- Mail Clam hasn't labelled yet is blocked whatever your rules say. The agent is told to try again shortly.

How rules match an email is on [How rules work](https://useclam.ai/docs/policies/how-rules-work).

## What your agent sees when blocked

A blocked call answers "Action not allowed" and nothing else: not what the message said, not who sent it. The agent knows your Shell's rules in plain English, which Clam hands over when it connects, and is told a block is Clam working as intended, not an error. So it should tell you what it couldn't do and why.

How blocked *mail* shows up is the Shell's Transparency setting, on its **Settings** tab:

- **Invisible**: blocked mail doesn't exist, as far as your agent knows. It may draw the wrong conclusions from the gap.
- **Counted** (the default): a search says how many messages were held back, nothing more.
- **Outlined**: your agent sees the sender and time of blocked messages. Subject, body and attachments stay sealed.

Outside the Shell's office hours, a call is answered with the time the Shell reopens. That is a separate refusal, not a blocked rule.

Every call, allowed or blocked, is recorded. The Shell's Overview summarises recent activity, and **View in audit log** opens the mailbox's **Audit log** with each call.

## What Clam does not cover

Clam governs what goes through Clam. If your agent has another way into your mail, or other connections you authorised separately (another Gmail connector, Drive, a browser), we can't see or stop what happens there. Connect only the services you mean your agent to use, and review their permissions on their own.

Next: [have your agent ask before it acts](https://useclam.ai/docs/trust/approvals).