> ## Documentation Index
> Fetch the complete documentation index at: https://docs.connect-crm.com/llms.txt
> Use this file to discover all available pages before exploring further.

# What your assistant can do

> The tools a connected AI client uses, and the difference between read-only and read-and-write access.

A connected AI client works through a small, fixed set of **tools**. You don't call these directly — you ask in plain language, and the assistant picks the right tool. Knowing what they are helps you understand exactly what an assistant can and can't do with your CRM.

Everything is scoped to the **one Workspace** you chose when connecting, and the assistant only ever sees data you can already see in the app.

## Reading your data

These tools are always available to a connected client. They never change anything.

| Tool                | What it does                                                                                                 |
| ------------------- | ------------------------------------------------------------------------------------------------------------ |
| **List objects**    | Lists the Workspace's Objects (e.g. People, Companies, Deals) so the assistant knows what it can work with.  |
| **List attributes** | Lists the fields on an Object (e.g. name, email, stage) so the assistant uses the right ones.                |
| **List options**    | Lists the choices for a dropdown field (e.g. a contact's type or tags) so the assistant picks a valid one.   |
| **List stages**     | Lists the stages of a pipeline (e.g. a deal's stages, in order) so the assistant can place a deal correctly. |
| **Get a record**    | Fetches a single Record and its current values.                                                              |
| **Search records**  | Returns a page of Records from an Object. Large result sets are paged automatically.                         |

So an assistant can answer questions like *"Which companies are tagged enterprise?"* or *"Show me the contact details for this person"* using only read access.

## Reading your conversations

A connected client can also read your inbox — the same conversations and
messages you see in the app, with your visibility.

| Tool                              | What it does                                                                                     |
| --------------------------------- | ------------------------------------------------------------------------------------------------ |
| **List conversations**            | Lists your conversations, optionally filtered to unread, starred, archived, or a single channel. |
| **Get a conversation**            | Opens one conversation and reads its full message thread.                                        |
| **List a record's conversations** | Lists the conversations a particular Record (a person or company) is part of.                    |

This lets an assistant answer questions like *"What's the latest from Acme
Corp?"* or *"Summarise my unread conversations."*

<Note>
  Conversations you've blocked never appear, and an assistant can only open the
  threads in your own inbox. When you list a Record's conversations it may show
  ones that belong to a teammate's inbox so you know they exist, but the
  assistant can't open or read those.
</Note>

## Changing your data

These tools are only available when you grant **read-and-write** access while connecting.

| Tool                | What it does                                                 |
| ------------------- | ------------------------------------------------------------ |
| **Create a record** | Adds a new Record to an Object with the values you describe. |
| **Update a record** | Changes the values on an existing Record.                    |
| **Delete a record** | Permanently removes a Record. This cannot be undone.         |

With write access, an assistant can act on requests like *"Add Jane Doe as a new lead"* or *"Move the Acme deal to the Negotiation stage"*.

<Note>
  Every create and update goes through exactly the same rules and validation as
  the app — required fields, unique constraints, and your automations all still
  apply. Changes an assistant makes appear in a Record's timeline just like a
  change you make by hand.
</Note>

<Warning>
  **Delete a record** is permanent — the Record is gone for good, just as if you
  deleted it in the app, and references to it from other Records are cleared. An
  assistant only has this ability when you grant read-and-write access.
</Warning>

## Drafting and sending messages

Messaging follows a **draft-then-send** model: an assistant writes a draft, and
sending it is always a separate, more carefully guarded step. A draft is never
sent on its own — it shows up in your inbox exactly like a draft you started
yourself, so you can review (and even add attachments) before anything goes out.

| Tool               | What it does                                                                       |
| ------------------ | ---------------------------------------------------------------------------------- |
| **Create a draft** | Composes a draft message — a reply in an existing conversation or a brand-new one. |
| **Update a draft** | Edits a draft that hasn't been sent yet.                                           |
| **Delete a draft** | Removes a draft.                                                                   |
| **Send a message** | Sends an existing draft.                                                           |

Drafting and editing need **read-and-write** access. **Sending** needs a
separate **send** permission on top of that — so an assistant you've allowed to
draft replies still can't send anything until you've explicitly let it.

<Note>
  When an assistant adds a recipient, it shows you whether each address matches
  a Record in your CRM, so you can spot a mistaken or made-up address before
  sending. Sending only reports a message was **accepted** for delivery — the
  assistant confirms it actually sent by re-reading the conversation.
</Note>

## How much access you grant

Access comes in three levels, each one a deliberate step up:

* **Read-only** — the assistant can look things up and read your conversations, but cannot change or send anything. A good default for an assistant you mainly use to ask questions.
* **Read-and-write** — the assistant can also create and update Records and compose drafts.
* **Send** — on top of read-and-write, the assistant can send drafts as messages.

An assistant is only ever given the level you've allowed: a read-only assistant that tries to write is refused, and an assistant that can draft but hasn't been given send permission can't send.

## What an assistant can never do

* **Delete Records without write access.** Deleting is part of read-and-write access; a read-only assistant can never delete a Record. (When you do grant write access, deletion is permanent — review the **Changing your data** section above.)
* **Send without permission.** Sending is a separate trust level, and even then only ever sends a draft you can review first.
* **Forward a thread or attach files when sending.** An assistant can't forward an existing conversation to a new address, and attachment content is never sent over an AI connection.
* **Reach another Workspace.** A connection is locked to the single Workspace you selected.
* **See more than you can.** The assistant acts as you, with your visibility — nothing more.

<Card title="Manage your connections" icon="shield-halved" href="/integrations/ai-clients/managing">
  Review which assistants are connected and revoke access at any time.
</Card>