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

# Attensira in ChatGPT

> Connect Attensira to ChatGPT and clear what needs you from the Today card: Approve or Decline each waiting change without leaving the chat.

When Attensira is connected to ChatGPT, ChatGPT shows your day as a card. The card is called **Today**. It shows what is waiting for you, and you can approve or decline each waiting change from the card. Connect it the way [Connect your assistant](https://docs.attensira.com/mcp/connect.md#chatgpt) describes, with the signed-in server at `https://mcp.attensira.com/mcp`.

This is not the free audit.

## The Today card

Ask ChatGPT something like "What needs me in Attensira today?". It calls [`brief`](https://docs.attensira.com/mcp/tools#brief) and shows the result as the Today card:

* **How many things need you.** If nothing is waiting, the card says "Nothing needs you today."
* **The first three waiting items**, by title. Each **change** has **Approve** and **Decline**. A **question** has no buttons; answer it in the chat.
* **Whether the agents are running.** If they are not, the card gives the reason. It also shows how many wins you have not opened yet.
* **A link to your Inbox** in the Attensira app, with how many more items are waiting there.

## Approve and Decline

Each button makes one [`act`](https://docs.attensira.com/mcp/tools#act) call through ChatGPT. The call uses your own connection, so the button can do exactly what asking ChatGPT to approve the change can do, and nothing more.

* **Approve** approves the change. The card then shows "Approved", or "Approved. It is queued to ship." when the change ships later.
* **Decline** asks for a reason first: **Not now**, **Already done** or **Never touch this**. The agents learn from the reason. To decline a change because it is wrong, say so in the chat. A "wrong" reason needs a note, and the chat is where you write it.
* Each item and action uses one idempotency key. If you tap twice, or ChatGPT retries, the change is still acted on only once.

### When the card cannot approve

A ChatGPT connection approves up to reversible work in any lane (`approve:*:reversible`). It never approves irreversible work or work that a connection started. The rules are in [Who can approve what](https://docs.attensira.com/agent/approval.md#who-can-approve-what). When a change is above that limit, the card shows Attensira's answer as it was sent. The answer says what kind of work the change is and where a person approves it: in the app, at the item's Inbox link. The item stays open in your Inbox. The card removes **Approve** and keeps **Decline**, because a connection that cannot approve a change can still decline it.

## Without the card

The card only displays the answer. It does not change it. Claude, Cursor, scripts and any other client that does not show cards get the same `brief` answer: the same text and the same `structuredContent`. Over REST, the same answer is [`GET /v1/brief`](https://docs.attensira.com/api/endpoints#get-/v1/brief), and approving or declining is [`POST /v1/actions`](https://docs.attensira.com/api/endpoints#post-/v1/actions).

## What the card can reach

The card fetches nothing from the network and loads nothing from other sites. It writes every value as plain text. Its only link goes to `https://app.attensira.com`, and ChatGPT opens that link.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.