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

# Agents in Slack

> Ask your agents for work from Slack through the Connie app: how to address an agent, who can ask and approve, what an agent shares in a channel, how to stop it, and where its work shows in Connie.

> [!NOTE]
>
> Agents in Slack need the Connie app installed in your Slack workspace and your own Slack account connected to Connie. See [Connect your accounts](https://docs.connie.ai/tools/connect-accounts.md#slack).

When choosing an account or channel for an agent to watch, search ignores letter case and surrounding spaces, including private-channel labels.

Your agents answer in Slack through the **Connie** app. You don't install an app per agent: one Connie app serves every agent in your workspace. Slack shows each agent's reply as a message from the Connie app, starting with the agent's name in bold, for example **Chief:**, so you can tell which agent wrote it.

## Ask an agent

Mention Connie and name the agent:

> @Connie ask Chief to draft the renewal email for Acme

Chief answers in the same thread, through the Connie app, with **Chief:** at the start of its message. You can also:

* Start a message in a thread with the agent's name: **Chief, make it shorter**.
* Reply right under the agent's message. A plain follow-up in a thread where an agent is working goes to that agent when it's meant for it; people talking to each other get no reply.
* Ask two agents at once: **@Connie ask Chief and Researcher for their views**. Each answers separately.
* Write to the Connie app in its **Messages** tab: **ask Chief what's on my plate today**. The answer stays private to you.

If you mention Connie without naming an agent, Connie answers as before, on your own accounts. In a channel it follows the same rule as an agent (below): it doesn't bring your private email, calendar or private Slack messages, what was said or learned in your Connie rooms (even rooms open to the whole workspace), your workspace's Records, Lists or Workflows, or personal memory into the channel.

While an agent works, Slack shows **Chief is working…** under the thread. It clears when the agent answers, stops or fails. If it couldn't start because of a problem on our side, it says in the thread "I could not start on this because of a problem on our side. Try again in a moment." with a reference to give support; if it was refused for a reason you can act on, it says that reason instead.

## Who can ask

Only people whose own Slack account is connected to a member of this Connie workspace can ask an agent. Anyone else who names an agent gets a private **Finish connecting in Connie** link, and nothing runs. Slack guests can't ask agents, and people from other organizations in a shared (Slack Connect) channel get no answer.

An agent works for the person who asked, with the agent owner's connected accounts (minus any the owner excluded). Asking someone's agent never gives you their permissions. See [Agents and permissions](https://docs.connie.ai/agents/agents-and-permissions.md).

## What an agent shares in a channel

Everyone in a channel can read what an agent posts there, including people who aren't in your Connie workspace. So in a channel, an agent uses only what's in the thread, public information and what you ask it to share. It never brings in its owner's private email, calendar or private Slack messages, what was said or learned in your Connie rooms (even rooms open to the whole workspace), your workspace's Records, Lists or Workflows, or anyone's personal memory. If you ask for that in a channel, it says it can't share it there and offers to send it privately in Connie instead.

In your direct messages with the Connie app, only you read the answer.

An agent can still act for you in a channel (send, create, update) when you ask, with approval when the action needs it.

## Approvals in Slack

When an agent wants to do something that needs approval, its reply shows **Approve** and **Decline**. Anyone in that channel whose Slack account is connected to a member of this Connie workspace can decide, including for an action that uses the agent owner's account. Connie checks, at the moment of the click, that the person is in the channel, isn't a Slack guest and isn't from another organization.

The decision is credited to the person who clicked, and the buttons change to **Approved by \[name].** (or **Declined by \[name].**). Anyone who clicks after that sees **Already approved by \[name].** The same approval shows under **Notifications** in Connie for the people it was offered to, labeled **\[agent] in Slack**; deciding it there decides it in Slack too, and the thread says so.

When an agent needs an account connected rather than an approval, its reply shows a **Connect \[App]** button for each app it needs, or **Connect in Connie** when it doesn't name the app; each opens **Tools** in Connie. There's nothing to approve or decline for it, in Slack or under **Notifications**. The reply ends "Connie carries on in this thread once it is connected. If nothing happens, reply here." When connecting can't carry the request on by itself, for example when the reply doesn't name the app or the app is an MCP server, the reply ends "Once it is connected, reply here and I'll carry on." instead: connect the account, then reply in the thread.

## Stop an agent

* Click Slack's **Stop** control on the thread.
* Or tell the agent in the thread, for example **Chief, stop**.

The agent replies **Stopped.** and posts nothing more for that request. Only the person who asked and the agent's owner can stop it; anyone else is told privately that they can't. Stopping can't undo anything already done, such as an email already sent.

## Your Slack work in Connie

Connie doesn't copy Slack threads. While an agent works on a Slack request, the request shows in the agent's profile under **In progress**, titled by where it was asked (for example **Request in Slack #renewals**), with a Slack mark that opens the thread and the time the agent has worked on it. It leaves the list when it's done; its answer stays in Slack.

If Connie is removed from your Slack workspace or from the channel, or its access is revoked, the agent's Slack work there stops and the person who asked is told in Connie: **\[agent] stopped your Slack request**.

## Good to know

* **Credits.** An agent's work from Slack uses your workspace's credits, like any other agent work. See [Credits](https://docs.connie.ai/billing/credits.md).
* **Names in Slack.** Every agent message comes from the Connie app, with Connie's name and picture, and starts with the agent's name in bold. Slack doesn't show the agent's own name and face as the poster: the Connie app doesn't have Slack's permission to post under another name.

## Related

* [Agents and rooms](https://docs.connie.ai/agents/overview.md)
* [Work in progress](https://docs.connie.ai/agents/tasks-and-work.md)
* [Connect your accounts: Slack](https://docs.connie.ai/tools/connect-accounts.md#slack)


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