> ## 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 and rooms

> Talk to your agents in a private direct message or in rooms with your team: what agents and rooms are, your starter agent Assistant, and where Connie is now.

> [!NOTE]
>
> Every account has agents and rooms: the sidebar shows an **Agents** area and no **Ask Connie** button. Standalone AI Chat and its Earlier chats archive are retired. Signed-out visitors start at the homepage and create an account before meeting their Chief of Staff.

In Connie, the work happens with **agents**: AI teammates that each have a name, a purpose you give them in conversation, an owner, and the owner's connected accounts. You talk to an agent in two places:

* **Its direct message (DM).** Each person has one private conversation with each agent. Only you can read your DM with an agent.
* **Rooms.** A room is a conversation with people and agents. It has its own members and its own history. You can have several rooms with the same agent, and each one keeps its own history.

Messages show their recorded send date and time in your local timezone. Hover over a timestamp to see the full date, time and timezone.

When you create a room, each active agent introduces itself in turn, with its own wording and role where that information is visible to everyone in the room. An agent added later also introduces itself once. Adding people or reopening the room does not repeat the greetings, and introductions do not start any work.

Your private agent conversation has an address like `/<workspace>/agent/<conversation>`. A room uses `/<workspace>/room/<conversation>`, including a room with just you and one agent. The conversation ID stays the same. Older `/rooms/` links and chat links still open the same conversation when you have access; copying a link does not grant access.

## Your starter agent, Assistant

A new conversation’s opening greeting appears at the top of the message column and writes out once on its first view. Reopening or reloading shows the saved greeting in full. Reduced-motion preferences skip the animation. In a new agent chat, the agent greets you by first name when known, introduces itself, offers one concrete first piece of work that fits its role and your business (from your workspace's business profile when there is one) and asks whether to start there, with 2–5 suggested options and a **Something else** text field. Click an option to send it immediately, or enter your own request in **Something else** and tap **Submit** or press Enter. **Submit** appears when you type an answer and disappears if you clear it. Suggestions draw on your onboarding answers and, with Memory on, your saved preferences, readable past requests and agent memory. The agent does not list or mention the tools it can use; it may only suggest work they make possible. Sending your first request removes these welcome choices; the saved greeting remains in the chat.

If an existing agent chat never received its greeting and has no message history, loading the sidebar recovers that welcome. Its saved message then appears as the agent’s sidebar preview. This also covers empty chats created before greetings were introduced. Chats with earlier messages, including deleted messages, do not receive another welcome. An agent you have never opened still starts its chat only when you open it.

Your starter agent is **Assistant**. Connie makes it for you the first time you open your workspace once the workspace has [free credits](https://docs.connie.ai/billing/free-credits.md), a paid plan (or an active trial started before November 2026), existing credit-only access, or **Beta / credit-only** or **Internal** access. Each member gets their own; guests get none. It's yours: you own it, it works on your connected accounts, and you can rename it by clicking its name in its details (click its name at the top of your DM with it to open its details, or choose **Rename agent** in its sidebar menu). It isn't listed in anyone else's sidebar, and no one else can talk to it, workspace owners and admins included, unless you share it with them; see [Who can talk to an agent](https://docs.connie.ai/agents/agents-and-permissions.md#who-can-talk-to-an-agent).

**Chief of Staff** is a different agent, not your starter agent. New workspaces meet it during onboarding. It remembers its coordination role. It researches your website or business description in the background, saves it as your workspace's [business profile](https://docs.connie.ai/agents/memory-and-privacy.md#what-every-agent-knows-about-your-business) so every agent knows what your company is, and helps you hire **Outbound SDR**, **Media Buyer**, **Ghostwriter**, **SEO Expert**, **ABM Marketer**, **Account Manager**, or an employee for your own described job. Each hired employee prepares a Page about its job. The Chief creates no onboarding document. Click **Message from \[agent]** above a published reply to open your chat with that employee, including while onboarding is still in progress. New workspaces run on their 1,000 [free credits](https://docs.connie.ai/billing/free-credits.md); your employees and their Pages stay available, and you get **Assistant** as well.

Opening [connie.ai](https://connie.ai) or your workspace's home page opens your last opened agent in this browser. If none is remembered, or it is no longer available to you, the first available agent in your sidebar opens. This preference is separate for each account and workspace. A prompt you typed on our website or in the setup card, an **Ask Connie** button clicked before the sidebar shows your agents still open your starter agent. A prompt you brought with you waits in the message box; it is never sent for you. A link that used to open a page with Connie beside it now opens just the page. Your DM with Assistant is created the first time you open it, not before.

In a workspace with no plan or usable credits, there's no Assistant to open: the sidebar lists none, and opening it from any of the ways above shows **Choose a plan to use your agents.** instead. Owners and admins can choose a plan; members can ask an admin. When a workspace has used its free credits, every agent's message box is replaced by **Choose a plan to keep going**. See [Free credits](https://docs.connie.ai/billing/free-credits.md).

A page’s **What do you want to build?** creation prompt chooses a suitable existing agent using your request and your readable DM context. Your words wait in that agent’s composer, unsent.

## The sidebar

The expanded sidebar’s Connie header has 24px of padding on each side. The **Search...** field has 10px of padding on the left and right.

Navigation rows such as **Inbox**, **Artifacts** and **Plugins** have 11px of padding on the left and right.
The workspace card at the bottom of the sidebar also has 11px of padding on each side. The agents and rooms list uses a thin scrollbar when its contents overflow. Scroll the list to reach more agents and rooms; there are no up-arrow or down-arrow buttons.

| Item | What it does |
| - | - |
| **Search...** | Finds your work, including the rooms and agent DMs you're in and rooms open to the workspace, by name or, from 3 characters, by what their messages say. An agent result opens your DM with it once you have one. Press ⌘K on Mac or Ctrl+K on Windows. |
| **Notifications** (bell beside your workspace at the bottom) | Approvals and updates waiting for you. See [Notifications](https://docs.connie.ai/notifications/overview.md). |
| **Tasks** | Your tasks and your agents' work. A red count shows how many things wait on you. See [Tasks](https://docs.connie.ai/tasks/overview.md). |
| **Inbox** | Email, LinkedIn and other conversations, **Meetings** and **Calls**. See [Inbox](https://docs.connie.ai/inbox/overview.md). |
| **Artifacts** | Your Records Lists, Workflows, Pages and Apps in one place. Each opens at its own page, with its own access. |
| **Plugins** | **Tools**, **Skills**, **Connectors** and **Connect my agent**. See [Tools](https://docs.connie.ai/tools/overview.md). |
| **Agents** | Your agents and rooms. An **agent row** shows the agent's face and opens your DM with it. A **room row** shows your picture with the room's agents and opens that room. Click **+** (**New agent or room**) for **New agent** and **New room**. While an agent is working on something you can see, its face uses the same animation as the chat, including in room pictures. The preview shows its current public activity, such as **Thinking**, **Searching the web** or **Writing**, while it works. In the collapsed sidebar, hover or focus the face to see that activity. Reduced-motion settings keep the animation still; see [In progress](https://docs.connie.ai/agents/tasks-and-work.md#in-progress). |

When the sidebar is collapsed, and on a phone, each agent and room shows as its face only: hover a face, or move to it with the keyboard, to see its name; on a touch screen, touch and hold the face (a quick tap opens it).

Your personal sections keep their saved order. Rows outside sections can be dragged within **Unassigned**, or directly in the Agents list before you create any sections.

Agent positions stay fixed when you send or receive messages, rename an agent or pause it. Drag an agent row to change its position; your order is saved for you in this workspace and also appears in the collapsed sidebar and on your other devices. With a row focused, press **Space**, use the arrow keys to move it, then press **Space** to drop it or **Escape** to cancel. On a touch screen, hold the row briefly before dragging. New entries appear above existing unpinned entries; pinned rooms stay first.

Room pictures overlap without borders around the people, agent faces or the **+N** count for additional participants.

Each row shows the agent or room name and a one-line preview of the latest message. An agent's Records statistics appear as a readable summary, such as **291 Records · 21 Lists**, instead of result JSON, including in **Archived rooms**. Opening the conversation still shows the original message; JSON you send remains unchanged. Unread messages have an 8 × 8px blue dot in the expanded sidebar; the unread count is also announced to screen readers. The selected row has a pale gray background. Your unread counts are the same on every device. A deleted agent leaves the list. A paused agent stays in the list with the latest message preview, without a pause label or tag in the sidebar. Its DM header shows **Paused** beside its name. Archived rooms move under **Archived rooms** at the end of the list, marked **Archived**, where you can still open them.

Deleting the agent you have open takes you to the next available agent and shows **\[Agent name] successfully deleted**. When no agents remain, you get a new private **Chief of Staff**. See [Delete an agent](https://docs.connie.ai/agents/agents-and-permissions.md#delete-an-agent).

If you select another agent or room while an earlier one is opening, Connie follows your latest selection. Returning to a recently opened conversation can show its last loaded messages immediately while Connie checks for updates and access. If you no longer have access, those messages are cleared and the conversation shows **Access restricted**.

The sidebar catches up when the connection returns and also checks periodically while the tab is visible. You do not need to reload after a brief connection interruption.

Click a row to open it. A three-dot options button appears when you hover or focus the row. Click that button to open its menu; hovering over it doesn't open it. With the keyboard, Tab to the button and press Enter or Space; on a touch screen, tap it. You can also right-click an agent row, or focus it and press Shift+F10 or the Menu key, to open the same dropdown without opening its conversation. Right-clicking a room row and touch-and-holding a row do not open the options menu. Closing the dropdown returns keyboard focus to its three-dot button.

* On a room: **Members**, where people and agents are added; and, if you are a non-guest participant, **Rename room**, **Duplicate** (a new room with the same saved room guidance, the same people with equal room controls and the agents you can use, starting with an empty conversation) and **Delete room**. A muted room's count turns gray. Delete removes the room for everyone in it; its history is kept but nobody can open it.
* On an agent: **Rooms**, which holds **New room with \[agent]** and your rooms with it, **Manage accounts** and **Add people**. If you own the agent: **Rename agent**, **Share agent**, **Transfer ownership** (it opens a compact dialog with a **New owner** dropdown, **Cancel** and **Transfer ownership**), **Pause agent** (or **Resume agent**) and **Delete agent**. During onboarding, **Delete agent** is disabled. Hover over it to see “Finish onboarding before deleting an agent.” It becomes available to the owner after onboarding is complete. An agent listed only for your DM with it after it stopped being shared with you offers no **New room with \[agent]**.

In the agents-and-rooms sidebar, a copy appears with **Duplicating…** while the server saves it. The pending copy cannot be opened or changed. Once saved, the same entry shows the saved name and can be opened without waiting for the next sidebar refresh. If saving fails, the pending entry disappears and the original stays unchanged.

Agent replies appear as plain text in sidebar previews: heading markers, bold markers and Markdown link syntax are removed, while real hashtags such as **#gaming** remain. This also applies to saved and archived previews. The full conversation keeps its formatting and original message content; text you send stays literal in the preview.

## Connie, outside the app

Once you've signed up, Connie has no chat inside the app. Connie is the assistant you reach from outside it, in Slack and in the AI apps you connect through MCP, such as Claude and ChatGPT. There, Connie works on your own connected accounts and with your own permissions. Set it up with **Connect my agent**. See [Connie in your AI apps](https://docs.connie.ai/mcp/overview.md).

In Slack, Connie also brings your agents in: **@Connie ask Chief to …** and Chief answers in the thread through the Connie app, with its name in bold at the start of each message. See [Agents in Slack](https://docs.connie.ai/agents/agents-in-slack.md).

## Old links

* Old standalone chat entries open an agent. Links to current rooms and agent DMs keep opening their conversations.
* The old Agents page (`https://connie.ai/<workspace-id>/agents`) opens your DM with Assistant. A link to one agent offers **Message \[agent]** and **Details**; a link to one of its conversations opens that conversation.
* The old Projects list (`https://connie.ai/<workspace-id>/projects`) opens **Artifacts**. A link to one project still opens that project.
* **Duplicate** on a shared page, file or app puts your copy in your DM with Assistant instead of a new chat, and opens it there, beside the conversation.

## Learn more

> [card] [Rooms and people](https://docs.connie.ai/agents/rooms-and-people.md)
>
> Create rooms, add people, mentions and who answers.

> [card] [Agents and permissions](https://docs.connie.ai/agents/agents-and-permissions.md)
>
> Create agents, the owner's accounts, what an agent can use, its model, responsibilities, pause, duplicate, transfer and delete.

> [card] [Work in progress](https://docs.connie.ai/agents/tasks-and-work.md)
>
> What an agent is working on, In progress and a conversation's Tasks, follow-ups, notifications, stopping work and approvals in rooms.

> [card] [Memory and privacy](https://docs.connie.ai/agents/memory-and-privacy.md)
>
> What an agent may use in which conversation.

> [card] [Agents in Slack](https://docs.connie.ai/agents/agents-in-slack.md)
>
> Ask your agents from Slack, approve their actions there, and stop them.

## Related

* [Find your way around Connie](https://docs.connie.ai/navigating-connie.md)
* [Glossary](https://docs.connie.ai/glossary.md)
* [Chat](https://docs.connie.ai/chat/overview.md) (current conversation entry points)

Agent DMs keep these actions in the sidebar menu rather than the chat header. Click the agent’s three-dot button in the sidebar (or Tab to it and press **Enter** or **Space**) for its options: **Manage accounts** opens the accounts it may use, and **Add people** starts a separate room with that agent. Click the agent's name at the top of the conversation to open **Chat details**, the right-side panel with its cover, avatar, members, **In progress**, **Tasks** (where your workspace has the [Tasks](https://docs.connie.ai/tasks/overview.md) page), **Routines**, **Artifacts** and **Context**. **In progress** is hidden when empty; **Routines** keeps saved schedules and event triggers with their enable/disable switches. It opens even when the agent has no work, tasks or artifacts yet.

Active agent avatars gently glance up, down, and diagonally, with pauses between glances. Their full silhouette stays visible without circular cropping. Paused and deleted agents stay still, and reduced-motion settings disable these animations.

Agent avatar outlines follow the agent’s silhouette, including in room participant badges.

Idle agents do not blink. Blinking indicates active work, an unread message, or an unsent message in that agent’s composer.

Hover over an agent in the sidebar to have its eyes follow your pointer. They resume looking around when the pointer leaves. Reduced-motion settings disable pointer tracking.

Generated artifact cards appear outside the agent’s response bubble in agent DMs and rooms.

List lookups appear in a collapsed **Looked at N Lists** summary instead of preview cards. Expand it to see which Lists the agent read, with their names when available, and open a List beside the conversation. If its name cannot be read, it stays labeled **List**. A List the agent creates or changes appears once per task, using its latest confirmed result. You can say **no List cards, just one link** to keep that task's List results as links, including after an approval or continuation. Quotes and approvals that need your decision remain available.

Agent-to-agent exchanges appear as a centered **X messages with \[agent]** summary in agent DMs and rooms. Click an agent’s name to read the attributed exchange, then close it to return to your draft. Internal handoffs and replies stay in that transcript, without extra **Sent by** messages or redundant **Messaged** receipts. A confirmed send without a readable exchange summary, and a deliberate message posted to a room, retain their delivery receipt. Direct answers to your requests, deliberate messages posted to a room, and questions or approvals that need your decision stay in the conversation. Transcript access is checked again whenever you open or refresh it.

While a card’s preview loads, it shows a loading skeleton and **Open** stays available. An empty page shows its title and **No content yet** in the preview until content is saved. Loading previews appear immediately, with **Open** still usable.

In **Artifacts**, choose **New page**, **New list**, **New workflow** or **New app** from **New**. Connie chooses a suitable existing agent and opens your private DM with it, with the artifact preview beside the chat. The choice considers the artifact type, the agent’s name and purpose, and a few recent messages from your own readable DM with it. A creation prompt also supplies your request. If the choice is uncertain or unavailable, Connie uses your existing starter agent or another active agent you can use. No new agent or shared room is created. Review and send the prepared request, then describe what you want to build. A new workflow starts as a blank workflow you can edit in the side panel; the other previews are replaced when their artifact is created. Drag the divider to resize the chat. If no active agent is available, create one from the sidebar or ask its owner to share one, then retry.

Artifact creation chats restore their empty preview when reopened from the sidebar in the same browser. Closing the preview hides it until you reopen that chat; clicking its current sidebar entry also restores it. This applies to pages, lists, workflows and apps.

Hovering or focusing a known chat in the sidebar prepares its route without creating a conversation. Remembered artifact creation previews stay visible while the chat loads; conversation access is still checked before messages appear.

Choosing a new artifact immediately opens its chat skeleton and matching preview while Connie selects an existing agent and opens your DM with it. If creation fails, the directory returns with a retryable error.

Reactions on your sent messages overlap the bubble's bottom edge by 9px. Spacing to the next message starts below the final reaction row.

After you answer a question, its original card keeps the question and your selected answer without the unused choices or a generic introduction. Use Back and Next to review saved answers when the card contains several questions. The Thinking indicator sits 12px after the latest message.
Question cards fill their available wrapper within the conversation's shared width cap, including short questions and cards waiting for someone else. Submitting preserves that width and styling. The answered card stays outside the message bubble, without an extra border or padding around it. The current design has no file-upload button inside a question card. Files already shown on a card can be removed and otherwise go with its answer; files in the message box belong to your next message. Unselected choices and their empty space disappear; multiple questions retain progress dots and Back/Next controls. Any additional agent reply appears in a separate consecutive message bubble.

Tool approval buttons use the shared black primary appearance. Workspace-wide grants read **Always allow**; one-call, session and delete approvals read **Allow**. Delete approvals require a second confirmation before the action runs; other changes run as soon as you choose **Allow**.

Tool approval cards omit Canvas and Title metadata lines, including in saved chats. Their Allow/Reject controls and permission behavior are unchanged.

Selected question-card badges match the agent’s color, with white letters. Cards without an agent color use the default blue.

Answered question cards keep the question and selected answer visible in a read-only card. For multiple questions, use the dots and Back/Next buttons to review the recorded answers.

Incoming and outgoing messages and cards in agent chats and rooms have a maximum width of 470px on screens 1440–1769px wide, and 580px on screens 1770px and wider. They shrink to fit narrower panels.

Chat composers expand to 810px on screens 1440–1769px wide and 1000px on screens 1770px and wider, while fitting within open chat panels.

Drag files anywhere over a chat you can post in to see the blue **Drag and drop to attach** area. Dropping adds them to your message box, ready for you to review and send. It does not send a message or answer an open question. Moving outside the chat or cancelling the drag dismisses the highlight.

A failed attachment keeps sending disabled until you remove it or retry it successfully. Its chip explains the error and next step, such as **Remove empty.csv before sending.** If **Retry** is available, it says **Retry or remove archive.zip before sending.** Remove each failed file with its × button, or use **Retry** where offered. Your message text and valid attachments stay in place; Connie does not silently skip a failed file. For an empty file, attach a copy with content if you still want to send it.

For a single question with one selectable answer, choosing an option sends it immediately. Type a custom answer and press Enter. Multi-select and multi-question cards retain their submission controls.

### Review message drafts

Agent email, Slack and LinkedIn drafts appear as editable cards in the chat.
You can compose a draft without connecting an account; a connected account is
required only when you ask the agent to send it or save it with that provider.
Review the sender, recipient or channel and message; email cards also include
subject and copy recipients. Draft edits are kept on your device for your
signed-in account and that individual draft. Switching accounts or opening a
different draft loads its own edits or the published draft. Older browser
edits without an account identity are not restored. **Discard** removes the
local draft. **Send** asks the agent to
send the edited draft through the normal connected-account and approval flow;
a request is not a delivery confirmation. Saved tool previews remain read-only.

Draft cards fill the incoming-message width without an outer message bubble.
Email, Slack and LinkedIn draft cards offer **Discard** and **Send**, without a
**Needs approval** menu. Manage tool permissions in **Plugins**; sending still
uses the normal approval flow.


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