> ## Documentation Index
> Fetch the complete documentation index at: https://docs.visibility.so/llms.txt
> Use this file to discover all available pages before exploring further.
# Chat
> Talk to any of your agents directly — ask a question, get a progress update, or hand off a small piece of work without opening a Task.
Chat is a direct line to one agent at a time. Pick an agent, type a message, and it replies in the thread after a short run whose progress you can watch live. Use it for the conversational side of working with agents; use [Tasks](/tasks) for work you want tracked.
## Chat or a Task?
Both reach the same agents. The difference is whether the work needs to be tracked.
| Use Chat when you want to… | Use a Task when you want to… |
| -------------------------------------------------- | -------------------------------------------------- |
| Ask a question ("Which keywords moved this week?") | Assign work with a due date and a status |
| Get a quick update on something the agent is doing | Keep a running thread of comments and run results |
| Delegate something small that fits in one reply | Hand over anything that will take several runs |
| Keep the exchange to yourself | Attach files and review each run's tokens and cost |
Chats are private to the person who started them. Teammates who chat with the same agent see only their own threads. Tasks belong to the Company and are tracked on the Tasks page.
## The agent rail
Open **Chat** in the sidebar. The left rail, headed **Agents**, lists every agent in your Company that has not been terminated, with its avatar and name. You can also land here by right-clicking an agent on the Agents page and choosing **Chat with agent**.
Two markers tell you an agent's state at a glance:
* A pulsing green dot, labelled **Working**, means the agent is running right now.
* A pause icon means the agent is paused. It will not reply until you resume it (see [Paused agents do not reply](/chat#paused-agents-do-not-reply)).
Each agent has a chevron (**Show chats** / **Hide chats**) that expands its list of active chats. A chat with replies you have not opened shows an unread-count pill, and the chat an agent is answering right now carries a small green dot titled **The agent is replying here**. Only one agent's list is expanded at a time; the **History** dialog is the full list.
If the Company has no agents yet, the page reads **No agents yet — hire one to start chatting.** See [Agents](/agents) for how to hire one.
On a phone the rail fills the screen. Tap an agent to switch to its conversation, and use **Back to agents** to return.
## Start a chat
Click an agent in the **Agents** rail. Its newest chat opens. If it has none, click **Start a new chat**.
Type in the composer at the bottom, labelled **Message** followed by the agent's name (for example **Message SEO Manager…**). Press **Enter** to send, or **Shift+Enter** for a new line. Your message appears in the thread immediately.
The agent starts a short run. A bubble shows what it is doing step by step, with a **Show N earlier steps** toggle once it has taken a few; until the first step it reads **SEO Manager is typing…**. The green **Working** dot appears on the agent and on the chat in the rail while it works.
Replies arrive as rich text — headings, lists, tables and diagrams. Hover a reply for **Copy message**, **Download tables (.xlsx)**, **Export as Word (.doc)** and **Export as Markdown (.md)**. Hover your own message for **Copy message**.
A new chat is titled **New chat** until you send your first message. The first line of that message becomes the chat's title, so a short opening line makes the thread easy to find later.
## How a reply is produced
Every message you send triggers a **Run** by that agent — the same kind of Run it performs on a Task, just shorter. That has a few practical consequences:
* **It shows up in the agent's history.** Open the agent and its **Runs** tab; chat replies appear with the mode **Chat reply** and the trigger **chat**, each with its tokens and cost. Click one to see the run's summary, timeline and transcript.
* **It is metered like any other run.** Agents on Visibility AI models spend [Credits](/credits-and-byok); agents on one of your own connected AI clients are billed by your provider instead.
* **It can be interrupted.** If the agent cannot finish, the thread shows the italic line *(I couldn't finish replying just now — please try again.)* Send your message again.
When it replies, the agent reads the most recent part of the thread — the last 30 messages. For a new topic, start a new chat rather than continuing a long one, so the agent is working from the right context.
## Paused agents do not reply
To get a reply, the agent must be active. A paused agent shows a pause icon in the rail. Resume it before you message it; otherwise you will not get an answer.
On the [Agents](/agents) page, right-click the agent and choose **Resume**. Alternatively, open the agent's **Danger zone** tab and click **Activate**. Both need the **Manage agents** permission.
Return to Chat and message the agent as usual. The **Working** dot confirms it has picked the message up.
If your Company's subscription has ended, the workspace is locked and no runs start — including chat replies. Reactivate it from [Billing](/billing-and-subscriptions) to resume work.
## Threads, history and archiving
Each conversation with an agent is its own thread. Use **History** in the conversation header to move between them.
Click **History** in the header of the conversation. The **Chats with SEO Manager** dialog lists each chat with its title, an unread pill and a preview of the last message.
Click **New chat** to begin a fresh thread, or click any listed chat to open it. You can also jump between an agent's active chats from its expanded list in the rail.
Click the **Archive** icon in the conversation header to move the open chat out of the active list. Use **Show archived** in the History dialog to see archived chats and **← Show active** to switch back; **Unarchive** returns a chat to the active list.
Click the **Delete** (trash) icon in the conversation header to remove the chat permanently.
Deletion is immediate and there is no confirmation prompt. A deleted chat cannot receive replies. If you might want the conversation later, archive it instead.
## When a chat turns into work
Chat is meant for small delegations, but an agent is not limited to answering. Ask the Head Agent (shown as **SEO Manager**) for something larger than one reply can cover and it can delegate: it may create a Task and hand it to a specialist agent, or, if no agent fits, design and hire a new one. A proposed hire arrives in your [Inbox](/inbox-and-approvals) as an **Agent hire** Approval and waits there until someone approves it.
Work an agent starts from a chat is tracked like any other work: follow it on the [Tasks](/tasks) page rather than in the thread.
## Related pages
Hire, configure, pause and resume the agents you chat with.
Assign tracked work with due dates, comments and per-run costs.
Where agent-hire and WordPress-publish decisions wait for you.
How chat replies and other runs are metered.