--- title: How to add the Agent to chat url: https://tideswell.xyz/docs/analyst/add-the-assistant-to-chat description: Add the Agent to your team's chat so anyone can write @Agent in a channel and get an answer only they can read. --- > Documentation index: fetch https://tideswell.xyz/llms.txt to discover every page before exploring further. # How to add the Agent to chat The **Agent** can join your team's chat as a member called Agent. An admin adds it from **Brain › Automations**, on the **Agent** card, and from then on anyone can write @Agent in a channel to ask it a question. See for the Agent itself. In chat it answers questions about sales, stock, demand, suppliers, purchase orders, marketing and decisions, as in . Each answer is written from what the person asking is allowed to see. ## Before you start [#before-you-start] You need **Manage automations** on **Settings › Permissions**, which admins hold by default. The **Brain › Automations** row only appears to people who hold it, per . You also need **See what the Brain has noticed**. Without it, the page shows "Brain access is needed to manage the chat agent." The **Automations** and **Brain** modules must both be on, on **Settings › Modules**, per . ## How do you add the Agent to chat? [#how-do-you-add-the-agent-to-chat] 1. Open **Brain › Automations**. 2. Find the **Agent** card, near the top of the page below the line that says whether everything is running. It reads "A teammate in chat. It answers when a message names it with @Agent. It only ever sees what the person asking is allowed to see, so two people asking the same question can get different answers." 3. Press **Add the agent**. The card's pill now reads **Off**. 4. Press **Switch on**. The pill now reads **On**. Switching it on also starts its background check, which runs every five minutes. Once that check has run, the card shows a line such as "Last looked for new work 3m ago: Nothing waiting." If that check is ever switched off while the Agent is on, the card warns "Its background check is switched off. New @Agent mentions still start immediately." Press **Switch the check back on**. ## How do you ask it a question in chat? [#how-do-you-ask-it-a-question-in-chat] Write **@Agent** in a message, pick Agent from the list that opens, add your question and send it. The Agent starts on it as soon as you send, and replies to your message in the same channel or thread where you asked. Mentions work the same way as for any colleague, per . For example: "@Agent which purchase orders from Portside Mills are late?" ## Why does the reply in a channel hold a link instead of the answer? [#why-does-the-reply-in-a-channel-hold-a-link-instead-of-the-answer] Because the answer is written from what you are allowed to see, and everyone in the channel can read the reply. So the reply says it has answered and that the answer is on a page only you can open. Press the link and the answer opens in your Agent panel. Anyone else who presses it is told "This answer isn't yours to read" and asked to put the same question themselves, so each person gets an answer from their own permissions. In a direct message with Agent, write @Agent the same way. You are the only reader there, so the answer arrives in the conversation itself. ## Which channels does it answer in? [#which-channels-does-it-answer-in] Every channel the Agent can see. To use it in a private channel, add Agent to that channel as a member first, the same way you add a colleague, per . ## How do you choose when it answers? [#how-do-you-choose-when-it-answers] Use **When it answers** on the **Agent** card. **When named** is the setting for chat. Its hint reads "It answers a message that names it with @Agent, in a channel or a direct message. A message that doesn't name it gets no answer." **Never** keeps it in the workspace but silent. Its hint reads "Silent, without deleting it. Its past messages stay where they are." ## Can it check in on late tasks? [#can-it-check-in-on-late-tasks] Yes. The **Check in on late tasks** switch on the **Agent** card has it ask each person about their late tasks each working morning, in their Agent Inbox or a direct message. It starts off. See . ## Can it follow up on what the background jobs find? [#can-it-follow-up-on-what-the-background-jobs-find] Yes, with **Follow up on saved findings**, the first setting on the **Agent** card. Its hint reads "Read results from existing background jobs and ask what to do in Agent Inbox. Financial, stock and external actions keep their approval steps." 1. Turn on the **Follow up on saved findings** switch. 2. Under **Default recipients**, tick the people whose Agent Inbox the findings go to, up to 50. Each needs access to what the finding is about. 3. Press **Save follow-ups**. It covers low stock, late purchase orders, unowned overdue work, waiting outreach, overdue supplier invoices, waiting approvals and failed automations. **Result history** under the button shows when each was last read. Nothing is posted to a channel. Its pill reads **On** once it is running. **Paused** means something is missing, and the lines under the switch say what, such as "Agent Inbox is switched off." The Agent Inbox is covered in . ## Can it draft a chat message for you? [#can-it-draft-a-chat-message-for-you] Yes. In the Agent panel, ask it to write a post for a channel. It drafts the message for you to check, edit and approve, per . ## Who can ask it questions? [#who-can-ask-it-questions] Everyone who holds **Use the agent** on **Settings › Permissions**, which every role holds by default. What each answer contains then follows that person's own permissions, per . For questions about tasks, chat, documents or outreach, or for **Selling › Reports** figures, ask in the Agent panel instead, per . ## What else does it do once it is on? [#what-else-does-it-do-once-it-is-on] It suggests a task to chase each late purchase order, such as "Call Portside Mills about PO-1042". When the supplier has a **Primary** contact, the task names them. The task is assigned to the order's owner. Nothing is created until someone approves it. The suggestion's header reads **Suggested by the watcher**, and it waits in the Agent panel under **History › Updates › Waiting on a decision**, per . The late orders are found once a day by the **What needs attention** job, under **Background jobs** on **Brain › Automations**. Switch that job on as well, per , and see for working through late orders by hand. ## How often does it chase the same order? [#how-often-does-it-chase-the-same-order] Once every three weeks at most. An order that is still late three weeks after its last suggestion gets a new one, unless a chase task for it is still open. At most three chases wait for a decision at once, most overdue first. Approve or refuse them and the next ones follow. ## Where are late-order chases filed? [#where-are-late-order-chases-filed] Under the order's own team when it has one. Next comes the team of the person who owns the order, then of the person who created it, when that person belongs to just one team. Otherwise the chase goes under the team you choose in **Late orders with no team** on the **Agent** card, which starts as **Not chosen**. Its hint reads "When a purchase order runs late and has no team of its own, its suggested follow-up goes to this team. Not chosen: those orders get no suggestion." Choose a team so every late order gets a chase. If the team you chose is later archived, the card says so in red, for example "Production was archived. Choose another team, or late orders with no team of their own get no suggestion." Choose another team. ## What does it cost to run? [#what-does-it-cost-to-run] The card's last line starts "Cost:" and shows what model usage across the whole workspace cost over the last 30 days, over how many calls, and the dates it covers. When nothing has used the model, it reads "No model calls in the last 30 days." ## How do you switch it off? [#how-do-you-switch-it-off] Press **Switch off** on the **Agent** card. The pill then reads **Off**. Its past messages stay where they are. Late-order chases and late-task check-ins stop until you press **Switch on** again. ## What should you check when it does not answer? [#what-should-you-check-when-it-does-not-answer] Work through these in order: * The **Agent** card on **Brain › Automations** has its pill reading **On**. * **When it answers** on that card is set to **When named**. * The card shows a recent "Last looked for new work" line, not "Its background check is switched off." * Your message names it with **@Agent** picked from the list, not typed as plain text. * The Agent can see the channel. For a private channel, add it as a member. * You hold **Use the agent**. * A minute or two have passed since you sent the message. If all of those hold, see .