Copilots: the AI that works for your team, inside the chat and in Team + AI

When a company thinks about AI for its support, it almost always thinks of the same place: a bot that replies on WhatsApp. That's the visible part. The other half of the work —rereading a long case before picking it up, preparing tomorrow's meeting, knowing how many complaints came in during August, putting together the proposal the customer asked for yesterday, training whoever joined last week— is still done by hand, with the platform in one tab and the spreadsheet in another.

AsisteClick's Copilots are that other half: AI Agents that work for your team, not for your customers. They don't reply to anyone outside. They live inside every conversation, in Team + AI —the platform's internal chat— and in CRM notes; they know which customer is being discussed without anyone explaining it, and they return a summary, a reply to review, an Excel file, a report with charts or a proposal with your company's brand.

Between September and October 2026 they changed a lot: they became available from the Company plan, added seven new templates —five of which query the account's data with the permissions of whoever asks—, learned to generate files that open next to the chat and gained their own panel inside every conversation. This guide covers everything you can do with them today: where they're used, what they know, which templates exist, what they don't do and what they cost.

Table of contents

• Two different AIs: the one that serves and the one that helps
• Four places where a Copilot works
• Inside the chat: the Copilot panel
• The private note: asking without leaving the conversation
• Team + AI: the internal chat where the Copilot is one more teammate
• From the CRM: a note that also talks to the AI
• What a Copilot knows without being told
• Sixteen templates: a Copilot for every job
• Copilots that query the account, with your permissions
• Files that open next to the chat
• Training without putting a customer at risk
• Privacy: who sees what
• What a Copilot costs
• What a Copilot doesn't do
• How to get started
• The summary
• Keep reading

Two different AIs: the one that serves and the one that helps

Two families of AI Agents live side by side in AsisteClick. The AI Agent serves your customers on a channel: it replies, resolves, transfers and closes according to its rules. The Copilot works for your team: someone on the inside asks it for something and it answers that person. They look alike inside —both are built in the same builder, with instructions, a knowledge base and tools—, but they do opposite jobs.

The separation isn't a technicality. Instructions written to help an agent ("give me two reply options", "this is what you should tell them") don't work for talking to a customer, and vice versa. That's why, since October 10, a Copilot can no longer be chosen as the bot of any channel: it's AI for the inside. The separation also sorts out the plan: the AI that serves customers stays in IA Plus, while Copilots work from Company, with your own account's OpenAI key.

Four places where a Copilot works

A Copilot doesn't have its own screen you have to go to. It shows up where the person who needs it is already working:

1. The chat panel: to the right of every customer conversation, next to the contact record, their deals and their tasks.
2. The private note: typing @ and the Copilot's name at the start of a note, inside the conversation.
3. Team + AI: a conversation of your own with the Copilot, as with a teammate, or adding it with @ to a chat with someone else.
4. CRM notes: mentioning it in a note on a contact, a company, a deal or a task.

The next four sections go through each door. If you're only going to use one, make it the first: it's the one that saves the most time day to day.

Inside the chat: the Copilot panel

Since October 4, the Copilot opens from the conversation itself. Its avatar appears in the chat's right-hand column, below Info, Contact, Company, Deals and Tasks, in the chats of its department. One click opens a conversation with the Copilot that already knows which chat it's about: no need to copy messages or explain who the customer is. You can ask it for a summary, a reply, a translation or whatever that Copilot does.

What's worth knowing about the panel:

• Each agent has their own thread with the Copilot in each chat. When you come back to the conversation, the thread picks up where it was. If the reply arrives with the panel closed, the avatar shows an unread counter; if you're in another chat, you get a notification with the link.
• Insert into reply moves the Copilot's text into the customer chat's reply box. There you review it, adjust it and send it yourself: nothing goes out to the customer unless you send it.
• Update reruns the request with the conversation up to date. It's useful when the customer wrote again while you were reading the reply.
• It accepts text, voice notes and attachments of up to 40 MB per file, like the chat's paperclip. Whatever you wrote and didn't send is saved.
• The next question, with TAB. The Copilot suggests what to ask next and leaves it in grey in the text box; the TAB key types it.
• New conversation clears the thread with that Copilot in that chat and starts from scratch. Continue in Team copies the conversation to Team + AI, where the Copilot still knows which chat it's about.
• Mentioning another Copilot with @ inside the panel makes the mentioned one answer, under its own name. For example, asking one for the summary and another for the translation without leaving the thread.
• Each chat shows up to three Copilots. If the conversation moves to another department, the Copilots follow it; if one can no longer follow, the panel says why.

Request on open: the summary that's already done

Each Copilot's settings have an option called Request on open, which comes turned off. If you turn it on, the Copilot starts by answering a fixed request as soon as the panel opens; for example, "give me a summary of this chat". Turned off, the conversation starts with the agent's first message and opening the panel costs nothing.

It's a cost-versus-convenience decision. Turned on, opening the panel triggers a request that's charged to your OpenAI key, even if nobody ends up reading the reply. It makes sense in a summary Copilot for a team that picks up a lot of other people's cases —second-level support, the afternoon shift that receives what the morning shift left—, where the summary is always the first thing asked for. For everything else, better off.

How to turn it on

An account administrator does it: in Setup → AI Agents and bots, in the Copilot's settings, the option Chat panel. Next to it is Request on open. From then on, the avatar appears in the chats of the Copilot's department.

The private note: asking without leaving the conversation

Before the panel, the way to use a Copilot inside a chat was the private note, and it still works the same. In the reply box you type @, pick the Copilot from the selector —it appears with the green Copilot badge, and you only see the ones from your departments— and then your request:

@summary what's still pending in this case?
@sales how do I handle the price objection?
@translator translate the last message into Spanish

The note is sent with the yellow Private note button (or Ctrl + N). The mention only works in a private note: if you pressed the send-to-customer button with a message that mentions a Copilot, the platform stops you and asks whether the message is for the customer or is a note.

The Copilot answers in another private note. While it works, an indicator appears at the bottom of the chat with its name, a phrase about what you asked and a timer —for example, @summary · Summarizing the case 0:12—, so you can tell a long request from a stuck one. To send the customer what it drafted, the pencil Use in reply to customer moves the text down to the reply box, without the mention at the start.

Three useful details:

• A specific message as context. If you use Reply to this message on a customer message and mention the Copilot in the note, that message travels with your request.
• Several Copilots in the same note. Each one answers separately, with its own note.
• The conversation continues. The Copilot remembers what it's been discussing with you in that chat, so all it takes is @summary make it shorter or @sales now in a more formal tone.

Panel or private note?

The difference is who sees the reply. The panel's reply is yours: each agent has their own thread. The private note's reply is seen by every agent with access to that conversation. That's why the panel is for thinking —trying drafts, asking for a second opinion, asking something you don't know— and the note is for leaving a record: the summary whoever takes the case tomorrow will read, or the internal closing notes.

Team + AI: the internal chat where the Copilot is one more teammate

To ask a teammate something about a customer, many teams end up leaving the platform: a message in another app, a screenshot, a "did you see María's chat?". The context ends up scattered and later nobody knows what was decided. Team + AI is the answer to that: the last tab in Conversations, a one-to-one internal chat with anyone in your account and with the Copilots. Everything written there is internal.

Chatting with teammates works on every plan and for every role. Copilots answer from Company; on a lower plan they show a padlock and, when clicked, explain how to activate them.

A conversation with a Copilot in Team + AI is private: only you see it. You can have several with the same Copilot —one per topic, each with its own thread and memory—, name them and delete them when they're no longer useful. What you can do there:

• Send it files. With the paperclip, by dragging or pasting a screenshot, up to five files per message and 40 MB each. It reads images, PDFs, spreadsheets and text documents, and transcribes audio and voice notes. It doesn't watch videos.
• Bring it a customer conversation. Typing # and the chat number —#1NJDUC summarize this case and tell me what's pending—, the Copilot loads it with its full context, as long as it belongs to your departments or is assigned to you. Before the reply, a line appears: Context: ticket #1NJDUC · customer · channel: if it doesn't appear, nothing was loaded.
• Add it to a chat with a teammate. In a conversation with another person, @ offers only Copilots. The mentioned one answers in that same chat and both of you see it: @summary summarize what we've been discussing.
• Forward messages to it. A Team message, with text or an attachment, can be forwarded to up to five people or Copilots, with a comment that's mandatory for a Copilot, because it's the request. And from a customer chat, any message —from the customer, a team reply or a note— can be forwarded to Team with its origin: "Forwarded from customer X".
• Pass it a report as data. In a conversation with a Copilot, typing / or tapping the chart icon lets you attach one of the account's saved reports, with the period of your choice. It travels as a spreadsheet, with the same table you see in Reports, and the Copilot does the math on it.

In Team + AI, also, every contact, company, deal or task the Copilot names appears at the bottom of its reply as a card that opens the record in the right-hand column, without leaving the conversation.

From the CRM: a note that also talks to the AI

The fourth door is in AsisteCRM. In the notes box of a contact, a company, a deal or a task, typing @ and choosing someone turns the note into a notification: it's saved on the record, as always, and it also reaches that person through Team + AI with a card for the record. The button changes to Add note and notify @name, so you know who it will reach.

If the one you mention is a Copilot, the AI receives your note together with the record and answers knowing which contact, company, deal or task you're talking about. The reply arrives in Team + AI, in a conversation dedicated to that record and titled with its name. Every new note in which you mention that Copilot from the same record continues the same thread, with memory, and below the notes box Conversation with @copilot appears as a shortcut.

Two uses that work well:

• Asking about an account without building the context. In a company's note: @data how are their conversations this quarter compared with the last one?
• Leaving a teammate a summary. If you mention a person and a Copilot in the same note, the message reaches the conversation with that person and the Copilot answers right there: both of you see the reply. It's the fastest way to hand an account over to someone with a summary made by the AI.

The conversation with the AI about a record belongs to each user: if someone else mentions the same Copilot from the same record, they have their own thread.

What a Copilot knows without being told

Until October, a Copilot only saw what its instructions requested with variables: if the prompt didn't ask for the history, it answered without seeing the conversation, and gave no warning. Since October 4, every request reaches it with the full context, without configuring anything:

What doesn't exist isn't sent: without the CRM module there's no record, and in Team + AI, without a conversation brought in with #, there's no history or customer. But when the data exists, it arrives. That changes how a Copilot's instructions are written: it's enough to say what it has to do and with what criteria —"summarize in five lines, always flag the promises we made to the customer"—, without wrestling with variables.

Memory

A Copilot remembers each conversation for 25 days from its last reply and for up to 40 chained replies. After that it starts over and relies on recent messages. In a very long thread it may not have the oldest things in mind, even if you see them on screen: for a new topic, better a new conversation.

What you tell it

The context tells it where it is; the instructions tell it what to do. That's where what only your company knows goes: the tone, the policies that can't be broken, the products and their prices, what counts as an urgent case. And if you connect a knowledge base with your manuals and procedures, it answers with them. Each Copilot also has a Description tab to tell the team what it's for; the agent sees it in the details of their conversation with that Copilot.

Sixteen templates: a Copilot for every job

No Copilot has to be built from scratch. In Setup → AI Agents and bots, the Templates tab has a Copilots filter with sixteen templates, each with a specific job and instructions that already work. They can be used as they are or adapted to your business.

In the conversation with the customer

The first nine work on the chat in front of you. They're the ones used most from the panel and from the private note.

Two rules come written into the templates and are worth keeping when you adapt them. Sales Opportunity never suggests selling if the customer has an unresolved problem: first you resolve, then you sell. And Internal Notes separates the promises made to the customer from the rest of the to-dos, because they're what gets lost most when a case changes hands.

To sell

The three sales templates work on the whole CRM, not on a chat. They're used mostly from Team + AI, though they can also be mentioned from a customer's chat so they work on that contact.

Meeting Prep. Before sitting down with a customer, it puts together a brief that works as a script: how to open, what to achieve and what to ask for at the close; three questions for that account —what the CRM is missing becomes questions—; the risks and how to handle them; the deal visually, with the stage, the amount and the days it's been there; the account's timeline, and what's happening at the company according to its website, with the source. All on one page with your company's brand, which can be printed or saved as a PDF. You ask it the way you'd ask an assistant: "I'm seeing Acme tomorrow" or "what meetings do I have tomorrow?". On the web it only looks up the company, never the people.

Proposal Builder. It takes a deal from the CRM and returns two documents ready to send: the quote, formal and brief, with the itemized detail, the total and the terms; and the proposal, the one that sells, with what we understood about the customer, their goals and how we'll measure them, the solution tied to those goals, the implementation, why us and the next steps, built with the account's real history. Both come out with your logo, your colors and your fonts. In the chat it leaves a summary of what's worth reviewing before sending them, and it never makes up a price or a feature: it takes them from what you loaded.

Upsell Detector. It goes through the portfolio and lists the accounts ready for an additional sale —a higher plan, more users, a product they don't have yet—, from clearest to least clear, each with what to offer, the signals that support it and the next step. It looks at the account's health and risk, the expansion potential and NPS, what they bought versus what you sell, the users they use versus the ones contracted, whether their conversations grow or fall in the last 90 days versus the previous 90, whether the renewal is near and how their previous deals went. Accounts at risk go separately: those get offered nothing, they need to be retained first. And if an account already has an open deal, it suggests adding the offer there instead of opening another.

To manage and measure

Ask Your Data. An analyst for the team: it answers business questions in natural language about conversations, contacts, companies, deals and tasks. "How many chats did we have per channel in August?", "which salesperson has the most stalled deals?", "how is the pipeline doing this quarter?". It delivers the key figure and up to three findings, and explains how it calculated it: which entity, which filters, which period. When the result has a series or a comparison, it builds a dashboard with charts; if you ask, the data in Excel or CSV.

The detail of how it calculates matters more than the answer. A number that can't be checked is useless for deciding, and a language model that "calculates" by reading rows tends to round or make things up. That's why this template doesn't count by eye: it translates the question into a query, resolves the names of agents, departments and stages to their identifiers, and asks the platform for the aggregate, where the number is exact. For rankings, cross-tabs and variations it uses a code interpreter. It never estimates.

Agent Coach. A private coach for every person on the team. It reviews the conversations they handled in a period —by default, the last seven days of whoever writes to it— and gives short feedback: two things they did well, each with a quote from their own messages and the conversation number; one single thing to improve, with the reply rewritten, and a tip to practice that week. You ask it "how did I do this week?", or a supervisor asks for the feedback of someone on their team, always within their permissions. If you load your quality guidelines —what good support means in your company—, it measures against them.

To train

Customer Simulator and Devil's Advocate don't help with a real case: they invent one, for practice. They have their own section further down.

Copilots that query the account, with your permissions

Five of the new templates —Ask Your Data, Upsell Detector, Meeting Prep, Proposal Builder and Agent Coach— aren't limited to the open conversation: they directly query the account's conversations, CRM and statistics. And they do it in a particular way: with the identity of whoever writes to the Copilot. Through the Copilot, each person sees exactly what they see on the platform. No more, no less.

This solves the classic problem of putting AI on top of a company's data: who sees what. A Copilot with its own permissions —a key with access to the whole account— forces you to choose between giving it only to management or opening up to anyone information they shouldn't see. One that asks with the identity of whoever uses it doesn't have that dilemma. An agent gets the Coach's feedback on their own chats; a supervisor can ask for their team's. A salesperson sees in the Upsell Detector the accounts they could already see. And the CRM's pipeline permissions, which have applied across the whole platform since September, also apply to what the Copilot sees: whoever doesn't have access to a pipeline won't see it show up in the answers.

Three more guarantees, built in:

• They only read. These five templates query; they don't create, change or delete anything. Meeting Prep sees the calendar, but doesn't book or cancel appointments.
• No token needed. Creating them doesn't ask for the account's access key: they connect with the session of whoever writes to them.
• They show where the data comes from. Ask Your Data says how it calculated each figure; the Upsell Detector, which signal supports each recommendation; the Coach quotes the messages. What they say can be checked.

Files that open next to the chat

Some requests aren't answered well with a paragraph: a spreadsheet with the month's orders, a report with charts, a Word draft, a presentation for a meeting. When the Copilot builds a file, its reply comes with a card with the name; one click opens it in the Artifact tab of the right-hand panel, next to the conversation, with options to expand it to full screen, open it in a new tab or download it.

Not everything has to be a file: when the Copilot replies with a table in the text, it shows as a table in the same bubble, and if it has more than ten rows it folds with a See the remaining rows. In a spreadsheet, the panel shows up to 10,000 rows; for more, you download the full file.

To generate files, the Copilot needs the tools that do it: Code Interpreter, for spreadsheets, PDFs, reports, documents, presentations and charts, and Image Generation, for images. Both work with the OpenAI provider. Some templates, like Ask Your Data and the Upsell Detector, already come with the code interpreter turned on; in the rest it's turned on from the builder, in Built-in tools. For HTML reports it's worth adding the HTML reports and dashboards tool from the catalog, which teaches the AI the rules of the panel where the report will be shown, so it doesn't come out empty or broken.

Who can open those files

An administrator sets it in the Copilot's settings, with the option Files the AI generates. By default, account users only: the link requires being logged in to your AsisteClick account, and anyone else sees a notice that the file is private. The alternative, anyone with the link, leaves the file at a public, hard-to-guess address; it's only useful if you're going to share those files outside the platform.

A practical caution: if you pass the customer the text of a reply that includes a private file, the customer gets a link they can't open. To send them a file, download it and attach it to your reply.

Your company's brand

Proposals, meeting briefs and reports come out with your brand because the brand is account data. In My account → Company, the Brand section stores the logo, colors, fonts, slogan, what the company does, who it speaks to, the tone, how it addresses the customer —informal, formal or the regional voseo—, the key messages and what not to say, with a preview. Every Copilot that has the Brand and design skill applies it: Proposal Builder and Meeting Prep already include it.

Training without putting a customer at risk

The usual way to train someone new in support is to let them handle customers and correct them afterwards. It works, but the mistakes are paid for by a real customer. The two practice templates change the order: first you practice with an invented customer, then you serve real ones.

Customer Simulator plays a difficult customer. The agent picks the profile —angry, confused, demanding, threatening to leave, in a hurry—, the difficulty —easy, medium or ruthless— and the case. The Copilot gets into character and responds like a real customer: it doesn't give all the information up front, reacts to how it's treated and has an underlying need the agent has to discover. When it's over, it steps out of character and gives feedback with a score for empathy, diagnosis, clarity and resolution, flags whether the agent promised anything outside the company's policies and proposes the ideal reply for the hardest moment.

Devil's Advocate does the same for sales: it plays a skeptical buyer. The salesperson picks the profile —procurement, skeptical technical buyer, owner in a hurry, a competitor's customer—, the difficulty and the scenario: first contact, price negotiation or renewal. The buyer has a hidden objection that only comes up if the salesperson asks good questions. The feedback scores discovery, objection handling, value proposition and closing, and rewrites the weakest moment.

Both are used from Team + AI: you open a conversation with the Copilot and type "start". "Pause", "we're done" or "feedback" break character at any time. And both accept a real case: if you bring a conversation with # and its number, the Simulator plays that customer, with their problem and their tone. It's useful for rehearsing before picking up a delicate case.

What makes them useful is what you load into them. With the objections your team really hears, your competitors, the frequent cases and the policies that can't be broken —refunds, deadlines, compensation—, the practice resembles your operation and not a generic manual.

Privacy: who sees what

An assistant the team tells its doubts, drafts and opinions about customers has to be truly private. These are the rules:

• Who can use a Copilot: only the people in its department, with no exceptions by role. When you create it, the platform asks who's going to use it.
• Who can create it: whoever has the Bot and AI agent builder permission. The Super administrator always has it; everyone else receives it from them.
• Your conversations in Team + AI are yours. They don't appear in the conversation inbox, don't go out through the account's webhooks, and aren't visible in the API or in the contact record's history. Nobody can open them, not even a Super administrator. In AI Observability you can see how much each Copilot consumes, its turns, its models and its errors; the content, never.
• What you ask inside a chat stays in the chat. A request by private note and its reply are notes in that conversation, visible to the agents who have access to it. The customer doesn't see them.
• The Copilot doesn't see more than you do. The templates that query the account do it with your permissions, as explained above.

What a Copilot costs

Copilots come with the Company plan and with IA Plus; the details of each plan are in plans and pricing. What each request consumes is charged by OpenAI directly to your account's key, depending on the model and the length of the conversation.

Since the history travels with every request, the cost depends mostly on the context: a short conversation is cheap; a weeks-long thread, a 150-message chat or a large spreadsheet weigh more. That's where the levers to control it come from:

• Request on open turned off, except where it's really always used.
• A new conversation per topic, instead of an endless thread.
• Files and reports when they're useful: an Excel file or a dashboard consumes more than a text reply and can take a few minutes.

Spending isn't estimated: it's measured. With the Company plan, AI Observability shows your Copilots' consumption —cost, breakdown by Copilot and by model, consumption per user— and separates the origin of each turn: Copilot in Team chat, what's asked from Team + AI, and Copilot in tickets, the mentions inside a chat.

What a Copilot doesn't do

Knowing the limits avoids using it for what it isn't:

• It doesn't talk to the customer. Since October 10, a Copilot can't be chosen as the bot of any channel. If an account had one set up that way, that channel stops serving with it: the chat goes to a person and the account owner sees a note explaining what to put in its place. To serve customers with AI there's the AI Agent.
• It doesn't route or close. A Copilot never transfers the conversation or opens it in the inbox. If its flow tried to route it, nothing happens and it leaves a notice for an administrator to fix it.
• It doesn't send anything for you. Insert into reply and Use in reply to customer only move the text into the box. The person reviews and sends.
• The data templates don't change the account. They query, calculate and build documents; they don't touch a record.
• It doesn't watch videos, and its memory has a limit: a few weeks and about forty replies per thread.
• It isn't instant on heavy requests. A report or a spreadsheet can take several minutes; each call to the model has a 15-minute cap, and if it doesn't make it, the Copilot lets you know.

And a note on scope: a Copilot multiplies the team's judgment, it doesn't replace it. A suggested reply about a poorly documented policy is still wrong; Coach feedback without quality guidelines measures against a generic standard. The quality of what it returns depends on what you load into it, just like in an AI Agent that works alongside people.

How to get started

Creating a Copilot takes a few minutes:

1. In Setup → AI Agents and bots, open the Templates tab and choose the Copilots filter.
2. Click Create Copilot on the template you want and fill in the name, a short hashbot —it's what the team will type after the @, such as summary or data—, the Default department, which defines who can use it, and your account's OpenAI key.
3. Review the instructions and adjust them to your business: your policies, your products, your tone. Fill in the Description tab so the team knows what it's for.
4. In the Copilot's settings, turn on Chat panel and decide whether you want Request on open.
5. If it's going to generate files, turn on Code Interpreter or Image Generation. If it's going to build documents with your brand, fill in your company's Brand section.
6. Click Deploy bot.

Three Copilots for the first week

There's no need to create all sixteen. Three are enough for each part of the team to see the value:

• For support: Summary, in the chat panel. Every case someone picks up starts with a five-line summary instead of fifty messages to read.
• For management: Ask Your Data, in Team + AI. The questions that today end up as a request to the analyst, or a spreadsheet built by hand on Monday morning, get answered on the spot and with the calculation in view.
• For newcomers: Customer Simulator. An hour of practice with difficult customers before the first real one.

If you'd rather not set them up alone, our team takes care of it: we choose with you which Copilots each area needs, adapt their instructions to your policies and the way you work, and leave them configured and tested with your real cases. You can see more on the page for AsisteCopilot.

The summary

The AI that serves customers is the visible one, but a good share of a support and sales team's time goes into inward-facing work: understanding a case, preparing a meeting, answering a question about the numbers, building a document, learning to handle a difficult customer. AsisteClick's Copilots put AI into that work, in the place where it's already done —inside the chat, in the internal chat and in CRM notes—, with the full context of each conversation, with each person's permissions and with files that open without leaving the platform. They never talk to the customer and never decide for the team: they leave the work done so the team can decide better.

Frequently asked questions

What is a Copilot in AsisteClick?

It's an AI Agent that works for your team and not for your customers. It's used from the panel of each conversation, with a private note that mentions it, from Team + AI —the platform's internal chat— and from CRM notes. It receives the context of the conversation, the customer and the CRM without configuring anything, and only answers the person who asked it for something.

How is a Copilot different from an AI Agent?

The AI Agent serves your customers on a channel and can transfer or close conversations; it requires the IA Plus plan. The Copilot helps your team, never writes to the customer or routes, and works from the Company plan. Since October 2026 a Copilot can't be chosen as a channel's bot.

Which plan do I need to use Copilots?

Copilots work from the Company plan, with your own account's OpenAI key, and also on IA Plus. The internal chat with teammates in Team + AI works on every plan.

Does the customer see what I discuss with the Copilot?

No. In a customer chat, the request and the reply are private notes, seen only by agents with access to that conversation, and the chat panel is a thread for each agent. Nothing goes out to the customer unless the person sends it.

Can anyone read my conversations with a Copilot?

No. Your conversations with a Copilot in Team + AI are private: they don't appear in the inbox, the API or the webhooks, and nobody can open them, not even a Super administrator. AI Observability shows how much they consume, but not their content.

Can the Copilot see data I can't see?

No. The templates that query the account —Ask Your Data, Upsell Detector, Meeting Prep, Proposal Builder and Agent Coach— do it with the identity of whoever writes to them, in read-only mode: through the Copilot, each person sees the same as on the platform, including the CRM's pipeline permissions.

What files can a Copilot generate?

With Code Interpreter, Excel and CSV spreadsheets, PDFs, Word documents, PowerPoint presentations, HTML reports with charts and videos; with Image Generation, images. They open inside the chat, in the Artifact tab, and by default only your account's users can open them.

How much does it cost to use a Copilot?

Copilots are included from the Company plan. What each request consumes is charged by OpenAI to your account's key, depending on the model and the length of the conversation, and it can be seen per Copilot and per user in AI Observability.

Keep reading

• AI copilot for agents: real-time replies — why it pays to assist people before replacing them
• Decisions: rules your AI Agent follows even when the model won't — the other half: the AI that serves customers, with rules that don't depend on the model
• AI Agent and human collaborator: the roles in your support — who does what when AI and people work together
• The 3 knowledge layers of an AI Agent — the knowledge base your Copilot can also use