Company administrator
Sees the whole operation: channels, agents, conversations, usage and everything the Squads have published. This is the person who creates Copilot chats and decides who else can use it.
HAL Copilot
The AI assistant for the people who run the operation
While channel agents take care of your customers, HAL Copilot takes care of you. It reads your own company's data and answers in plain language, with a table and a chart on screen, a PDF ready to go, and delivery by WhatsApp or email.
You don't open one report after another. You ask.
Available in the platform console for the company administrator and authorized supervisors.
HAL Copilot
Data Analyst gem
How did each queue perform this week?8:41 AM
I pulled the period broken down by queue. The table and the bar chart are right here on screen.8:41 AM
Send it to my WhatsApp as a PDF.8:42 AM
Generated and sent. Want me to run this same check every Monday morning?8:42 AM
Who it's for
Built for the people accountable for the operation
The Copilot doesn't talk to your customers. It talks to the people who have to decide: the company administrator and the supervisors who watch queues, agents and campaigns every day.
Supervisors
They follow their own queues and their own agents, with the same permission scope they already have in the call center — the Copilot respects that scope, it never widens it.
- Every conversation starts with a chosen gem, so the assistant begins with the right frame.
- Questions are asked in plain language: there is no query syntax to learn.
- The console is bilingual, English and Portuguese, and works on a phone.
Specialized gems
Five specialists, one assistant
When you create a chat, you pick the gem. It sets the Copilot's repertoire for that conversation: what it tends to look up, how it reasons about the subject and the format it answers in.
Data Analyst
The numbers behind the operation: conversation volume, behavior by queue, by channel and by period. It answers with a table and a chart, and explains what the chart is showing.
Routine Manager
What is scheduled and what actually ran: automated agent tasks, recurring checks and Squad executions. Good for spotting whatever stalled.
CRM Strategist
Looks at your book of business inside the intelligent CRM: customer profiles, AI-generated traits, open conversations and opportunities going cold.
HAL Support
Questions about the platform itself: where each setting lives, what a channel state means, how a feature behaves. It's the manual that answers by asking back.
HAL General
The open gem, for when the question doesn't fit in a single box — or when you're still working out what you want to know.
Memory in all of them
Each chat keeps the context of what has already been discussed. Today's conversation picks up where one from weeks ago left off — the same memory technology described in the patent filed with the USPTO.
How it answers
The answer comes out ready to circulate
It isn't a paragraph of text asking you to go open another screen. The Copilot hands over material that is already formatted and already on its way to whoever needs to read it.
Text, table and chart on screen
The answer is laid out in HTML inside the chat: sections, tables with the data and charts drawn on the spot. You read the number and see its shape, not just its value.
PDF export
The same content becomes a server-generated PDF containing the entire document — nothing truncated, no ellipsis halfway through a table.
Delivery by WhatsApp or email
Ask for it to be sent and it goes out on the channel you name, using the same official WhatsApp and email infrastructure the rest of the platform already runs on.
Refine it in the conversation
Changed your mind about the scope, the period or the breakdown by queue? Say so in the next sentence. It reworks the answer on the same context, without starting over.
From "can someone pull this number for me?" to a finished document
The question that used to become a ticket for the data team — and came back days later — turns into a two-minute conversation, with the result already in the format it will be presented in.
- Ask in plain language, no hand-written query
- Table and chart in the same place as the answer
- PDF of the complete document
- Delivery by WhatsApp or email
The screen below shows the Copilot in the middle of one of those conversations. On the left is the manager's list of chats, and each chat carries the gem that opened it — Data Analyst, Routine Manager, CRM Strategist, HAL Support — along with the day of the last message.
On the right is the answer to a question written in plain language: first the paragraph interpreting the result, then the chart of conversations by queue over the last seven days and the table with volume, waiting conversations and average time for each queue. The footer of the answer carries the time, the notice that a recurring check is scheduled for thirty minutes from now, and the buttons to export as PDF or send by email.
On the bottom bar, next to the message field, the Analysis selector keeps the cost in plain sight: the deep read, which goes through conversation after conversation, is a choice the manager makes on purpose — not a silent behavior.
Queues this week
Base specialist: Data Analyst
Clear Conversation
You · 09:12
HAL Copilot
There were 3,135 conversations across the four queues in the last 7 days. Sales has the highest volume, but the one holding things up is Support: 31 of the 49 waiting conversations and the worst average time of the week, 6m40s. Scheduling and Billing are within the expected range.
Conversations by queue — last 7 days
Expand chart
| Queue | Conversations | Waiting | Average time |
|---|---|---|---|
| Sales | 1,240 | 12 | 2m10s |
| Support | 980 | 31 | 6m40s |
| Scheduling | 610 | 4 | 1m50s |
| Billing | 305 | 2 | 3m05s |
| Total | 3,135 | 49 | 3m36s |
09:14 · Check scheduled for 30 minutes from now
What you're looking at
Just under the Copilot's name, a row of tags states the scope of the answer: the gem that opened the chat, the analysis mode used and the period consulted — seven days, four queues. In the chart, Support's amber bar and the legend beside it flag the one queue outside the norm.
In the table, the total row closes the math the paragraph summarized: 3,135 conversations, 49 waiting and a 3m36s average time. Every number quoted in the text is printed right below it, in the table it came from.
The advantage
The conclusion arrives with the math next to it, inside the same card. Checking it no longer means opening the source data and redoing the calculation by hand: the check fits on the same screen, while the conversation is still open.
And because the scope is written into the answer itself, whoever gets the PDF later knows which period and which queues the number came from — no second round of questions just to find out how it was pulled.
Only on Hal-AI
No "conversations by queue" report was ever designed for this screen. The question came in as plain language and the Copilot chose what to query, what format to show it in and which row deserved attention — including deciding that Scheduling and Billing needed no comment.
A workflow automation hands back the screen somebody designed for it. There is no decision tree drawn here: there is an agent with a repertoire, tools to read with, and the permission scope of whoever asked.
Autonomous work
Give it the mission and close the page
A long analysis shouldn't hold anyone hostage to a screen. The Copilot runs in its own process on the server: you describe what you want, leave the console and get a notification when it's ready.
You describe the mission
In one sentence: what to investigate, over which scope, and what you expect to receive at the end.
The work continues without you
Execution moves to the server. Closing the tab, walking away from the computer or switching screens interrupts nothing.
Your browser lets you know
When the result is ready, a push notification arrives in the browser. You go back to the chat and the document is there.
You send the result on
Read it on screen, export it as a PDF, or send it straight over WhatsApp or email to whoever needs to act.
Waiting in front of a screen is not part of the job.
The screen below is the Copilot's mission panel. At the top is the manager's queue with the state of every request — one queued, one running and two already done — and under it the mission running right now, with the record of what it has completed so far: the scope loaded, the queries made, the conversation-by-conversation read and the grouping of the reasons.
Background missions
Every mission runs in its own process on the server, outside your browser session.
+ New mission
| # | Mission | Gem | Requested | State |
|---|---|---|---|---|
| 13 | Cancellation reasons by channel | Data Analyst | today 9:24 AM | Queued |
| 12 | Why Support keeps hearing the same complaint | Data Analyst | today 9:20 AM | Running |
| 11 | Accounts untouched for over 30 days | CRM Strategist | today 8:05 AM | Done |
| 10 | Routines that didn't run this week | Routine Manager | yesterday 5:40 PM | Done |
Mission #12 · running
Stop
Support · last 14 days · conversation-by-conversation read
Running Mission #12 · requested in chat Step 5 of 7
09:20:03Context loaded — scope: Support queue, last 14 days
09:20:06api_get_conversations_by_queue(queue="Support", days=14)
09:20:11412 conversations in the queue; 96 with a repeated complaint
09:21:40api_get_conversation(id) — 96 of 96 read
09:24:52Reasons grouped into 5 themes
09:25:072 conversations with no transcript — left out of the grouping
09:25:09Building the document and the chart
96 conversations read · 94 grouped · 2 with no transcript running on the server
You can close this page
Push active in this browser
What you're looking at
Mission #12 is halfway through: it has already loaded the scope, queried the queue, read the 96 flagged conversations and grouped the reasons into five themes. The step tag and the clock on each line show where it stands without anyone having to ask.
The footer closes the math the last line opened: 96 read, 94 grouped and 2 with no transcript. The bottom strip says the essential part — the page can be closed, and the notice arrives by push.
The advantage
Reading conversation by conversation takes minutes, not seconds. Because execution lives on the server and not in the tab, that time stops being the manager's time: they ask, they leave, and they come back when the push arrives.
And because the mission queue stays in view with the state of each one, asking for a second analysis before the first has finished is routine — none of it depends on keeping the browser open.
Only on Hal-AI
The record is not a progress bar: it is the list of what the Copilot actually did, in the order it did it — the query, the number it returned, the decision to go down to the case-by-case level, and what was left out and why.
When the document arrives, the conclusion comes with the path traveled right beside it. You can check the reasoning, not just the result.
Recurrence and depth
Two ways to go beyond a one-off question
A good answer today helps once. The Copilot handles the two cases where once isn't enough: the check that needs to repeat, and the analysis that needs to go case by case.
Recurring checks
Turn a question that proved useful into a routine: the Copilot schedules the check and starts running it on its own, at the frequency you agreed on.
- The same scope, every time — nobody has to remember to ask again.
- The answer arrives on the channel you already set for that check.
- If nothing changed, that is an answer too: checking and finding nothing is a legitimate result.
Deep analysis, conversation by conversation
Instead of summarizing the set, deep analysis reads each conversation individually and then consolidates. It's the path to questions a total hides.
- Every conversation is read with its own reasoning, not skimmed as a sample.
- It surfaces the real reason behind a repeated complaint, not just how often it appears.
- It is the Copilot's heaviest kind of reading — which is why you decide when it's worth running.
Need the routine to not only analyze but also act and talk to the customer? That's the job of the Squads.
Limits by design
What the Copilot doesn't do
An assistant that reads a company's database needs explicit limits. Here is what it never does — and what always happens on every query.
It never
- Reads data from another company on the platform
- Changes, deletes or inserts a record in the operation
- Lets a question switch the company scope
- Queries a table outside the authorized list
- Takes over a customer conversation in the agent's place
- Launches a campaign or sends a message on its own
It always
- Queries only your own company's data
- Works in read-only mode
- Uses curated, parameterized queries, never free-form ones
- Stays inside a positive allowlist of tables
- Respects the permission scope of whoever is asking
- Leaves sending and exporting as an explicit request from the manager
Frequently asked questions
Questions from the people who will use it
Can the Copilot see data from other companies on the platform?
No. Every query is filtered by the company of the signed-in user, and that scope comes from the session — not from the text of the question. There is no company parameter that can be swapped mid-sentence.
Can it change or delete anything?
No. The Copilot works in read-only mode, with curated, parameterized queries instead of free-form querying, and a positive allowlist of the tables it may read.
Editing a customer, taking over a conversation or launching a campaign remains an action taken in the operations screens, with a named person responsible. The details are in Security.
Do I need to know how to build reports to use it?
No. You ask in plain language, the way you would ask someone on your team. It answers with text, a table and a chart on screen, exports to PDF and sends it by WhatsApp or email when you ask.
See the Copilot running on your operation
Our team walks you through HAL Copilot inside the console, alongside the channel agents and the Squads, and discusses the scope that makes sense for your company.