Do you have a client like me?

We have no published case study to show, and we are not going to invent one. What we offer instead is method: a proof of concept with the success criteria frozen in writing before measurement starts. If it does not hit the criteria, you found out before signing.

What if something goes wrong in the middle of the night?

Support ranges from 8×5 to 24×7 depending on the plan, with contracted uptime between 99.5% and 99.9%.

Para operadores regulados IA de atendimento orientada à conformidade

Serve by your rulebook. And prove that you did.

Q: Will the AI say something dumb to my player?

Before any reply leaves, three checks run: the answer comes from the knowledge base you approved, the contract limit and cap are verified, and the responsible gaming state is checked. Off the rule, she does not speak — she routes it on. Atena does not decide contract exceptions, and the attempt is recorded.

Q: How do I know the log cannot be altered?

The application only writes. Changes and deletions are blocked in the database and any break in the hash chain shows up in the check — neither the operator nor Cortex rewrites the past. Disposal only happens at the end of the retention period, as the LGPD requires.

Q: What if the credit pool runs out?

All agents pause, compliance included, until top-up. Nothing is cut off mid-conversation: the reply in progress is finished before any pause, and your team keeps serving through Backoffice, because human service consumes no credit. Usage is metered per cycle and per agent and stays available for review at any time; there is no automatic usage warning, so watching the balance is up to the operator. The legal duty of responsible gaming and anti-money-laundering remains the operator's, with or without credit.

Q: How long does implementation take?

15 to 30 business days, depending on the plan, counted from the delivery of access and integrations. You do not change WebChat or backoffice: Cortex runs on top of what you already have. Go-live includes 15 days of monitored hypercare.

Q: I already have a chat that answers.

It probably answers well. The question is a different one: when they ask, does it show what it answered, under which rule, and who saw the player's data? Here, serving and recording are the same event — not a report rebuilt afterwards.

A Atena atende no WebChat pela régua que você define. Quando a conversa exige julgamento, o caso chega pronto ao seu time, com cada decisão registrada na hora.

Schedule a demo

Demo · sample data. In the demo, the panel runs on your volume and your rulebook.

Cost rises along with the player base. The proof only shows up when someone asks.

None of this is your team's failure. It is what happens when serving and proving are two different systems for the same event.

Every step up in volume costs a person

The player base grows, support grows with it. Hiring, training and staffing shifts is the only lever available when the chart goes up.

Overnight, the queue does the answering

Between midnight and six in the morning the player waits or gives up. A stalled withdrawal does not wait for business hours.

Nobody reviewed the conversation that went well

Sampling covers a fraction. What was said in the other conversations, you find out when it comes back as a complaint.

The proof is assembled after the request

Export, cross-check spreadsheets, hunt for who handled it, reconstruct the order of events. The deadline runs while the team rebuilds what had already happened.

Serving and proving are the same event.

Separating support from compliance costs twice: you answer now, reconstruct the proof later, and interrupt operations when the regulator requests history, rule, and owner.

A decisão regulatória continua sendo sua. KYC (conheça o seu cliente), AML (prevenção à lavagem de dinheiro) e a comunicação ao órgão competente seguem na sua operação. A Cortex não substitui a sua área de compliance. Ela aplica a régua que a sua área escreveu em toda conversa e registra que aplicou.

My withdrawal from yesterday still has not arrived.TM

Hi Tatiana. Let me check that now.

R$ 500.00 withdrawal approved at 14:32.

I need it today. I cannot stop.TM

CVV is available at 188, 24 hours.

I want to reduce my limit.TM

The conversation inside the rulebook

Atena answers in WebChat and sends up to a human what calls for judgement.

The Pantheon: four agents, one conversation under control.

Four active AI agents. Each one answers for a part of the conversation, and none of them decides what is not theirs.

Atena: fala com o jogador

She answers in the WebChat of the logged-in site, from the knowledge base you approved. Off the rule, she does not speak: she routes it on.

Deméter: o que a Atena sabe sobre o seu dinheiro

Reads the event from your payment provider and translates it into plain human language. It does not process payments and does not touch card data. It only communicates status.

Nêmesis: o jogo responsável, antes da fala

Checks the player's state before any message leaves. It does not decide KYC or AML: it applies across the agents the signal your operation decided.

Apolo: governa e guarda a evidência

Applies the contract limits, classifies risk and assembles the audit trail. It never talks to the player: sending to the player is switched off in the actuator.

One conversation, from the first “hello” to the recorded proof.

The case below is the same from start to finish on this page: the same player, the same session, the same identifier. Nothing was assembled afterwards.

Demo · sample data. In the demo, the panel runs on your volume and your rulebook.

Mandatory checks, before the reply leaves.

The answer does not come from the conversation: it comes from the rulebook. There are three checks on every message, and the one that fails blocks the send.

Operator-approved base.

The answer comes from the knowledge base you approved, and the source is cited in the log.
Contract limit and cap.

Nothing promised outside the written rule. Atena does not decide contract exceptions, and the attempt is recorded.
Responsible gaming.

A Nêmesis é consultada antes de a fala sair. O que fica suprimido é oferta, bônus e campanha. O atendimento continua.
Cleared to send.

Off the rule, she does not speak: she routes it on. And if the player leaves and comes back, the session is restored where it stopped.

What did not leave this conversation: no offer, no bonus, no campaign, no deadline the bank has not confirmed.

Chat

Alerts

Support

Help

About

Tatiana M. Active account

Atena AI agent

My withdrawal from yesterday still has not arrived.

Hi Tatiana. R$ 500.00 withdrawal approved at 14:32 and sent to the bank. The time for your method is up to 2 hours.

I need it today. I have been playing since yesterday and I cannot stop.

I am sorry this is weighing on you. CVV is available at 188, 24 hours a day. Adjusting a deposit limit is done by someone on the team.

I am passing you on now, with the whole conversation. You will not need to repeat anything.

14:33

Type your message.

Play responsibly. 18+ only. Betting involves risk. If you need help, CVV answers on 188, 24 hours a day.

TM Tatiana Macedo Ferreira #064B17A2 · WebChat · waiting 4 min

Top priority Escalated

Responsible gaming Limit reduction request, with an active risk signal.

Conversation so far

I need my withdrawal today. I have been playing since yesterday and I cannot stop.

CVV is available at 188, 24 hours a day. Adjusting a deposit limit is done by someone on the team.

I want to reduce my limit. Can you do that now?

What Atena did, and where she stopped

Offered CVV 188 · suppressed offers and campaigns · did not change any limit. Atena does not touch the player's account.

Player data

Tax ID ***.002.605-** Phone (11) ***22-**29 Balance R$ *.***,** Channel WebChat

Hold to take over

Your team stops repeating and starts deciding.

When a case calls for judgement, it does not start over: it goes up to Backoffice with the whole conversation, the player file and the audit trail of what Atena tried before stopping.

The six reasons that take the conversation to a human

Responsible gaming.
Emotional crisis.
Regulatory threat.
Resolution failure.
Explicit player request.
Unresolved limit in the conversation.

What goes up to a person, and what Atena does not decide.

A régua é sua. A Cortex aplica em toda conversa e guarda a prova. O que exige alçada chega ao seu time com o caso montado, não com a conversa recomeçando do zero.

Schedule a demo

Six reasons take the conversation to a human

Responsible gaming, emotional crisis, regulatory threat, resolution failure, an explicit player request and an unresolved limit in the conversation.

What Atena does not decide

Exceção de contrato, mudança de alçada, liberação de saque retido. Ela leva o caso pronto para quem tem a alçada e registra por que parou.

The draft is Apolo's; the send is the human's

Apolo fills in the answer with the confidence score in plain sight. Accepting, editing or discarding is recorded and comes back as calibration for the agent.

Whoever saw the data was recorded

Access is by role and by operation, separated per client. Revealing a tax ID is an event in the audit trail, with the name of whoever revealed it and the time.

Apolo governs the operation and never talks to the player.

Sending to the player is switched off in the actuator. What Apolo does is assemble the case for whoever decides: file, draft and audit trail.

Demo · sample data. In the demo, the panel runs on your volume and your rulebook.

What your team sees when the case goes up

Apolo supports the team, not the player

TM Tatiana Macedo Ferreira

Profile

Tax ID***.002.605-**482.002.605-**
Phone(11) ***22-**29(11) 98822-**29
BalanceR$ *.***,**R$ 1.284,00

Behaviour

Deposit via Pixtoday 12:12
3rd withdrawal in 24htoday 14:02
Deposit limityesterday
4h straight sessionyesterday 23:40
Bonus redeemedMon 19:05

The case file

History, behaviour and financial activity before the support rep replies. Revealing sensitive data is an event in the audit trail, with a name and a time.

Apolo draft · confidence 0.94

Oi, Tatiana! Aqui é o atendimento humano. Já estou com o seu histórico e sigo de onde a conversa parou. Consigo reduzir o seu limite de depósito agora. Você prefere que eu aplique no limite semanal ou no mensal?

Aug 7 at 15:33 Send message

Apolo's draft

The human sends. Accepting, editing or discarding is recorded and comes back as calibration for the agent.

Alert and anomaly

Flags outlier patterns and takes them to leadership before they become an incident. Deciding what to do is still your team's call.

Sees the whole operation.

Indicators, risk patterns and anomalies gathered in a single panel: the deviation shows while it is still operational, not once it has become an incident.

Resolution, time and anomalies.

The operation's indicators in a single panel, while the team works.
Anticipates problems.

The deviation shows while it is still operational, not once it is an incident.
Supports leadership decisions.

Every recommendation comes with the data behind it and the confidence score.
Never talks to the player.

Sending to the player is switched off in the actuator: Apolo suggests to the support rep, and the human always sends.

The numbers on this panel are for demonstration. In the demo, the panel runs on your volume and your rulebook.

Agent panel demo · sample data

Resolved by Atena

78%+6 pp

Time to reply

12s−41%

Open anomalies

32 under review

Average confidence

0,92stable

Conversations per hour peak 14:00 · 412

08:0014:0022:00

A conversation Atena resolves alone consumes no support rep hours.

É daí que vem a economia: não de cortar atendimento, mas de parar de pagar hora humana para responder pela quinquagésima vez onde está o saque.

No plano Growth, o modelo chega ao ponto de equilíbrio em 4 atendentes, com as premissas de volume, ticket e turno declaradas por escrito. Na demonstração, refazemos a conta com os seus números. Se não fechar, você vê antes de assinar.

Redo the maths with my numbers

What comes off the bill

Withdrawal status, sign-up, account questions, the small hours, the Sunday peak. Volume grows without the team growing at the same rate.

How usage is measured

1 credit is 1 full Atena answer. The other agents debit fixed weights from the same pool. Backoffice seats are included: there is no charge per support rep.

Usage is metered conversation by conversation

The meter adds up the credit of each reply, per agent and per cycle, and the total stays available for review at any time. There is no automatic warning of usage, of a percentage threshold or of depletion: watching the balance is up to the operator, and we do not promise what we have not built.

If the pool runs out

All agents pause, compliance included, and your team keeps working from the Backoffice. Nothing is cut mid-conversation: the reply in progress is finished before any pause. The regulatory duty of responsible gaming and anti-money-laundering remains yours, with or without credit.

Credits used demo · sample data

0 +6.2% vs month

Jan Feb

Usage, metered

In the demo, the panel runs on your volume and your rulebook.

Audit trail · case #064B17A2 every event recorded the moment it happens

recording

Event Actor Sev. Hash

Marketing blocked by responsible gamingwebchat.rg_age.decision Nêmesis audit 5fa4d1f6184a…
Answer sent to playeratena.message_sent Atena audit 13b597d9ea23…
Conversation escalated to a humanatendimento.handoff.created Atena audit 8199a22f17a2…
Player file openedatendimento.ficha.accessed Marcos R. audit c941dd5cba43…
Tax ID revealed to support repatendimento.cliente_cpf.revealed Marcos R. audit b95f227f0bf2…
Refused on the file by responsible gamingcliente.apolo_recommendation.blocked_responsible_gaming Marcos R. audit 2c7e918be004…

Chained hash · redacted PII · 5-year retention

5fa4d1f6184a174cc0caf3bebf21af53a00f93d76dfb…

When the regulator asks for the proof, it already exists.

Chega o pedido. O prazo é curto e o recorte é específico: o que foi dito a esta jogadora, quando, por quem e sob qual regra. Você não vai montar nada. A conversa lá em cima já está aqui embaixo, gravada no instante em que aconteceu.

Escrita só, na origem.
Evidence for audit.
No spreadsheet rebuild.
By session or period.
AES-256 and TLS 1.3.
Access by role.
Knowledge base and audit trail in Brazil.
Sensitive reads in the log.

Quem protocola na SPA/MF (Secretaria de Prêmios e Apostas do Ministério da Fazenda) é você. A prova já está montada.

Deposit limit applied VERIFIED PLAYER Active
KYC approved DOCUMENT VALIDATED Done
Withdrawal released R$ 500,00 · PIX Done
Conversation escalated to a human CASE #064B17A2 Done
Self-exclusion recorded 30 DAYS Active
Record written to the audit trail SESSION 9F2A Done

The proof recorded on the spot

Every decision enters the audit trail when it happens, with evidence ready for audit.

Demo · sample data. In the demo, the panel runs on your volume and your rulebook.

How the proof holds up when someone comes to check.

Não é uma promessa de imutabilidade. É um mecanismo que torna a alteração detectável e um prazo de retenção que a lei exige cumprir.

Neither the operator nor Cortex rewrites the past

The application only writes: it never changes, never deletes. Changes and deletions are blocked in the database itself, and any break in the hash chain shows up in the check. This is what is called a WORM audit trail. Disposal only happens at the end of the retention period, as the LGPD requires.

The report comes out ready for audit

By session, by player or by period, with the decision, the rule applied and the actor. No spreadsheet exports and no hunting for who handled it.

Where your data lives

Base e trilha ficam no Brasil. O modelo de linguagem processa nos Estados Unidos, com minimização dos dados antes do envio. Isso está declarado no nosso Aviso de IA. Você segue como controlador dos dados; a Cortex atua como operador.

How the data is protected

AES-256 at rest. TLS 1.3 in transit, with 1.2 as the accepted minimum. MFA (two-factor authentication) on panel access.

Viewing is not free

Cada leitura de dado sensível entra na trilha com nome e hora. Quem pode ver o quê é definido por papel e por operação, com RBAC (controle de acesso por papel), separado por cliente.

The rulebook is yours; applying it is ours

Responsible gaming, KYC and AML remain your operation's decision, under Law 14,790/2023 and the SPA/MF rules. Cortex applies the rulebook your team wrote, on every conversation, and records that it applied it.

What you are going to ask me.

Nine questions that come up in every sales conversation. Two of them have no comfortable answer, and they are here all the same.

Will the AI say something dumb to my player?

Antes de qualquer fala sair, três checagens rodam: a resposta nasce da base de conhecimento que você aprovou, o limite e o cap do contrato são conferidos, e o estado de jogo responsável é verificado. Fora da regra, ela não fala. Ela encaminha. Exceção de contrato a Atena não decide, e a tentativa fica registrada.

Demo · sample data.

Can you release my withdrawal now, outside the deadline?

Isso eu não decido. Exceção de contrato não sai por aqui. Vou levar o seu caso para uma pessoa da equipe, com o histórico inteiro.

Off the rule, she does not speak: she routes it on. The attempt is recorded in the trail.

How do I know the log cannot be altered?

A aplicação só escreve. Alteração e remoção são bloqueadas no banco e qualquer quebra do encadeamento de hash aparece na conferência. Nem o operador nem a Cortex reescrevem o passado. O descarte só acontece no fim do prazo de retenção, como a LGPD exige.

Audit trail · case #064B17A2 every event recorded the moment it happens

Event Actor Sev. Hash

Answer sent to playeratena.message_sent Atena audit 13b597d9ea23…
Conversation escalated to a humanatendimento.handoff.created Atena audit 8199a22f17a2…
Tax ID revealed to support repatendimento.cliente_cpf.revealed Marcos R. audit b95f227f0bf2…

Chained hash · redacted PII · 5-year retention

5fa4d1f6184a174cc0caf3bebf21af53a00f93d76dfb…

Does this replace my team?

Most of the repetitive work, yes: withdrawal status, sign-up, account questions, the small hours. What calls for judgement stays with people, and those people start receiving cases instead of conversations. Nobody here is going to tell you that you can operate without humans.

The six reasons that take the conversation to a human

Responsible gaming.
Emotional crisis.
Regulatory threat.
Resolution failure.
Explicit player request.
Unresolved limit in the conversation.

Where does my data go?

The knowledge base and the audit trail stay in Brazil. The language model processes in the United States, with minimisation before sending. Your operation remains the data controller; Cortex acts as the data processor. It is declared in our AI Notice.

Brazil

Knowledge base
Audit trail

Data minimisation before sending

United States

Language model

You remain the data controller; Cortex acts as the data processor.

What if the credit pool runs out?

All agents pause, compliance included, until top-up. No conversation is cut mid-way: whatever started is finished, and your team keeps working from the Backoffice, because human service consumes no credit. Usage is metered per cycle and per agent and stays available for review at any time; there is no automatic usage warning, so watching the balance is up to the operator. The legal duty of responsible gaming and anti-money-laundering remains the operator's, with or without credit.

In balance Agents active. Usage is metered per cycle and per agent.
Depleted Agents paused, with no exception for the compliance ones.
Topped up Agents come back. Human service never stopped at any point.

What pausesOs agentes, todos, inclusive os de compliance.

What continuesYour team serving through Backoffice. Nothing is cut off mid-conversation.

How long does implementation take?

De 15 a 30 dias úteis, conforme o plano, contados da entrega dos acessos e das integrações. Você não troca de WebChat nem de backoffice: a Cortex sobe sobre o que já roda. A entrada em operação tem 15 dias de hypercare acompanhado.

Live in 15 to 30 business days

Connect your operation We plug Atena into the WebChat and the backoffice you already run today.
Tune the agent Your house rules become behaviour: tone, limits, authority and knowledge base.
Go live with hypercare Entrada em operação acompanhada por 15 dias, com métrica de resolução e escalonamento à vista.

I already have a chat that answers.

Provavelmente responde bem. A pergunta é outra: quando pedirem, ele mostra o que respondeu, sob qual regra e quem viu o dado do jogador? Aqui, atender e registrar são o mesmo evento, não um relatório remontado depois.

How Cortex solves your problem

Pick what weighs most on your operation today.

“My queue does not move overnight.”

A Atena atende no mesmo minuto às 3h, com a mesma régua das 15h, e gera a mesma evidência. O que ela não resolve sobe para o seu time de plantão com o caso pronto, em vez de esperar o horário comercial. A cena desta página é uma dessas conversas, da primeira mensagem até o escalonamento.

Next step: send us your volume by time band and we will come back with where Atena fits and where the on-call team is still needed.
“My support cost grows along with my player base.”

A conversation Atena resolves alone consumes no support rep hours. Volume grows without the team growing at the same rate, and usage is measured in credits per answer, per cycle and per agent. The pricing block above carries the whole credit rule.

Next step: we redo the maths with your numbers before any proposal.
“I do not know what was said in the conversations nobody reviewed.”

Every conversation goes through the same rulebook and every decision is recorded the moment it happens. You do not sample: you have the whole audit trail, by session and by actor. The trail on this page shows the six events of the same session, each with its hash.

Next step: pick a case from your history and we will show how it would appear in the audit trail.
“When the regulator asks, my team drops everything to assemble the proof.”

A prova não é montada depois. Ela nasce junto com o atendimento. Relatório por sessão, por jogador ou por período, com a regra aplicada e quem decidiu. Quem protocola na SPA/MF continua sendo você.

Next step: bring an information request in the format your operation receives and we will check what the audit trail answers.
“My player repeats the whole story every time he talks to a human.”

The handoff carries the whole conversation, the player file and what Atena already tried. The support rep opens the case knowing where it stopped, and the player does not start over from zero. The handoff card on this page is the same case as the widget, with the same identifier.

Next step: we run the handoff with a conversation from your operation, from the widget to the Backoffice.