Book a demoDemoTalk to a specialist

Dashboard manual · Intelligence

Your own AI: connect your company's AI models (OpenAI, Anthropic, Google, Azure, Bedrock) and choose which one does each task

"IA própria" (Own AI) is the screen where the company connects its own AI models, using the API key from the provider it chose, decides which model handles each support task and monitors the health, usage and security of those models.

When it appears: Only appears on an Enterprise plan account with your own AI (or on a Custom plan that uses the company's own AI, where the texts replace Enterprise with the plan name). It does not appear on an account contracted through a reseller; without the plan, the direct address goes back to "Painel" (Dashboard).

Who can use it

RoleWhat you see and do
OwnerSees everything. With two-step verification on, connects, tests, pauses and changes the key of the models, changes the routing and the rules, and is the only one who changes contingency, recording, permissions, alerts and the cap.
AdminSees everything. Changes models, routing and rules only if the account has not turned on "Só donos mudam modelos e regras" (Only owners change models and rules), and also with two-step verification on; the other settings stay with the Owner. Exporting the change log is allowed.
AgentDoes not see the item in the menu; anyone who opens the direct address goes back to the "Painel" (Dashboard).

Screens

"IA própria" (Own AI)

The "IA própria" (Own AI) screen is used to activate the plan with the company's models, connect and test each model, choose the primary and the fallback for each task, and track the month's fair use, the health and the security of the data.

Address in the dashboard: /app/ia-propria · Who uses it: Owner, Admin

How to get there: From the menu, in the "Inteligência" (Intelligence) group, the "IA própria" (Own AI) item with the Enterprise badge. Each tab has its own address, so you can send the link to a specific tab.

Header and contracting

The shortcut to the guide and, while the plan is being contracted, the notice with the current step (sign, pay or connect).

Buttons and actions

"Guia da IA própria" (Own AI guide)
Opens the guide for "IA própria" (Own AI) in "Central de ajuda" (Help center).
"Assinar agora" (Subscribe now)
Appears only for whoever signs for the company: opens the "Condições" (Terms) document for electronic signature.
"Pedir reenvio" (Request resend)
For the other Owners: asks for the document to be resent to whoever signs.
"Ler antes" (Read first)
Opens the "Condições do Plano Enterprise" (Enterprise Plan Terms) with the contract values, to read before signing.
"Pagar agora" (Pay now)
After signing, opens the payment link for the first monthly payment.
"Pagar a implantação" (Pay the setup fee)
Opens the payment link for the setup fee, when there is one.
"Conectar o que falta" (Connect what's missing)
Once signed and paid, opens the "Conectar um modelo" (Connect a model) drawer already on the first missing type; if only a choice is missing, it leads to "Roteamento" (Routing).
  • Contracting notices: "Assine as Condições do Plano Enterprise para ativar" (Sign the Enterprise Plan Terms to activate), "Falta pagar a primeira mensalidade" (The first monthly payment is still due) and "Enterprise ainda não ativo: as conversas vão para a sua equipe" (Enterprise not active yet: conversations go to your team).

Top of the page (hero)

The plan name, the activation status and, once active, the month's fair use.

  • The Enterprise · "IA própria" (Own AI) badge and the title "A inteligência é sua. A operação comercial roda aqui." (The intelligence is yours. Your sales operation runs here.)
  • Pills: "‹n› modelos conectados" (‹n› models connected) (or "Nenhum modelo conectado" (No model connected)), "‹n› tarefas roteadas" (‹n› tasks routed) and the contingency ("Contingência: reserva, IA da plataforma, equipe" (Contingency: fallback, platform AI, team) or "Contingência: reserva, depois equipe" (Contingency: fallback, then team)).
  • Before activation: the pills "Enterprise ainda não ativo" (Enterprise not active yet) and "Conversas com a sua equipe" (Conversations with your team), and "‹n› modelos conectados, faltam ‹n›" (‹n› models connected, ‹n› missing).
  • The "Ativação do Enterprise" (Enterprise activation) panel: pending, "‹n› de ‹total› prontos" (‹n› of ‹total› ready), the "Modelos obrigatórios prontos" (Required models ready) meter with the types (conversation, search, transcription, speech) and the sentence about what is missing. When contracting, the steps include signature and payment.
  • The "Uso justo do mês" (Fair use this month) panel (once active): the badge on pace, above pace or cap reached, ‹n› of ‹n› messages, the meter with the 80% alert mark and the forecast "No ritmo atual, o mês fecha em ‹n› mensagens" (At the current pace, the month closes at ‹n› messages).
  • With no open cycle: "O primeiro ciclo começa quando a assinatura fica ativa." (The first cycle starts when the subscription becomes active.)

Notices below the header

What blocks the change or the activation, with the shortcut to resolve it.

Buttons and actions

"Abrir Configurações → Segurança" (Open Settings → Security)
Appears when the page is read-only because two-step verification is missing. Leads to the "Segurança" (Security) section of "Configurações" (Settings).
"Conectar o que falta" (Connect what's missing)
In the "Enterprise ainda não ativo: as conversas vão para a sua equipe" (Enterprise not active yet: conversations go to your team) notice, opens the drawer on the first missing model.
"Pedir ajuda da implantação" (Ask for setup help)
Opens an email to support with the subject about the search and voice setup.
  • Read-only notice (padlock): the sentence explaining why you can read but not change.
  • Once active, if the persona started replying to audio with audio without a speech model: "A sua persona responde áudio com áudio, mas não há modelo de fala: a IA responde por texto." (Your persona replies to audio with audio, but there is no speech model: the AI replies by text.)

Tabs

Five tabs; the chosen one stays in the page address.

Buttons and actions

"Visão geral" (Overview)
Activation, orchestration, indicators and model health.
"Modelos" (Models)
The connected models, with the count next to the tab name.
"Roteamento" (Routing)
Who does each task, rules by lead type and what happens if the models fail.
"Consumo" (Usage)
The month's messages, per day, per model and the fair-use alerts.
"Segurança e dados" (Security and data)
Permissions, what goes out to the models, recording and the change log.
  • Keyboard: the left and right arrows switch tabs; Home goes to the first and End to the last.

"Visão geral" (Overview) · "Para ativar o Enterprise" (To activate Enterprise)

The card with each required model and what is missing. It disappears when the account activates.

Buttons and actions

"Ler o guia" (Read the guide)
Opens the guide at the activation section.
"Conectar" (Connect)
Opens the "Conectar um modelo" (Connect a model) drawer already on that step's type.
"Escolher em Roteamento" (Choose in Routing)
For an already connected model that has not been chosen yet: leads to the "Roteamento" (Routing) tab.
"PDF assinado" (Signed PDF)
When contracting, downloads the PDF of the signed Terms.
  • Steps: "Modelo de conversa" (Conversation model), "Modelo de busca (embeddings)" (Search model (embeddings)), "Modelo de transcrição" (Transcription model) and "Modelo de fala" (Speech model).
  • Badge for each step: ready, not needed (speech, when the persona replies by text) or "Lendo a base: ‹n›%" (Reading the knowledge base: ‹n›%).
  • When contracting, the signature and payment steps come before the models, with the badges signed, paid and connected.
  • Footer showing who offers each type: search by OpenAI, Google, Azure OpenAI, Amazon Bedrock or your own server; transcription by OpenAI, Google, Azure OpenAI or your own server; speech by OpenAI, Azure OpenAI or your own server. Anthropic has no search or audio models.

"Visão geral" (Overview) · "Como a sua IA trabalha aqui dentro" (How your AI works in here)

The orchestration diagram: your conversation models on the left, the platform core in the middle and the channels on the right.

  • Up to three models appear; the others become "+‹n› modelos, Ver em Modelos" (+‹n› models, View in Models). The main conversation model carries "Principal" (Primary).
  • With no model: the node "A sua equipe, Enterprise ainda não ativo" (Your team, Enterprise not active yet).
  • In the core: "Intenção" (Intent), "Qualificação BANT" (BANT qualification), "Perfil ideal" (Ideal profile), "Conferência da resposta" (Response check), "Próxima melhor ação" (Next best action) and "Transbordos" (Handoffs).
  • On the right: WhatsApp and Instagram ("Atendimento" (Service)), "Agenda" (Calendar) ("Visitas e reuniões" (Visits and meetings)) and "Funil e CRM" (Pipeline and CRM) ("Até a venda" (Through to the sale)).

"Visão geral" (Overview) · "Indicadores" (Indicators)

Four numbers from the last 7 days.

  • "Tempo de resposta" (Response time): 7-day median, with the difference from the previous period.
  • "Aprovadas na conferência" (Approved in the check): the approved percentage, with how many were rewritten and how many became a safe reply.
  • "Falhas dos seus modelos" (Failures of your models): the percentage and how many were covered by the fallback.
  • "Custo estimado no seu provedor" (Estimated cost at your provider): based on the prices you entered; without a price, enter the prices in "Modelos" (Models).

"Visão geral" (Overview) · "Saúde dos modelos agora" (Model health now)

The last 24 hours of each conversation model.

Buttons and actions

"Ver modelos" (View models)
Leads to the "Modelos" (Models) tab.
"Conectar um modelo" (Connect a model)
With no connected model, leads to the "Modelos" (Models) tab to connect the first one.
  • Each row: provider, name, status (live, slow, down or paused), tasks, p50, success and calls.
  • Red notice "O ‹modelo› está fora do ar: a reserva e a contingência cobrem as tarefas dele" (‹model› is down: the fallback and the contingency cover its tasks).
  • Yellow notice when a model is slower than the median of the last 7 days, saying whether this delays the reply to the lead.

"Modelos" (Models)

One card per conversation model, the card to connect another one and the "Busca e voz" (Search and voice) grid.

Buttons and actions

"Testar agora" (Test now)
Opens the dialog to test the model again.
"Trocar chave" (Change key)
Opens the dialog to paste a new key (cloud providers).
"Editar" (Edit)
On an own-server model, opens the drawer to change the address, model, name or price.
"Pausar" (Pause)
Takes the model off the tasks without deleting it; the button becomes "Retomar" (Resume).
"Retomar" (Resume)
Puts the paused model back on the tasks.
"Conectar um modelo" (Connect a model)
Card at the end of the grid: opens the drawer on the conversation type. Lists OpenAI, Anthropic, Google, Azure, Bedrock or your own server.
"Ler o guia" (Read the guide)
With no model, opens the guide at the connecting section.
  • Model card: provider, name, the source line (masked key, address or region, and since when) and the status.
  • Numbers: p50, success and the price per million (or no price; on your own server, your own server).
  • Test badges: "Português" (Portuguese), "JSON estruturado" (Structured JSON), "JSON com correção" (JSON with correction) or "JSON reprovado" (JSON failed), the proven context and the simultaneous calls.
  • Search and voice: cards with the "Busca" (Search), "Transcrição" (Transcription) or "Fala" (Speech) badge; search shows searches, positions and price; transcription and speech show time, success and price per minute. A model with no task shows as having no task.
  • With no search, transcription or speech model: "Nenhum modelo de busca, de transcrição ou de fala" (No search, transcription or speech model), with "Conectar o que falta" (Connect what's missing).

"Testar ‹modelo› de novo" (Test ‹model› again) dialog

Runs the model's checks again.

Buttons and actions

"Testar agora" (Test now)
Runs the checks. The test cost shows in the footer beforehand; if you need to wait, the sentence counts the seconds and the button unlocks at the end.
"Cancelar" (Cancel)
Closes without testing; once the test has started, the button becomes "Fechar" (Close).
  • Conversation: "A chave conecta" (The key connects), "Responde em português do Brasil" (Replies in Brazilian Portuguese), "Devolve JSON no formato pedido" (Returns JSON in the requested format), "Aguenta uma conversa longa com a base" (Handles a long conversation with the knowledge base) and "Atende chamadas ao mesmo tempo" (Handles simultaneous calls).
  • If the key fails: "A chave não conectou. Confira a chave e o modelo e teste de novo." (The key didn't connect. Check the key and the model and test again.)

"Trocar a chave d‹o modelo›" (Change the key of ‹the model›) dialog

Pastes a new key, which is tested with the provider before it takes effect.

Fields and filters

"Chave de API" (API key)required
The new key ("Cole a chave nova" (Paste the new key)). It is stored encrypted; once saved, only the beginning and the end appear. On Bedrock, it accepts the Bedrock key or the access key pair.

Buttons and actions

"Trocar chave" (Change key)
Tests and saves the new key; shows "Chave trocada" (Key changed). If the provider refuses it, the reason appears in the field itself.
"Cancelar" (Cancel)
Closes and clears what was typed.

"Conectar um modelo" (Connect a model) drawer

Three steps: choose what it is for and the provider, fill in the key and the model, test and save. The same drawer opens from every "Conectar" (Connect) button on the page.

Fields and filters

"Chave de API" (API key)required
The provider's key. It is stored encrypted and only used by your tasks.
"Modelo" (Model)
List of the provider's suggested models; "Outro modelo…" (Other model…) opens the "Nome do modelo" (Model name) field.
"Nome do modelo" (Model name)
The exact name of the model, when it is not on the list.
"Nome da implantação" (Deployment name)
On Azure OpenAI, the deployment name in place of the model.
"ID do modelo" (Model ID)
On Amazon Bedrock, the model ID.
"Endereço" (Endpoint)required
On your own server, the address compatible with the OpenAI API.
"Endereço do recurso" (Resource endpoint)required
On Azure OpenAI, the resource address.
"Região da AWS" (AWS region)required
On Amazon Bedrock, the account's region.
"Seu preço por milhão de tokens (opcional)" (Your price per million tokens (optional))
Input and output price. Only used to estimate the cost on the screens; billing is done by your provider.
"Seu preço por minuto de áudio (opcional)" (Your price per minute of audio (optional))
On the transcription model, to estimate the cost.
"Seu preço por minuto de fala (opcional)" (Your price per minute of speech (optional))
On the speech model, to estimate the cost.
"Nome nesta conta" (Name in this account)
How the model appears on the screens and in the "Roteamento" (Routing) table.

Buttons and actions

"Conversa e tarefas" (Conversation and tasks)
Step 1, "Para que é este modelo" (What is this model for): replies, qualifies, summarizes.
"Busca na base" (Knowledge base search)
Step 1: knowledge base embeddings model.
"Transcrição do áudio" (Audio transcription)
Step 1: listens to the audio the lead sends.
"Resposta falada" (Spoken reply)
Step 1: replies to audio with audio.
OpenAI
"Provedor" (Provider). Each card says what it offers for the chosen type or why it does not.
Anthropic
Conversation-only provider: it has no search or audio models.
Google
Provider for conversation, search and transcription; speech not yet.
Azure OpenAI
Provider under the company's own subscription, for all types.
Amazon Bedrock
Provider in the company's AWS account, for conversation and search; transcription and speech are left out.
"Seu servidor" (Your server)
The company's server, compatible with the OpenAI API.
"Colar outra chave" (Paste another key)
For search and voice, replaces "A mesma d‹o modelo›" (The same as ‹the model›) with a new key.
"Continuar" (Continue)
Goes from step 1 to step 2.
"Voltar" (Back)
Goes back one step; on the first, the button is "Cancelar" (Cancel) and closes the drawer.
"Testar o modelo" (Test the model)
Runs the checks for the chosen type. The test cost shows in the footer beforehand.
"Testar de novo" (Test again)
When the test fails, runs it again after you fix the problem.
"Salvar modelo" (Save model)
Saves the approved model; shows "Modelo conectado" (Model connected). No task changes until you choose in "Roteamento" (Routing).
"Salvar" (Save)
When editing: name and price save directly; a changed address or model is tested again first.
  • When editing, the title becomes "Editar ‹o modelo›" (Edit ‹the model›) and the old address appears as "Antes: ‹endereço›" (Before: ‹address›); changing the address asks for the key again.
  • Results: "Pronto para usar" (Ready to use) (or "Pronto para a busca" (Ready for search), "Pronto para transcrever" (Ready to transcribe), "Pronto para falar" (Ready to speak)), "Pronto, com uma limitação" (Ready, with one limitation) (no JSON in the requested format) and "A chave não conectou" (The key didn't connect).
  • Search: "Gera o vetor de um trecho em português" (Generates the vector of a passage in Portuguese), "Acha o trecho certo numa busca de exemplo" (Finds the right passage in a sample search) and "Responde rápido o bastante para a conversa e a leitura de um lote de trechos" (Responds fast enough for the conversation and for reading a batch of passages), with the "Busca de exemplo" (Sample search).
  • "Transcrição" (Transcription): a sample audio in Portuguese, the word accuracy (with "Ouvir" (Listen)) and the WhatsApp audio. "Fala" (Speech): a short utterance in Portuguese, in the format WhatsApp accepts, with "Ouvir a fala" (Listen to the speech) in the persona's voice.
  • On a search or voice failure: "Veja acima o que falhou. Ele não é salvo." (See above what failed. It is not saved.)

"Roteamento" (Routing) · "Quais dos seus modelos leem imagem" (Which of your models read images)

The models that work for image reading, verified in the test with a sample photo.

  • Each model with the badge reads images, text only, search model, transcription model, speech model or test again.

"Roteamento" (Routing) · "Quem faz cada tarefa" (Who does each task)

The table that chooses the primary model and the fallback, and shows each task's timeout. Each choice saves instantly.

Fields and filters

"Tarefa" (Task)
The name and description of the task, with the own-route badge when it does not follow the conversation.
"Modelo principal" (Primary model)
Who handles the task. Options: your allowed models, "IA da plataforma (incluída)" (Platform AI (included)) or, for intent, "Decisor da plataforma (incluído)" (Platform decider (included)).
"Reserva" (Fallback)
Who takes over if the primary fails or runs over time; "Nenhuma" (None) leaves the task without a fallback. Only unlocks once a primary is chosen.
"Tempo-limite" (Timeout)
The maximum time of the call before the fallback takes over; the conversation's or the task's default when inherited.

Buttons and actions

"Seguir a conversa (‹modelo›)" (Follow the conversation (‹model›))
On tasks that can follow the conversation: uses the primary, the fallback and the timeout of "Conversa com o lead" (Conversation with the lead).
"Conectar" (Connect)
On a search, transcription, speech or image row with no model: opens the drawer on that row's type.
"Escolha o modelo de busca" (Choose the search model)
On the search row: choosing a model opens the knowledge base reading dialog.
"Mostrar todas" (Show all)
Shows the groups "Seguem a conversa" (Follow the conversation), "Bastidores" (Behind the scenes) and "Funcionam por regras, sem modelo de IA" (Run on rules, no AI model); the button becomes "Mostrar menos" (Show less).
"Usar como principal" (Use as primary)
Confirms in the "Usar ‹modelo› como principal?" (Use ‹model› as primary?) dialog when the chosen model was the task's fallback, which is then left without a fallback.
"Cancelar" (Cancel)
Closes the confirmation without changing anything.
  • At the top, the estimate "‹valor› / mês estimado" (‹value› / month estimated), based on the prices entered in "Modelos" (Models).
  • Search, voice and image: each row only accepts models of its type. With no model it shows "falta um modelo de busca" (a search model is missing) (or the transcription, speech or vision equivalent). The search fallback is always "Busca por palavras, sem modelo" (Keyword search, no model); the speech fallback can be "Nenhuma (responde por texto)" (None (replies by text)); the image one, "Nenhuma (a equipe vê a foto)" (None (the team sees the photo)).
  • "Bastidores" (Behind the scenes): tasks that do not talk to the lead and run in the background to prepare the account.
  • "Funcionam por regras, sem modelo de IA" (Run on rules, no AI model): "Por regras, sem custo" (By rules, no cost); they are listed there to show everything that serves the lead.
  • Footer with how many tasks follow the conversation model and how many have their own route.

"Roteamento" (Routing) · "Leitura da base (troca do modelo de busca)" (Knowledge base reading (search model switch))

Switching the search model makes the knowledge base be read again; the screen shows the estimate beforehand.

Buttons and actions

"Trocar e ler a base de novo" (Switch and re-read the knowledge base)
Confirms the switch and starts the reading; search keeps using the current model until the new base is complete.
"Voltar agora" (Switch back now)
Goes back to a search model used in the last 7 days: the switch is immediate and free of cost.
"Manter ‹modelo›" (Keep ‹model›)
Closes without switching; on the first choice, the button is "Agora não" (Not now).
"Cancelar a troca" (Cancel the switch)
During the reading, gives up and keeps the current model.
"Retomar de onde parou" (Resume where it left off)
If the reading stopped because of a provider failure, continues from where it was.
"Desistir da troca" (Give up the switch)
If the reading stopped, gives up and keeps the current model.
"Entendi" (Got it)
Closes the reading-complete notice.
  • The dialog shows the passages and sources, the tokens and the estimated cost at your provider, and how long it takes.
  • During the reading: "Lendo a base de novo com ‹modelo›, ‹n› de ‹n› trechos" (Re-reading the knowledge base with ‹model›, ‹n› of ‹n› passages), with a progress bar.
  • Today's base is kept for 7 days so you can go back at no cost.

"Roteamento" (Routing) · "Regras por tipo de lead" (Rules by lead type)

Rules that switch the conversation model for certain leads; they apply on top of the table, from top to bottom, and the first one that matches decides.

Fields and filters

"Quando a regra vale" (When the rule applies)
The condition: ideal profile, if there is an objection or if the card has the tag.
"Nota mínima do perfil ideal" (Minimum ideal-profile score)
With the ideal profile condition: ‹n› or more.
"Objeção" (Objection)
With the objection condition: the type of objection (or the saved combination).
"Etiqueta" (Tag)
With the tag condition: which card tag.
"Modelo da regra" (Rule's model)
The conversation model that serves the rule's leads.

Buttons and actions

"Nova regra" (New rule)
Adds a rule. With no connected model, it is dimmed with the hint "Conecte um modelo para criar regras" (Connect a model to create rules).
"Arrastar para mudar a ordem" (Drag to reorder)
Handle to drag the rule. By keyboard: space picks up, up and down arrows move, space drops and Esc cancels.
"Remover regra" (Remove rule)
Opens the "Remover esta regra?" (Remove this rule?) confirmation.
"Remover" (Remove)
Confirms: the rule's conversations go back to following the table.
  • With no rules: "Nenhuma regra. Todas as conversas seguem a tabela acima." (No rules. All conversations follow the table above.)

"Roteamento" (Routing) · "Se os seus modelos falharem" (If your models fail)

The order a message follows until it is answered; the lead is never left without a reply.

Buttons and actions

"A IA da plataforma" (The platform's AI)Owner
The 2nd step is an on/off switch: when on, the platform AI answers only while your models are down, with pay-per-use cost in the month's report.
  • 1st: the task's fallback: always on.
  • 3rd: your team: the conversation goes to the queue with the notice that the models are down.
  • On search and voice: search falls back to keyword search (the platform AI does not come in); transcription falls back to the fallback, the platform AI (if on) and the team, who receives the audio; speech falls back to the fallback, the platform AI (if on) and, lastly, the text reply.

"Consumo" (Usage)

The cycle's messages, the forecast, the usage per model and the fair-use alerts.

Fields and filters

"Teto de excedente" (Overage cap)
How far the account can go over the included messages: "Sem teto" (No cap) or ‹n› thousand. Above it, neither the platform AI nor your models answer and the conversation goes to the team.

Buttons and actions

"Este mês" (This month)
Shows the "Mensagens por dia" (Messages per day) chart for the current month.
"Mês passado" (Last month)
Shows the previous month; with no data, "Sem mensagens no mês passado" (No messages last month).
"Aviso aos 80%" (Alert at 80%)Owner
Switch for "Aos 80% das mensagens" (At 80% of messages): email and bell for the account Owners.
"Aviso ao passar" (Alert when exceeded)Owner
Switch for "Ao passar das incluídas" (When exceeding the included ones): warns when the overage starts to enter the month's invoice.
  • "Mensagens no mês" (Messages this month): ‹n› of ‹n› included · ‹n›%.
  • "Previsão de fechamento" (Closing forecast): within fair use, above fair use or cap reached.
  • "Leads e transbordos" (Leads and handoffs): ‹n› leads, ‹n› hot, ‹n› handoffs, at no charge.
  • "Mensagens por dia" (Messages per day): a message is each lead utterance answered by the AI, on any channel. Legend: past days, today and forecast.
  • "Por modelo" (By model): table with "Modelo" (Model), "Entrada" (Input), "Saída" (Output) and "Custo" (Cost), in tokens and cost estimated from your prices.
  • When the contingency was used, the lines for what ran on the platform AI (messages, background tasks and audios) with the amount.
  • Cap reached: "O teto de excedente do mês foi atingido; as conversas vão para a equipe até o mês virar ou até o teto mudar." (The month's overage cap has been reached; conversations go to the team until the month turns over or the cap changes.)

"Segurança e dados" (Security and data)

Page permissions, what goes out to the models, recording and the change log.

Fields and filters

"Gravação" (Recording)
What is stored from the calls to the models: the text of the calls or only numbers (tokens and time).

Buttons and actions

"Só donos" (Owners only)Owner
Switch for "Só donos mudam modelos e regras" (Only owners change models and rules): when on, Admins can see but not edit.
"Duas etapas" (Two steps)Owner
Switch for "Exigir verificação em duas etapas para mudar esta página" (Require two-step verification to change this page).
"Rotação" (Rotation)Owner
Switch for "Lembrar de trocar as chaves, a cada 90 dias" (Remind to change keys, every 90 days).
"Exportar" (Export)
Downloads the change log. Allowed for anyone who can read the page.
"Baixar PDF assinado" (Download signed PDF)
On the "Condições do Plano Enterprise · ‹número›" (Enterprise Plan Terms · ‹number›) card, downloads the signed contract.
  • "Chaves e acesso" (Keys and access): keys are stored encrypted and never appear in full again.
  • "O que sai para os seus modelos" (What goes out to your models): "Vai" (Goes) (recent conversation, company context, knowledge base passages and, in the background, dashboard numbers with no name or contact) and "Não vai" (Doesn't go) (the lead's phone, email and document, replaced by placeholders).
  • Audio note: the lead's audio goes as is to the transcription model and the reply text to the speech model; in search, the question goes with personal data masked.
  • "Trilha de alterações" (Change log): everything that changed on the page, who changed it and when; with no items, "Nenhuma alteração ainda" (No changes yet).

Screen states

Loading
"Carregando a IA própria…" (Loading your own AI…), with gray blocks in place of the content.
Nothing here yet
With no model, the "Modelos" (Models) tab shows "Conecte o primeiro modelo" (Connect the first model), explaining that conversations go to the team until the models are connected and tested, with "Conectar um modelo" (Connect a model) and "Ler o guia" (Read the guide).
If something goes wrong
"Não conseguimos carregar a IA própria agora" (We couldn't load your own AI right now), with the "Tentar de novo" (Try again) button.
Plan not active yet
The pills "Enterprise ainda não ativo" (Enterprise not active yet) and "Conversas com a sua equipe" (Conversations with your team) and the yellow notice: the AI answers no lead and each conversation enters the team queue with the reason.
Account without permission to change
The padlock notice explaining the reason; the change buttons are dimmed.
Overage cap reached
The cap reached badge at the top and the notice in the "Consumo" (Usage) tab to raise or remove the cap.

Step by step

How to connect an OpenAI, Anthropic or Google model with the API key

  1. Turn on two-step verification in "Configurações" (Settings) → "Segurança" (Security), if it is not on yet.
  2. In "IA própria" (Own AI), "Modelos" (Models) tab, click "Conectar um modelo" (Connect a model).
  3. Choose "Conversa e tarefas" (Conversation and tasks) and the provider, and click "Continuar" (Continue).
  4. Paste the "Chave de API" (API key), choose the "Modelo" (Model) and, if you like, enter the price and the "Nome nesta conta" (Name in this account).
  5. Click "Testar o modelo" (Test the model) and, once approved, "Salvar modelo" (Save model).

How to activate the Enterprise plan with your own AI

  1. In "Visão geral" (Overview), follow the "Para ativar o Enterprise" (To activate Enterprise) card from top to bottom.
  2. When contracting, use "Assinar agora" (Subscribe now) and then "Pagar agora" (Pay now).
  3. Click "Conectar" (Connect) on each missing model (conversation, search, transcription and, if the persona speaks, speech).
  4. For an already connected model, use "Escolher em Roteamento" (Choose in Routing) and choose it on the task row.

How to choose which model answers each task, and the fallback

  1. Open the "Roteamento" (Routing) tab.
  2. In the "Quem faz cada tarefa" (Who does each task) table, choose the task's "Modelo principal" (Primary model).
  3. Choose the "Reserva" (Fallback), which takes over if the primary fails or runs over time.
  4. Click "Mostrar todas" (Show all) to see the tasks that follow the conversation and the background ones.

How to change a model's API key

  1. On the "Modelos" (Models) tab, click "Trocar chave" (Change key) on the model's card.
  2. Paste the new key into the "Chave de API" (API key) field.
  3. Click "Trocar chave" (Change key): it is tested with the provider and takes effect immediately.

How to change the knowledge base's search model

  1. On the "Roteamento" (Routing) tab, on the search row, choose the new model.
  2. Read the estimate of passages, tokens and time in the dialog.
  3. Click "Trocar e ler a base de novo" (Switch and re-read the knowledge base); search keeps using the current model until the reading finishes.
  4. If the reading stops, use "Retomar de onde parou" (Resume where it left off) or "Desistir da troca" (Give up the switch).

How to use a different model for ideal-profile leads

  1. On the "Roteamento" (Routing) tab, in the "Regras por tipo de lead" (Rules by lead type) card, click "Nova regra" (New rule).
  2. Under "Quando" (When), choose the ideal profile and the minimum score.
  3. Choose the model that will serve these leads.
  4. Drag the rules to put the most important one on top: the first match decides.

How to set a cap on overage messages

  1. Open the "Consumo" (Usage) tab.
  2. Under "Avisos do uso justo" (Fair-use alerts), choose the "Teto de excedente" (Overage cap) (or "Sem teto" (No cap)).
  3. Turn on "Aviso aos 80%" (Alert at 80%) and "Aviso ao passar" (Alert when exceeded) to be warned in advance (Owner only).

Frequently asked questions

Which AI providers can I connect?

OpenAI, Anthropic, Google, Azure OpenAI, Amazon Bedrock, or a server from your company that is compatible with the OpenAI API. Each model type (conversation, search, transcription, and speech) has the providers that offer it.

Why can't the agent see "IA própria" (Own AI)?

The page is for Owners and Admins. For the agent, the item does not appear in the menu, and the direct address goes back to the "Painel" (Dashboard).

Does saving a model already change how conversations are handled?

No. The saved model appears in "Roteamento" (Routing), but no task changes until you choose it as the primary or fallback model.

What is the difference between "Pausar" (Pause) and changing the model in "Roteamento" (Routing)?

"Pausar" (Pause) removes the model from all tasks at once, without deleting anything, and "Retomar" (Resume) brings it back. In "Roteamento" (Routing) you change the model of a specific task.

Does "Testar o modelo" (Test the model) cost anything?

The test makes real calls to your provider. The test cost appears in the footer before you click "Testar" (Test), and the provider is the one that charges for it.

What happens to the key after it is saved?

It is stored encrypted and never appears in full again: the screen shows only the beginning and the end. To change it, use "Trocar chave" (Change key).

Can I send the link to a specific tab?

Yes. The open tab is part of the page address, so the link opens directly on "Modelos" (Models), "Roteamento" (Routing), "Consumo" (Usage), or "Segurança e dados" (Security and data).

What is the change log?

It is the record of everything that changed on the page, including who changed it and when. It can be exported by anyone who can read the page.

Help center guide

"IA própria" (Own AI)

Connect your company's models, choose which one handles each task, and track health, usage, and security. The commercial operation keeps running here.

On the Enterprise plan, the intelligence belongs to your company: you connect the models you already use or sign up for, choose which one handles each task, and the platform takes care of the conversation on WhatsApp and Instagram, qualification, the pipeline, the calendar, and the handoff. Every agent that uses an AI model for your account uses your models — the ones that talk to the lead and the ones that work behind the scenes. Leads, qualifications, and handoffs are not charged on your plan; model usage is billed by your provider, directly on your account with them.

The screen, part by part

On screen: "Visão geral" (Overview) of "IA própria" (Own AI): the fair use for the month at the top, the orchestration diagram, and the indicators. The "Visão geral" (Overview) answers, at a glance, whether your AI is healthy and how much of the month has been used.

  1. What is turned on: How many models are connected, how many tasks can go to your models, and what happens if they fail.
  2. Fair use for the month: The messages answered by the AI in the month, how many are included, and where the month will close at the current pace. At 80% you get a warning.
  3. How your AI works in here: Your models on one side, the channels on the other, and in the middle the orchestration: company context, knowledge base, lead memory, reply check, and rules.
  4. The four numbers:
"Saúde dos modelos agora" (Model health now)
For each model: whether it is "no ar" (online), "lento" (slow), "fora do ar" (offline), or "pausado" (paused), which tasks it works on, and the time (p50), success rate, and calls from the last 24 hours.
"Lento" (Slow)
In the last hour, the model took more than double its own median from the last 7 days. The screen says which of its tasks delay the reply to the lead and how long until the fallback takes over.
"Fora do ar" (Offline)
The automatic hourly test could not reach the model twice in a row. The fallback and the contingency cover its tasks, and the Owners receive a warning.

Activating Enterprise: sign, pay, and connect

Enterprise turns on when three things are ready: the "Condições do Plano Enterprise" (Enterprise Plan Terms) signed, the first monthly payment paid, and your company's models connected. On Enterprise, everything the AI does for your account runs on your models — including reading the knowledge base and listening to audio messages — so a conversation model and the search, transcription, and speech ones come in, connected, tested, and chosen in "Roteamento" (Routing). While something is missing, a notice at the top of the page says what it is, and the "Para ativar o Enterprise" (To activate Enterprise) card shows the steps in order, with what has already been done checked off — including the models you connected before signing.

  1. Sign the Enterprise Plan Terms: The document arrives by electronic signature for the person who signs for your company, with a summary of what was agreed: tier, monthly fee, included messages, and commitment period. Only that person sees "Assinar agora" (Subscribe now); the other Owners see who the document was sent to and can "Pedir reenvio" (Request resend). "Ler antes" (Read first) opens the same text here in the dashboard. The link is valid for 30 days.
  2. Pay the first monthly fee: The link arrives by email to the account Owners as soon as both parties have signed — before that, the step appears locked, with "depois da assinatura" (after signing). It covers the setup and the first monthly payment. The 7 days to withdraw count from the payment, with a full refund.
  3. Connect search: And the conversation model: knowledge base search only counts when it is ready. If it is still missing, the button opens the conversation step first.
  4. Connect transcription: The lead's audio becomes text on your model.
  5. Connect speech: It only appears if your persona replies to audio with audio; without that, the AI replies by text.

On screen: "IA própria" (Own AI) during signup: the notice to sign the Terms at the top, the hero with "1 de 4 prontos" (1 of 4 ready), and the "Para ativar o Enterprise" (To activate Enterprise) card with sign, pay, search already connected, and transcription.

  1. What is missing now: One notice at a time: "Assinar agora" (Subscribe now) and "Ler antes" (Read first) (or "Pedir reenvio" (Request resend), for the other Owners), then "Pagar agora" (Pay now), then "Conectar o que falta" (Connect what's missing).
  2. Enterprise activation: How many steps are done: signature, payment, and the models.
  3. To activate Enterprise: The steps in order. Payment stays locked until both signatures are in; models connected before signing already appear checked off.
Who has each type
Search: OpenAI, Google, Azure OpenAI, Amazon Bedrock, or your own server. Transcription: OpenAI, Google, Azure OpenAI, or your own server. Speech: OpenAI, Azure OpenAI, or your own server. Anthropic has no search or audio models.
The same key
If you have already connected an OpenAI conversation model (or one from the same Azure resource), the drawer offers "A mesma do …" (The same as …): the stored key works for search and audio, with no need to paste it again. It never goes to another provider or another endpoint.

"Conectar um modelo" (Connect a model)

  1. Open "Modelos" (Models) and click "Conectar um modelo" (Connect a model): Choose what the model is for — "Conversa e tarefas" (Conversation and tasks), "Busca na base" (Knowledge base search), "Transcrição do áudio" (Audio transcription), or "Resposta falada" (Spoken reply) — and the provider: OpenAI, Anthropic, Google, Azure OpenAI, Amazon Bedrock, or your own server (compatible with the OpenAI API). A provider that does not have that type appears grayed out, with the reason why.
  2. Paste the key (or use the same one) and choose the model: The key is encrypted and, once saved, only the beginning and the end are shown. For Azure, enter the "Endereço do recurso" (Resource endpoint) and the "Nome da implantação" (Deployment name); for Bedrock, the "Região" (Region) and the "ID do modelo" (Model ID); for your own server, the "Endereço" (Endpoint) (https only). The price is optional — per million tokens for conversation and search, per minute for transcription and speech — and is only used to estimate cost on the screens.
  3. Test the model: For conversation we check five things: the key connects, it replies in Brazilian Portuguese, it returns JSON in the requested format, it handles a long conversation with the knowledge base, and it handles simultaneous calls (about 10 thousand tokens from your account). For search: whether it generates the vector for a passage, finds the right passage in a sample search, responds quickly, and reads a batch of 100 passages (about 3 thousand tokens). For transcription: a sample audio recorded on WhatsApp (about 70 seconds of audio), with the text it wrote and how many words it got wrong. For speech: a short utterance in your persona's voice (about 10 seconds of speech), which you can listen to.
  4. Save: It appears in the "Roteamento" (Routing) table. No task changes until you choose.

On screen: The "Conectar um modelo" (Connect a model) drawer at the key step, with the test cost notice in the footer.

  1. The key: Encrypted when saved. Nobody on the team sees the full key again — not even you.
  2. The cost of the test: The test really calls your model. If you run it many times in a row, the screen asks you to wait a few seconds.
"JSON com correção" (JSON with correction)
The model got the format right on the second attempt. It works, just a little slower on tasks that ask for JSON.
"Contexto" (Context)
How much the test PROVED the model can handle (for example, "8 mil de contexto" (8 thousand context)), not the window advertised by the provider.
"Simultâneas" (Simultaneous)
How many of the 20 simultaneous calls the model answered. Fewer than 20 is highlighted.

On screen: The cards for the connected models, with time, success rate, price, and the tested capabilities.

  1. Each model: Time (p50) and success rate for the last 24 hours, the price you entered, and what the test proved. "Testar agora" (Test now) repeats the test; "Trocar chave" (Change key) tests the new key before swapping; "Pausar" (Pause) takes the model off the tasks without deleting anything.

On screen: The drawer for connecting a search model at the end of the test, with the checks passed, the sample search, and the notice that it is ready for search.

  1. What search needs: The five search checks, with the result of each one.
  2. The sample search: A lead's question and the passage the model found. For transcription, the audio appears so you can listen, along with what the model wrote; for speech, the utterance in the persona's voice.

On screen: The "Busca e voz" (Search and voice) grid on the "Modelos" (Models) tab, with the search, transcription, and speech models.

  1. "Busca e voz" (Search and voice): Each one only appears on its own type's row in "Roteamento" (Routing). Search shows the time per search and the vector size; transcription, how long it takes per minute of audio; speech, how long until the speech is ready. "sem tarefa" (no task): connected, but with no row in "Roteamento" (Routing).

Who does each task

In "Roteamento" (Routing), each task has a "Modelo principal" (Primary model), a "Reserva" (Fallback), and a "Tempo-limite" (Timeout). If the primary fails or goes over time, the fallback takes over. Intent and handoff can stay with the "decisor da plataforma (incluído)" (platform decider, included). Search, transcription, speech, and image reading are in the "Busca, voz e imagem" (Search, voice, and image) group, on your models of each type (see Search and voice in Routing). Every change takes effect immediately and is recorded in the trail. "Mostrar todas" (Show all) reveals every agent that serves your account, in three groups:

"Seguem a conversa" (Follow the conversation)
Sentiment, objections, lead memory, follow-up, lead profile, and Instagram comments run during the conversation. Without their own route, they use the model, fallback, and timeout of the lead conversation; choose a different model and the task gets its "rota própria" (own route).
"Bastidores" (Behind the scenes)
They do not talk to the lead: they run behind the scenes to prepare your account — ideal customer learning, the group specialist, the qualification adjustment, the industry generator, "Pergunte à IA" (Ask the AI), and the voice profile. Without their own route, they use the conversation's model and fallback, but with the default timeout for each task (from 10 to 60 seconds, depending on the size of the job); you choose their own route the same way.
"Funcionam por regras, sem modelo de IA" (Run on rules, no AI model)
The humanizer, the scheduler, the payment link, the handoff policy, and LGPD use no model at all: there is nothing to choose and they cost nothing. They are in the table so you can see everyone who serves the lead.

On screen: The "Roteamento" (Routing) table opened with "Mostrar todas" (Show all), with the tasks that follow the conversation, the behind-the-scenes ones, and the agents that run on rules.

  1. Primary, fallback, and time: Choose the model for each task. The fallback must be a different model.
  2. "Busca e voz" (Search and voice): Search, transcription, and speech, each only with models of its own type. Search changes by reading the knowledge base again.
  3. "Seguem a conversa" (Follow the conversation): They use the model, fallback, and time of the lead conversation — until you choose a different model, and then the task gets its "rota própria" (own route).
  4. "Bastidores" (Behind the scenes): The tasks that prepare your account without talking to the lead. They also go to your models.
  5. By rules: The agents that do not use an AI model: nothing to choose and no cost.
Rules by lead type
They apply on top of the table, from top to bottom: when the ideal profile goes above a score, when there is an objection, or when the card has a tag, that lead's conversation uses a different model. Drag to change the order (also by keyboard: space, arrows, and space again).
If your models fail
The lead is never left without a reply.

Search and voice in Routing

In the "Busca e voz" (Search and voice) group, each row only accepts models of its own type and none runs on the platform's AI. Search has no model fallback: if your model fails, the message uses keyword search, on the same base, at no cost. Transcription and speech have a primary and a fallback like the other tasks; without a speech model, the reply goes out as text. A lead's audio above 20 MB or 20 minutes does not go to the transcription model (providers refuse it): the conversation goes to your team to listen.

"Leitura de imagem" (Image reading) is the new row in the group and is optional. It only accepts a conversation model that passed the "Lê uma foto de exemplo" (Reads a sample photo) check in the connection test; the "Quais dos seus modelos leem imagem" (Which of your models read images) card, at the top of "Roteamento" (Routing), shows which ones passed. It has no fallback: with your models offline, the photo goes to your team. Without the row, the AI replies based on the photo's caption or asks the lead to write. The image goes to your model as it arrived, like audio: the personal-data masking applies to the text. See Image reading.

On screen: The dialog for changing the search model, with the passages, the cost, and the time to read the base.

  1. Before reading the base again: How many passages, how much it costs on your provider, and how long it takes. Going back to the saved base's model (within 7 days), the switch is immediate and free of cost.

Usage and fair use

On screen: The "Consumo" (Usage) tab with the month's messages, the bars per day, the cost per model, and the warnings.

  1. The month in three numbers: Messages used, the closing forecast, and the month's leads and handoffs — with no charge.
  2. Warnings and cap: The warnings at 80% and when you go past the included messages go by email to the Owners. The "Teto de excedente" (Overage cap) is a limit of your own: above it, neither the platform's AI nor your models respond, and the conversation goes to the team. Raise or remove the cap and the AI goes back to responding on the next message; conversations that already went to the queue stay in the team's queue.

"Segurança e dados" (Security and data)

On screen: The "Segurança e dados" (Security and data) tab with the keys and access, what is sent to the models, and the change log.

  1. Keys and access: Only Owners can change models and rules; require two-step verification to change the page; remember to rotate the keys every 90 days.
  2. What is sent to your models: What goes out is the recent conversation, the company context, and passages from the knowledge base; in background tasks, also the dashboard numbers and the won and lost deals, without names or contact details. Phone, email, and ID number of the lead are not sent: they become placeholders and the reply comes back with the right value. The lead's audio goes as is to your transcription model, and the reply text goes to your speech model; in search, the lead's question goes with phone, email, and ID number masked. You choose whether we store the text of the calls or only the numbers.
  3. Change log: Everything that changed on this page, who changed it, and when. "Exportar" (Export) downloads the log as a CSV.

Once the Terms are signed, the top of this tab shows "Condições do Plano Enterprise" (Enterprise Plan Terms) with the contract number, who signed, and when, plus "Baixar PDF assinado" (Download signed PDF) — the same PDF found in "Carteira" (Wallet), so whoever handles the technical side can find it without going there.

Frequently asked questions

If no model is connected, does the AI stop?
Yes. While Enterprise is not active, the AI does not reply to any lead: conversations go to your team, with the reason. On Enterprise, nothing in your account runs on the platform's AI.
Who pays for the tokens of my models?
You do, directly to your provider. The costs shown here are estimates based on the prices you entered.
Why can't I change anything?
The account requires two-step verification to change this page. Turn it on under "Configurações" (Settings) → "Segurança" (Security). If the account allows only Owners to make changes, Admins can read without editing.
What happens when the model goes down?
The fallback takes over immediately; if it also goes down, the platform's AI replies (if contingency is on) or the conversation goes to your team. For background tasks, without contingency, the task waits for your models to come back.
Do the background agents also use my models?
Yes. Every agent that uses an AI model for your account uses your models: by default, the same one as the conversation; in the table you can give each one its own route.
Does Anthropic work for Enterprise?
It works for the conversation and the tasks. Search and audio need another provider (OpenAI, Google, Azure OpenAI, or your own server) — our implementation team can help.
I use Bedrock. What about audio?
Search runs on Titan, within Bedrock itself. Transcription and speech, in this version, go through OpenAI, Azure OpenAI, or your own server (transcription also through Google).
How much does it cost to re-read the knowledge base?
The screen shows it before you switch: the passages, the tokens, and the cost at the price you entered under "Modelos" (Models). Your provider is the one that bills.
Can I go back to the old search model?
Yes, within 7 days after the switch: choose the old model again and search goes back immediately, at no cost.

In the dashboard: "Abrir a IA própria" (Open Own AI)