To create a WhatsApp AI agent in ConverZap, open “Agentes de IA” (AI Agents) in the menu, click “+ Novo agente” (+ New agent), choose a starter preset, fill in the form tabs, test it in the “Simulador” (Simulator) and click “Ativar” (Activate). The agent is built through a form, with no prompt writing: ConverZap generates the AI instructions from what you fill in and shows the result in the “Prompt gerado” (Generated prompt) tab. The agent replies on the WhatsApp number connected through the official API and uses the workspace’s own AI key (Gemini, Anthropic or OpenAI), added under “Configurações” (Settings) → “Integrações” (Integrations).
Choose a starter preset
When you click “+ Novo agente”, ConverZap offers four presets, all editable later: “SDR — qualificação e agendamento” (SDR — qualification and scheduling), “Atendimento e dúvidas (FAQ)” (Customer service and questions, FAQ), “Agendamento de horários” (Appointment scheduling) and “Em branco” (Blank).
Fill in the AI agent tabs
- “Identidade” (Identity): internal name, the name the agent uses in the conversation, company, role, introduction, tone of voice, form of address, emoji use and language. The same tab holds the “Objetivo” (Goal) section (main goal, audience and what a successful conversation looks like) and the message style, such as “Uma pergunta por vez” (One question at a time).
- “Roteiro” (Script): steps in order (“+ Etapa”, meaning + Step), each with a goal, “Como conduzir” (How to lead), a sample message, the questions to collect and when the step ends. Without a script, the agent converses freely toward the goal.
- “Perguntas” (Questions): the data to collect (“+ Pergunta”, meaning + Question), with a type (text, number, email, phone, CNPJ [Brazilian company tax ID], options or date), validation and where the answer is saved on the contact (name, email, company, city, state or a custom field). This is also where you turn on rule-based “Qualificação automática” (Automatic qualification), a decision made by the system, not by the AI.
- “Produtos & objeções” (Products & objections): products and differentiators, pricing policy, frequently asked questions and responses to objections. The toggles “Usar a marca (produtos, ofertas vigentes, provas)” (Use the brand: products, current offers, proof) and “Consultar a base de conhecimento (documentos) para dúvidas” (Check the knowledge base documents for questions) use what is registered under “Marca & Conhecimento” (Brand & Knowledge).
- “Regras & humano” (Rules & human): what the agent always and never does, human handoff and the “Operação” (Operation) section: wait time to group messages, maximum number of lead messages, business hours, understanding of audio, images and documents, conversation retention and “Modelo de IA” (AI model).
- “Ferramentas” (Tools): actions the AI performs during the conversation, described below.
- “Canal” (Channel): choose “API oficial da Meta (direto)” (Meta official API, direct) and the “Número que o agente atende” (Number the agent handles).
AI model
Under “Regras & humano” → “Operação”, the “Modelo de IA” field accepts “Automático — rápido e barato” (Automatic — fast and low-cost), “Automático — mais inteligente” (Automatic — smarter) or a specific model available through the workspace’s keys. If the chosen model fails, the agent falls back to the automatic option.
Tools: calendar, pipeline and integrations
Saving the answers to questions and moving forward in the script are always on. In the “Ferramentas” tab you can turn on: checking the knowledge base, handing off to a human, ending the conversation, “Usar o funil” (Use the pipeline), which moves the deal on the Kanban board when the lead is qualified or disqualified, “Permitir marcar tags” (Allow tagging), “Agendar reuniões” (Schedule meetings) through Cal.com (requires Cal.com connected under “Integrações” and the event type ID), and custom tools that call one of the business’s https URLs, such as a CRM or an order system.
Human handoff
In “Regras & humano”, define whether the agent hands off when the lead asks to talk to a person, when the lead is qualified, or when it receives trigger words that transfer the conversation immediately, bypassing the AI. You can also write the “Mensagem ao transferir” (Handoff message) and turn on “Pausar o agente quando um atendente responder a conversa” (Pause the agent when a team member replies to the conversation). A handed-off conversation appears as “Aguardando humano” (Waiting for human) in the inbox, with a summary in an internal note.
Test and activate the AI agent
- Click “Salvar” (Save). The simulator always uses the saved version.
- Open the “Simulador” tab and chat as if you were the lead. Nothing is sent to WhatsApp; pipeline, tags, calendar and custom tools are only simulated. Use “Reiniciar” (Restart) to start over.
- Check the “Para ativar falta:” (Still needed to activate:) list, which flags pending items such as the agent’s name, goal or channel number.
- Click “Ativar”. From then on, the agent replies to messages received on the number. To stop it, click “Pausar” (Pause).
Only one agent can be active per number. In the agent’s “Conversas” (Conversations) tab, you can follow its conversations and use “Assumir (pausar agente)” (Take over, pausing the agent), “Devolver ao agente” (Return to agent) and “Encerrar” (End).
Tip: outside the 24-hour window, the agent can’t start a conversation with free-form text. To reach contacts who haven’t messaged recently, use an approved template in a bulk messaging campaign.