Select before explaining.
Deterministic engineThe normalizer organizes the data. The selector applies the credit range, fee and installment type. The model receives calculated options.
An agent that understands the request, checks the catalogue and helps compare the options. From the first hello to the next step, inside WhatsApp.
Hi! I can help you compare consórcio options. What would you like to simulate?
I want a R$ 60k car.
Someone asks for a R$ 60k car. The catalogue returns dozens of combinations of credit, term, fee and installment.
The product decision: organize the possibilities before explaining them.{
"idTaxaPlano": "1000111",
"nome": "FX 1 - LINEAR",
"codigoPlano": "IMO111",
"prazo": 120,
"valorParcela": 842.67,
"percentualTaxaAdmMensal": 0.0022,
"percentualTaxaAdmPlano": 26.4,
"percentualSeguroPrestamista": 0.0324,
"tipoParcela": "INTEGRAL"
}Excerpt of the API format, with demo values
Verifiable rules narrow the catalogue to up to three options. Every choice has a reason; every number has a source.
R$ 596,40/mo
Lowest fee among reduced installments; installment breaks ties.
FX 3 - REDUZIDA
A reduced installment is not a fixed installment for the entire term. Check type, term and group conditions.
R$ 852,00/mo
Lowest fee among full installments; installment breaks ties.
FX 3 - LINEAR
A reduced installment is not a fixed installment for the entire term. Check type, term and group conditions.
R$ 663,13/mo
Next option when sorted by fee and installment.
FX 2 - REDUZIDA
A reduced installment is not a fixed installment for the entire term. Check type, term and group conditions.
Demonstration values. Check the group conditions.
Aggregated visualization. Counts and options come from the selector.
The original product’s journey, reconstructed in five moments. Explore the chapters or follow the conversation.
Hi, I saw the ad for a car consórcio
Is the car for daily use or a long-held dream?
Hi, I saw the ad for a car consórcio
Is the car for daily use or a long-held dream?
Daily use. Around R$ 60k
Got it: R$ 60k for a car. I’ll look up the options.
The engine selected three options. The first has a reduced installment. Let’s look at its card?
First selected option, with a reduced installment. Check the group conditions before making a decision.
Option 1
I can pass your choice and context to a specialist. Would you like to continue?
Yes
All set. The next step is to review the conditions with the specialist.
Record, listen and send your voice note. The audio stays in the conversation, and the agent can reply by voice too.
Listen before sending. Your message stays as audio.
The normalizer organizes the data. The selector applies the credit range, fee and installment type. The model receives calculated options.
One question at a time, short messages and retained context. Text and voice belong to the same conversation.
Plans become readable cards. In the original product, the choice reaches a specialist with context; here, you can explore the simulation.
Normalizer and selector, ported from the n8n kit with tests
Catalogued tools, budgets, execution journal and per-visitor limits
Model routed by catalogue; cost accounted per turn
The client's product runs on n8n with prompts and a serverless image generator. The agent on this page is a demo built on the same engine, not the production system.
Short messages, one idea at a time, human rhythm. The prompt forbids long blocks on WhatsApp.
documented in the delivered kitThe person sends a voice note; the agent transcribes it and keeps the flow, never asking them to type.
stated by the authorAn ad screenshot, a photo of the car or the home: the agent extracts context and continues qualifying.
stated by the authorA video explanation at the right point in the conversation, before the simulation, as the prompt requires.
documented in the delivered kitA serverless function draws the image with the options and the closer sends it as media in the chat.
documented in the delivered kitsimular_parcelas queries the catalogue and the selector. The model never invents an installment.
documented in the delivered kitQualifies in minutes and passes the lead to a specialist to close. The agent signs no contract.
documented in the delivered kitA demonstration of the work using the same selection engine. Commercial service is provided through Rodobens channels.
Demo catalogue while the Rodobens API is not connected. Once it is, the quote says "fetched just now".
Two, read-only: quote (catalogue + selector) and list available values. Neither writes anything.
12 messages per session, 3 sessions per visitor per day, 600 characters, 2 tools per turn, 25 s per turn and a global daily budget.
This demonstration does not request your name, phone or ID. Messages are processed by the AI provider and kept in server memory. Sessions expire after 24 hours; content is removed when the process restarts. No database storage.
Routed through a catalogue (OpenAI or OpenRouter); the identifier is this site's configuration, not the client's choice.
Listen before sending. Sending uploads the audio file to the server and transcription service so the agent can understand it. The recording remains playable on this page until you reset or leave. The server temporarily retains the transcript in the session and does not save the audio file. Spoken replies use a public catalogue voice.
Logo obtained from the official Rodobens website. This interface presents the case and does not provide commercial customer service. rodobens.com.br
Reconstructed from the prompt and the kit’s image generator — not a capture of a real conversation. No personal data.
The original prompt includes dated commercial conditions and campaign urgency. They aren't reproduced here: this page doesn't sell.
Start with a value, a question or a voice note.
Granvinhas