Introduction
Brosbots ships ready-made industry templates that let you build a production-grade voice agent in less than a minute. Instead of writing prompts from scratch: you pick the template for your industry, fill in a short form, tick the agent’s capabilities - and the platform generates a fully configured agent: system prompt, first utterance, conversation flow and enabled tools. Your form answers are woven directly into the prompt - you don’t get a half-finished product with placeholders to fill in later.Choosing a template
Go to Create agent to open the template gallery. You can:- search templates by name, industry or tag,
- filter by industry (list) or category chips,
- click Use template to enter the guided setup,
- paste your website address into the banner above the gallery or click AI Generate - the AI picks the template and fills in the form for you (see Generate an agent with AI),
- click I have a call script if you already have a script for your sales team (see Agent from a call script),
- start from a blank agent.
Available templates
Campaign templates
The appointment reminder, the cash-on-delivery confirmation, the payment reminder and the NPS survey call the people on a campaign list, so their call direction defaults to outbound. What differs from one call to the next - the appointment date, the order number, the invoice amount - comes from the columns of the contact list. The prompt refers to them as template variables with a fallback, such as{{amount_due | fallback:(not on the list)}}, and fills them from that
person’s row on every call.
The name is recognised on its own (a column called name, Name,
Imię i nazwisko and similar). Name the other columns as in the table:
When a column is missing the agent sees “not on the list” and is told not to
guess: instead of making up an amount or a time, it says the team will get
back with it. The agent does not read appointments out of a calendar by itself
either - the campaign list is rows you prepare. With a calendar connected it
can check free times and book a new one when someone reschedules; the team
removes the old appointment.
The payment reminder opens with the company name only and mentions the
invoice once it knows it is speaking to the right person. It never mentions
court, bailiffs, debt collection, debtor registers or interest, and agrees
only to what you enter under What the agent may agree to.
The setup form
The form is a step-by-step wizard: one panel at a time, Back and Next between them, and a row of section chips you can click to jump straight to a panel. Nothing is lost when you step back and forth - the answers stay until you create the agent. The four sections below are the panels, with What your agent can do split into several so no panel runs long.1. Basics
Facts about the business the agent is grounded in. At the top of the form you’ll find the “Autofill from your website” card - provide your website address and the AI fills in the description, services, hours and address (only empty fields are filled). The AI also reads what is not in the page text: numbers behind “Call” buttons, e-mail addresses and schema.org structured data (address, opening hours, prices). Fields depend on the industry, but typically include:- Company name (required),
- Main goal of the assistant - what the agent is on the phone to achieve; it goes into the first paragraph of the prompt, in the same sentence as the role (see Main goal),
- Company description - what you do, for whom, what sets you apart; it goes straight into the prompt and acts as an anchor against hallucinations,
- Services / products, Opening hours, Address,
- industry-specific fields, e.g. Cuisine type + Menu (restaurant), Specialization (clinic), Car brands (dealership).
For menu and service fields the agent mentions only what you provide -
it never invents dishes, prices, treatments or staff.
2. Contact details
Phone, email, website and address. They go into the ABOUT block of the prompt together with an instruction to give them out when a caller asks for them. The address only appears here when the template doesn’t already ask for it in Basics. The website autofill fills these fields too.3. What your agent can do (capabilities)
The agent asks for these one at a time: a single question, the caller’s
answer, an acknowledgement, and only then the next question. It never reads the
whole list out in one turn - on the phone the caller would remember only the
last item and the rest of the fields would come back empty.
book_appointment action - see the Cal.com integration. |
| SMS, email, CRM | In-call and after-call notifications, plus contact lookup in HubSpot / Pipedrive. |
| Write every call to Google Sheets | A row is appended to the spreadsheet after every call. |
| Take notes into Google Sheets during the call | The agent writes a row while still on the phone - a callback request, a message to pass on. You give it the tab, what the note must contain and the phrases that should make it reach for one; the columns are fine-tuned later in the agent’s Integrations tab - see Taking a note during the call. |
| Look things up in Google Sheets | The agent reads the sheet mid-call - an order status, a price, whether this caller is already a customer - see Reading a sheet during the call. |
| Google Calendar, Slack, Zapier / Make | An appointment straight into your calendar, a message to a channel, your scenario fired with what the call collected. |
| Weather | The agent can answer what the weather is doing - useful for tours, deliveries and site visits. |
All three Google Sheets capabilities talk to one spreadsheet, so the wizard asks
for the spreadsheet once - under whichever of them you switch on first. The
others show only their own settings. The agent writes through the Google
account signed in on the Integrations page, so sign in there first; the wizard
no longer asks for a service account key.
Main goal of the assistant
The first paragraph of the prompt says who the agent is and what it is there to accomplish:You are a helpful AI customer support agent for AWParts.pl. Your task is to take in complaints, answer product questions and check order status.Leave the field empty and the template’s default goal is used, so the opening paragraph always states a purpose. The goal is repeated in the Main Agenda node as “Main goal”.
Form of address
Languages with a formal/informal distinction force this choice, and models happily say a literal “Sir/Madam”. Each option therefore states the form to use and the forms never to say out loud:Conversation structure
A support agent needs no plan: it waits for the caller to say what is wrong and reacts. A bank, a garage or a clinic often has to say or check something before asking for anything - quote a price, flag a condition, confirm the case even qualifies. That is what How the conversation should go is for. One step per line, in plain words:This is a plan, not a script to read out. The agent covers each step in its
own words, finishes it before moving on, and when the caller jumps ahead or asks
about something else it answers them and returns to where it left off.
conversation_structure variable, so you can edit it later
without running the wizard again. Leave it empty when callers always open with
their own request.
Knowledge base and FAQ
These are two separate fields because they do different jobs:- Knowledge base holds facts (return policy, pricing notes, rules). Paste them, or click Import from file (PDF, DOCX, TXT, MD, HTML - up to 2 MB) and the platform converts the document to text and appends it to the field.
- Most common questions (FAQ) holds question + answer pairs, which go into their own prompt section with the rule: answer those cases as instructed, and fall back to the knowledge base for anything else.
4. Voice and language
- Voice - female · warm, female · bright, male · calm, male · confident.
- Language - Polish, English, Spanish, French, German, Italian, Portuguese, Dutch, or multilingual (auto-detection).
Prompt generation and language
After you click Create agent, the platform builds the prompt via one of two paths:1
Deterministic build (Polish and English templates)
An agent made from a ready template in Polish or English gets a prompt
assembled straight from your answers, with no LLM rewrite. It is the same,
tested prompt every time: the same answers give the same prompt, so it is easy
to see where the agent’s behaviour comes from.
2
AI generator (your own idea and other languages)
For I have my own idea, a blank template with a goal, an agent made from a
call script and the other languages, your answers go to an LLM, which writes
one complete, tailored system prompt - role, personality, knowledge, data
to collect, guardrails and phone conversation style. Without an available LLM
the other languages are based on English with a translation or a strong
language directive, and the multilingual mode tells the agent to match the
caller’s language.
Anti-hallucination guardrails
Every generated prompt contains explicit instructions preventing fabrication, e.g.:“Mention ONLY the services and details listed above. Do NOT invent treatments, prices or staff.”They are powered by the Company description, Menu / Services and Knowledge and FAQ fields.
The agent says it is an AI
The EU AI Act (art. 50) requires the caller to know they are talking to an AI. Every template’s greeting says so in its first sentence, inbound and outbound, in every language - for example “Hello! Welcome to Bella Vista, this is the AI assistant. How can I help you today?”. Every agent’s prompt also carries a rule, whatever the call direction: asked whether it is a person, the agent confirms it is an AI assistant and never claims to be a human, even when the caller insists. The rule is added at call time too, so a prompt you edit or write yourself cannot lose it. A greeting taken from your own call script is kept word for word; if it does not say the agent is an AI, the agent says so in its first reply.What you get
From a template you receive:- a system/global prompt in your language,
- a localized first utterance,
- a ready conversation flow with the typical stages (start → main agenda → summary → ending) - built underneath, you never have to look at it,
- enabled tools (e.g.
book_appointmentfor Cal.com), - voice, transcription and language settings.
What you can change later
Onboarding answers are woven into the prompt as text, so the agent is complete right away. The business facts (name, hours, prices, FAQ, contact details) also show in the Variables tab, but read-only and marked from the wizard: changing them there would not change what the agent says. To change, say, the opening hours, edit them in the prompt on the Role tab or create the agent again with the wizard. If you put{{opening_hours}} into
the prompt, the variable becomes live and editable again.
The data the agent collects lives in the same tab, in the Data collected
by the assistant card - review it, edit it, or add more. See
Template variables.