A Voice and Text Agent Built to Turn Visitors Into Leads
Intentport is the voice and text agent platform we built for websites. It answers visitors from a business's live catalogue, hours and knowledge base, books appointments, and hands qualified leads to the CRM, in five languages.
Showroom concierge
AI assistant
-
Hears
- Transcribes English while the visitor is still talking
- Catches the model and the day they want
- Hears a spoken yes to the read-back
-
Looks up
- Catalogue: 7-seat SUVs 2 in stock
- Opening hours: Tuesday closed
- Free slots: Wednesday 11:00
-
Replies
- Answers in the English voice, spoken and written at once
- Shows the two matching models
- Offers the next real slot instead of the closed day
-
Acts
- Books the test drive
- Sends the lead to the CRM Wants a 7-seat hybrid SUV. Test drive booked for Wednesday at 11:00. Conversation in English. Transcript and recording attached.
Category
AI Agents in Production + Voice AI Platform
Project Type
MicroPyramid product, live at intentport.com
Audience
Businesses that sell or book through their website
The Problem
Most website chat can answer a question, but it rarely moves a visitor toward a decision. Generic bots guess from the open web, return every match instead of helping someone choose, and hand the sales team a name with no context about what the visitor actually wanted.
Visitors also do not all speak the same language, and many would rather talk than type. Intentport was built to close that gap: answer from the business's own data, guide the visitor to a choice, book the visit or capture the lead, and pass the whole conversation to the people who follow up.
What We Built
- voice and text conversations in one embeddable widget, added to a site with a single script tag
- answers grounded in the business's live catalogue, opening hours, policies and approved knowledge base
- product cards and interactive choices shown alongside the spoken answer
- appointment booking checked against real opening hours and available slots
- lead capture with the visitor confirming details before anything is sent
- CRM handoff to Salesforce, HubSpot, Zoho, GoHighLevel, Odoo and SAP, or any API or webhook
- five languages (English, Arabic, Russian, German and Dutch), each with its own voice
- an Intent Log of transcripts, recordings and summaries of what visitors wanted
The Intent Log: What Visitors Wanted
Every conversation is kept with its transcript, a summary, and the lead it produced. Over a month, the log shows catalogue gaps, objections in the visitor's own words, and demand by product, location, and hour.
What web analytics shows
What visitors clicked
Pageviews, SUV range
Bounce rate
Time on page
Top site search: "price"
Contact form conversion
Every figure is accurate. None of them says why visitors left.
What the Intent Log shows
What visitors wanted, in their own words
- Test drive booked
Do you have a seven-seat hybrid in stock?
- Answered
Is the showroom open on Friday evening?
- Callback requested
Can I trade in my current car?
- Callback requested
هل يوجد تمويل بدون دفعة أولى؟
Is there financing with no down payment?
- Not in catalogue
Do you sell electric pickups?
Guardrails Built In
An agent that speaks for a business on its own website has to know what it is not allowed to do.
Honest about being an AI
Asked whether it is a person, it says plainly that it is an AI assistant.
Nothing written without a yes
Every booking and lead is read back to the visitor first, and only saved after they confirm.
No promises it cannot keep
A closed day or a past hour is refused with a reason. Without real hours, it registers interest for a callback instead.
Stays on your site and your topic
Only the business's own domain can start a conversation, and questions outside its field are politely declined.
Conversations stay yours
Each client's conversations are stored in isolation and kept until they ask for deletion, or held in their own database instead.
Why This Was More Than a Chatbot
An agent that talks to customers on a live website has to be fast, stay inside what the business can actually do, and leave a record the team can trust.
keeping a spoken conversation natural, with streaming speech-to-text, turn detection and replies fast enough not to feel like a phone tree
answering only from the business's own data, and taking a callback instead of guessing when the answer is not there
following the visitor into another language mid-conversation, including right-to-left Arabic in the interface
testing each change against recorded conversations in every language before it reaches a live website
Outcome
- live in production at intentport.com, serving clients in the UAE and India
- one platform running voice and text agents in five languages
- leads reach the sales team with the full transcript and a summary attached
What Carries Over to Your Project
You may not need a showroom concierge. If you are putting an AI agent in front of your customers, these are the parts of Intentport we bring with us.
Answers grounded in your systems
Live lookups into your catalogue, stock, calendars, and approved knowledge, with a callback instead of a guess when the answer is not there.
Actions with a confirmation step
The agent reads back what it is about to write to your CRM or calendar, so nothing lands on the strength of a misheard word.
Every conversation costed and timed
Cost broken down by speech, language model, and voice, plus turn-by-turn timings that show where a slow reply spent its time.
Models you can swap
Speech, language, and voice models are set per agent, so a better or cheaper one can replace the old one without a rebuild.
Relevant Service Lanes
Intentport is the same work we do for clients: an agent wired into real systems, measured on real conversations, and run in production.
AI Agents in Production
Agents that do real work inside your systems, with evaluations, audit trails and cost controls.
Open pageAI Knowledge Assistants
Assistants and search grounded in your own documents and data, with citations.
Open pageSaaS Development
Multi-tenant architecture, billing, dashboards, and APIs.
Open pageQuestions About This Build
Did MicroPyramid build Intentport?
Yes. Intentport is our product. Our team built the voice agent, the operator dashboard, the embeddable widget, and the website, and we run it in production.
Can you build a voice or text agent on our own data?
Yes. We connect the agent to the systems your staff already use (catalogue, calendar, CRM, knowledge base) and test it on the questions your visitors really ask. We start with a discovery sprint and give a fixed estimate after it.
Can the agent speak Arabic and other languages?
Intentport speaks English, Arabic, Russian, German, and Dutch today, each with its own voice and its own tested prompt, and it follows a visitor who switches language mid-conversation. Arabic runs right to left in the interface.
Where are the conversations stored?
Each client's conversations, text and voice, are stored in isolation and kept until the client asks for them to be deleted. They can also be stored in the client's own database and cloud storage instead.
Want an Agent Like This on Your Site?
If you are planning a voice or text agent that has to work with your real data and systems, MicroPyramid's senior engineers can scope it, build it, and run it in production.