Lyra by Felikon

An AI host that talks to your customers, and writes you what they said.

Lyra stands on an iPad in your shop or on your table at an event. It greets the people who stop, asks four or five questions steered by a one-paragraph brief you write, and sends you a short report on each conversation. After five, what kept coming up.

Request a demo See a week with Ava
Lyra's customer screen on a tablet, Willow Row in the corner: Ava, a woman with tied-back auburn hair in a leather jacket, listening. Beneath her the band shows her question, Understood! What kind of jacket were you hoping to find today that we didn't have?, and under it the customer's reply, You had the navy one but not in my size.
A whole conversation, 44 seconds, on the customer's screen. Recorded from a test browser; Ava's voice is as it played, the customer's replies were typed by a script. Sound is off until you turn it on.

It listens, it asks, it writes it down.

Lyra is a web page on a tablet or a laptop, with a camera and a microphone. You set it up from your own admin page and give it a brief.

It starts when someone says hello.

Nothing runs until a customer taps Say hello. Then the camera looks for the person in front of the screen and what they are wearing, Ava greets them and may say a word about it, and she says out loud that she is an AI and that the conversation is recorded. When the conversation ends, the camera and the microphone stop.

It asks what your brief says to ask.

The brief is one paragraph in a settings page: this week's window, next season's buy, a new collection, the offer on the rail. It steers every question, four or five of them, each short enough to answer standing up. Ava listens in English today; other languages have not been tried with customers yet, and if you ask we will say honestly whether it is weeks or months.

It writes down what it heard.

After every conversation, a short report: what they came in for, what they asked for that you did not have, what they thought of the offer, with the transcript word for word. After five conversations, what keeps coming up. The report can be written in English or Romanian while your customers speak English.

A week at Willow Row

Willow Row is our demonstration shop, not a customer. The conversations behind these screens were scripted so that the reports had something to report on, so read the shape of the week rather than the content.

  1. Mondaybefore opening

    The owner types the brief.

    Ninety seconds in the admin page. She is choosing what to buy for next season, so this week Ava asks about that. The brief also says what Ava must never do: invent stock, quote a price that is not written here, or say something is nearly sold out.

    The owner is choosing what to stock for next season and wants customers' opinions before she buys. Find out what the customer came in for today, what they wished the shop had and did not, which colours and shapes they would actually wear, what they would expect to pay for a jacket here, and whether they would like to hear when new pieces arrive. If they would, tell them the person at the till can take their name.

    The settings page of Lyra's admin for Willow Row: the assistant name Ava, the brand name Willow Row, and the campaign brief in a text box that begins You are Ava, the host of Willow Row.
    The settings page. The next conversation on any device using the shop's link picks up the change, with nothing to restart.
  2. Tuesdaylate morning

    A customer stops, and Ava says hello.

    She greets, asks what the customer came in for, and follows the brief from there. Each of her lines is a sentence or two, and her words appear beneath her face as she says them. The customer answers by talking. When the questions run out, Ava rounds off with what she heard and says goodbye.

    AvaOh, what a shame! What size would you usually wear in a jacket like that?

    CustomerI'm a large, sometimes extra large in jackets

    AvaThanks for letting us know! What would you expect to pay for a work jacket here?

    Three lines from the conversation at the top of the page, as the band beneath her showed them.

    Lyra's customer screen on a tablet, Willow Row in the corner. Ava looks at the customer. The band beneath her shows her question, Thanks for letting us know! What would you expect to pay for a work jacket here?, and under it the customer's reply, It happens a lot here, the bigger sizes go first. Below, a microphone button and End conversation.
    The customer's screen a few lines later in the same conversation. The small (i) mark opens the shop's own disclosure line.
  3. Tuesdaya few minutes later

    The report is on the owner's phone.

    Every conversation gets one: a summary, what the customer was interested in, what the brief asked to find out, any objections, and the transcript word for word. The owner reads it between customers. If it caught something she can act on, it is one line long.

    A conversation's report in Lyra's admin on a phone. It begins: This customer was looking for a long cardigan, specifically mentioning a cream one that was previously stocked. They expressed a preference for cream, navy, or grey colors and disliked cropped styles due to their height. Beneath it, sentiment and purchase intent, then the transcript line by line.
    Another conversation from the same day, opened on a phone: the report, then every word of the transcript. Any conversation can be deleted with one click.
  4. Fridayafter the fifth conversation

    What kept coming up.

    Once five conversations have a report, Lyra writes conclusions across them: a headline, the findings that recur, the interests heard more than once, and what to do about them. Beside each finding it says how many of the five conversations support it, and at five it calls its own confidence low, in so many words.

    The conclusions panel in Lyra's admin. The headline reads: Customers are looking for specific items and sizes, with recurring interests in navy, cream, and grey, but are frequently disappointed by stock availability and pricing. Below it, five numbered findings, each followed by how many of the five conversations the model says support it, then recurring interests, recommended actions, a confidence note that reads Low, and customer segments.
    The conclusions panel, rewritten every five conversations.
  5. Next Mondaythe window changes

    So does the brief, if you want it to.

    Whoever changes the window changes the paragraph, and the week's conversations are about the week's window. A shop that would rather write the brief once and leave it can do that too.

An evening at the stand

The same product on a table at a small-business evening, with a different brief. This is Felikon's own stand, and the five visitors were scripted like the customers at Willow Row, so read the shape rather than the content.

  1. Before the doors openthe brief

    The assistant is told whose table it is.

    Four questions, about a minute of a visitor's time: what business they run, which repetitive task costs them most, what tools they use today, and whether privacy or setup time matters more. Then an invitation to leave a card with the person at the table, because the assistant takes no names.

    We are Felikon, a small agency that sets up AI assistants for small businesses: customer-support chatbots, email drafting and document search. Find out what business the person runs, which repetitive tasks take most of their time, what tools they use today, and whether data privacy or setup time matters more to them. Close by inviting them to leave a card with the person at the table.

    The settings page of Lyra's admin for Felikon: the assistant name Felikon Assistant, the brand name Felikon, the campaign brief that begins We are Felikon, a small agency, and four questions per conversation.
    The same settings page, another company: four questions instead of five, and a hundred conversations for the evening.
  2. Seven o'clocka baker stops

    A minute at the table.

    The assistant greets, asks what business the visitor runs, and follows the brief from there. The person at the table listens with half an ear and reads the report on a phone once the visitor has moved on.

    AssistantA bakery, delightful! What repetitive tasks consume most of your time?

    VisitorAnswering the same questions on Facebook and the phone, opening hours and custom cakes, that eats hours every week

    AssistantThat's a lot of questions! What tools do you use to manage them now?

    Three lines from the conversation on the screen beside this, as the band beneath her showed them.

    The visitor's screen at Felikon's table: the same face, Felikon in the corner. The band beneath her shows the visitor's line, I run a small bakery, two shops and a market stall, and the assistant's reply, A bakery, delightful! What repetitive tasks consume most of your time? Below, a microphone button and End conversation.
    The visitor's screen, the moment after the baker answered.
  3. Between visitorson a phone

    The report says which service fits.

    This table's report focus asks three things: which of Felikon's three services fits this person, how urgent it sounds, and one sentence to open the follow-up with. The plumber's report: quotes and follow-ups written on the sofa at night, a phone, WhatsApp and a notebook, and setup time over privacy.

    A visitor's report in Lyra's admin on a phone. It begins: This customer is a self-employed plumber with one apprentice. He spends his evenings writing quotes and follow-up messages. He uses his phone, WhatsApp, and a notebook for these tasks. He prioritizes quick setup over data privacy for a new tool. Beneath it, sentiment and purchase intent, then the transcript.
    The plumber's report, read between visitors. The transcript follows beneath it.
  4. After the fifth visitorthe same evening

    Which service to lead with.

    Five visitors in, the conclusions: what recurred across a bakery, a law office, a plumber, a dentist and an online shop, and what to do about it. The same panel as the boutique's, written from a different brief.

    The conclusions panel for Felikon's stand. The headline reads: Small businesses are actively seeking solutions for repetitive tasks, with a clear divide in priorities between quick setup and data privacy. Below it, five numbered findings with how many of the five conversations support each, recurring interests, recommended actions, a confidence note that reads Low, and two customer segments.
    The conclusions after five visitors. At five it rates its own confidence low, and says so.

What the customer sees, and where it goes

Written as a description, not a promise. Everything here is how the product behaves today.

Before anything, a tap
Nothing runs until a customer taps Say hello. The camera and the microphone start then, and stop when the conversation ends. Nobody walking past is seen.
What she says first
The shop's own disclosure line, spoken and kept behind the (i) mark on the screen. Willow Row's reads: I am an AI assistant for Willow Row. This conversation is recorded so the shop can improve its service. Nothing is kept with your name. Each company writes its own.
The camera
Used to notice that a person is there and what they are wearing, so Ava can say hello and a word about it. The pictures themselves are not stored. What is kept is a phrase such as medium blue trousers.
The microphone
The customer's speech is turned into text by the browser's speech recognition. The text is what Ava answers and what the transcript holds. No audio recording is kept.
What is kept, and for how long
The transcript and the report, in the company's own account, for a period the company sets (90 days unless changed), and any conversation can be deleted one at a time. No names, emails or phone numbers are taken: anyone who wants telling about new pieces is sent to a person.
Where it runs
On Google Cloud, which does the seeing, the language and the voice, with the language and voice models in a European region and the Lyra server in a US region. If a shop runs it, the shop is the data controller, Felikon processes on its instructions, Google Cloud is the sub-processor, and a one-page agreement says so.
What it does not do
It does not sell, and it does not know your stock: it cannot say you have that in a twelve, and it will not invent it. It tells you what customers asked for; the buying stays with you.
When it gets something wrong
It can. The brief tells it what never to say, every word it said is in the transcript, and the conversation can be deleted. What was heard cannot be unsaid, which is why the first afternoon in a shop is with us standing beside it.

Ava on an iPad

The device at rest on a desk stand, photographed and filmed in September 2026. Nothing on the screen is a mock-up.

  • The iPad seen straight on: Ava is here, and the Say hello button.

    Straight on, at rest. The page is the same on a phone, a tablet and a laptop.

  • An iPad on a black desk stand shows Ava against a dark stage, the words Ava is here and a Say hello button beneath her, with a mirror and a window behind.

    In the room. She is rendered live in the browser, which is why she keeps moving at rest.

See it with your own brief.

Ten minutes on a laptop we bring, set up beforehand with whatever your window says this week. Nothing of yours is processed. If you want more, an afternoon in your shop with us standing beside it, where the first real count happens: how many people stop, how many talk, and what you learn from them.

What you need
An iPad or a laptop with Chrome, a stand, Wi-Fi and one link. For the afternoon we bring the kit.
What it costs today
There is no price list yet. The first shops run it as a pilot, and the price is being set with them.
What can carry your brand
The assistant's name, your shop's name, the brief, the disclosure line and the report language. Not yet her face, her voice or her clothes.
Who is behind it
Felikon, a small agency that sets up AI assistants for small businesses: customer-support chatbots, email drafting and document search. Lyra is its own product, and no shop runs it yet: the first afternoon in yours would be the first real measurement.

This opens an email in your own mail app, addressed to us; nothing is sent from this page. Or write to marius.peicu@felikon.com.