Some launches can be explained as an installation. Others say more about how a team works. Bringing Reception.ai to karmina.ai belongs to the second category.
We discovered it while working in Chrome. We spotted a promising tool, explored it and quickly saw how it could fit a very specific need: giving Karmina a conversational front door, through voice and text, available at any time. But finding a tool is not the same as turning it into a useful experience. Between the two sit knowledge, judgement, architecture, content, testing and a clear decision about what the agent should and should not do.
That is the development we want to share. Reception.ai is now part of Karmina’s website in Catalan, Spanish and English. More importantly, it has helped us validate an idea at the heart of our approach to artificial intelligence: when an organisation has structured its knowledge well, an agent can connect to it much more easily and start delivering value quickly.
What Reception.ai is
Reception.ai ↗ is a conversational reception solution powered by AI agents. It can be used to create a virtual receptionist that handles voice or text enquiries, explains business information, guides a visitor, gathers context, routes a conversation and, where appropriate, connects to tools such as calendars or management systems.
The idea is easy to describe: an initial point of contact available 24/7, without forcing people into a rigid form or making them wait until someone is free to respond. Yet its real value is not simply “a voice that talks”. Its value depends on whether that voice understands the company, uses accurate information, knows its boundaries and provides a clear handover to a person.
The technology behind it makes it possible to embed the experience in a website and combine voice and text. The official ElevenAgents widget documentation ↗ also covers visual customisation, authorised domains and connections to tools. It is a powerful foundation. But a foundation is not yet a receptionist ready to represent a brand.
Spotting the opportunity and putting it to work quickly
We often talk about agility, but not as a synonym for moving fast without judgement. In this case, agility meant shortening the distance between four moments: spotting an opportunity, understanding it, deciding whether it made sense for Karmina and testing it in a real environment.
We started by exploring the product in Chrome. Instead of stopping at a demo, we looked at how it could fit into the ecosystem we already have. The final implementation is not a browser extension attached superficially. It is integrated into karmina.ai. The agent appears as a conversational layer of the website itself and coexists with its navigation, content and other contact points.
We have prepared it to work across all three website languages. We have also defined where it makes sense to show it and where it does not. It stays out of spaces such as the main chat and legal pages, where another floating interface could create confusion. This decision is less visible than the voice, but it is essential. A good integration also knows when not to appear.
The difference is not the widget. It is the knowledge
Installing a widget can be quick. Enabling it to respond with sound judgement is a different job.
An agent does not know a company because it has read three slogans. It needs to understand what the organisation does, how it describes itself, which services it offers, who it serves, which language it uses, which promises it must not make and when it needs to acknowledge that human help is required. It needs clear sources, priorities and boundaries.
This is where the work Karmina had already done becomes important. Our Karmina Brain organises identity, proposition, criteria, services and useful knowledge. The Knowledge Foundation prepares that knowledge so AI systems can consult and apply it. These are not two labels added after the agent. They are the infrastructure that gives the agent something solid to work with.
That is why the connection has been agile. We have not had to invent Karmina from scratch to feed a new tool. We already had a content architecture, a clear proposition, routes in three languages, structured services and brand criteria. The agent can connect those pieces because the organisation had already turned them into accessible knowledge.
This is an important lesson for any business: implementation speed does not depend only on how technically easy a tool is. It depends on how ready the organisation is to explain itself to a machine without becoming unfaithful to who it is.
An agent does not start with a prompt
Discussions about AI agents often focus on finding the “perfect prompt”. It is a tempting simplification. The prompt matters, but it cannot replace a knowledge base, defined responsibilities, up-to-date sources or a system of oversight.
At Karmina, we believe that an AI agent does not start with a prompt. It starts with a business question: what problem should it solve? Then we need to define what knowledge it needs, which tools it can access, which actions it can take, what it must record and when it should hand over to a person.
Reception.ai turns that idea into a highly visible use case. The conversation is the layer the visitor sees. Underneath it, the system needs content, structure and rules. Without that invisible layer, the receptionist can sound convincing while remaining unhelpful or unreliable.
Testing AI at home before taking it to clients
We wanted to implement this tool on karmina.ai because we believe in a simple rule: before recommending a technology, we need to work with it ourselves.
Testing it on our own website forces us to answer real questions. How do we greet people without sounding generic? How do we explain an agency that brings together marketing, creativity, technology and AI? How do we maintain consistency across languages? What happens when an enquiry falls outside our scope? How do we prevent two conversational elements from competing on the same screen? What should the agent do when it does not have enough information?
This practice gives us experience that cannot be gained from watching a commercial demo. It lets us see the distance between the promise and real operation, understand the content work required and identify where human oversight remains essential.
We are not claiming that a first version handles all of Karmina’s reception needs, or that we already have metrics at scale. This is a real implementation that lets us learn from real traffic, improve the knowledge behind it and decide the next steps from evidence.
From reception to an agent ecosystem
An AI receptionist is a front door. Its potential grows when that door is part of a wider system.
An enquiry may begin with a general question, move towards the right service, prepare information for the team and end in a human conversation with much richer context. In other organisations, the journey could include a booking, incident classification, a catalogue search, preparation for a visit or access to internal documentation.
That is why we talk about agent ecosystems rather than isolated tools. Each agent can perform a specific role, but they should share sources, permissions, criteria and a governance model. If every tool receives a different version of the company, the organisation gains interfaces but loses coherence.
The model we are building starts from a shared foundation: the company’s knowledge. From there, different agents can consult it or trigger processes according to their role. The receptionist talks to website visitors. An internal agent could help the team find information. Another could prepare a proposal. A third could review brand consistency. They do not need to know or do everything. They need the right part of the knowledge and the right permissions.
What we will start implementing for clients
Our experience with Reception.ai opens a clear line of work for Karmina AI Studio: implementing similar conversational tools for clients when they solve a real need.
This does not mean installing the same receptionist on every website. The organisation’s context will always be the starting point. Potential use cases include:
- Reception and initial guidance. Answering frequent questions, explaining services and directing each person to the right channel.
- Enquiry qualification. Gathering the essential context before a sales or service team takes over.
- Bookings and availability. Connecting the agent to calendars or appointment systems when the process is well defined.
- Out-of-hours service. Providing a useful first response without promising a resolution that requires a person.
- Product or service support. Consulting controlled catalogues, manuals or knowledge bases.
- Internal assistance. Allowing teams to ask questions of company documentation with permissions and traceable sources.
Every implementation must answer the same questions: what information the agent may use, which source is authoritative, how often it is updated, what data it may collect, which actions it may perform and when it must hand over to a person. The technology may be shared; the knowledge, processes and responsibility are specific to each client.
Agility with judgement
Being ahead does not mean collecting tools. It means spotting a useful possibility early and having the ability to turn it into a coherent solution.
We were able to move quickly in this case because three capabilities came together. The first is exploration: paying attention to what is emerging and understanding what it may change. The second is implementation: knowing how to connect a technology to a real website, an experience and a set of constraints. The third, and the hardest to copy, is knowledge: having the organisation’s information structured well enough that the agent does not start from nothing.
That is what we want to bring to clients. Not only the selection of a tool, but the complete journey from opportunity to operation: diagnosis, knowledge foundation, agent definition, integration, testing, governance and continuous improvement.
The new feature is visible. The advantage is the system
Reception.ai is now part of karmina.ai. It can be seen, heard and tested. That is the visible development.
The more important conclusion is different. A tool can be discovered today and change tomorrow. What lasts is an organisation’s ability to structure what it knows, turn it into a shared foundation and connect new agents without rebuilding itself every time.
That is why this implementation says as much about Karmina as it does about Reception.ai. We spotted an opportunity, acted on it with agility and integrated it into an architecture that already existed. We are now beginning to bring that learning to clients that need more available reception, more consistent service or a new way to activate their knowledge.
Technology lets us move fast. Knowledge tells us where to go.