Most technology asks people to adapt to it.<\/p>\n\n\n\n
Download an app.<\/p>\n\n\n\n
Create an account.<\/p>\n\n\n\n
Remember another password.<\/p>\n\n\n\n
Learn a new interface.<\/p>\n\n\n\n
Understand where to tap.<\/p>\n\n\n\n
Keep the device updated.<\/p>\n\n\n\n
For younger users, those demands may barely register.<\/p>\n\n\n\n
For an older person who has spent most of their life without smartphones, apps or conversational AI, they can become the difference between a useful service and one that is never used.<\/p>\n\n\n\n
R\u016bpest\u0117lis starts from the opposite assumption: technology should learn how to communicate with the person.<\/strong><\/p>\n\n\n\n Built in Klaip\u0117da by founder Tomas Margelis<\/strong>, R\u016bpest\u0117lis is a Lithuanian-speaking voice AI assistant designed for older adults.<\/p>\n\n\n\n There is no new device to learn.<\/p>\n\n\n\n No app interface to master.<\/p>\n\n\n\n The service works through something almost everyone already knows how to use:<\/p>\n\n\n\n an ordinary telephone call.<\/strong><\/p>\n\n\n\n And Margelis is careful to make one thing clear from the beginning.<\/p>\n\n\n\n R\u016bpest\u0117lis is not a concept waiting to be built.<\/p>\n\n\n\n The product is already built, tested and operating in Lithuania.<\/strong><\/p>\n\n\n\n The idea behind R\u016bpest\u0117lis began with a simple question.<\/p>\n\n\n\n Why should older people have to learn new digital interfaces when the technology itself could adapt to them?<\/p>\n\n\n\n For Margelis, that question quickly became about more than usability.<\/p>\n\n\n\n It became about independence.<\/p>\n\n\n\n \u201cOur starting point is respect for the individual, their experience, independence and choices,\u201d he says.<\/p>\n\n\n\n That principle runs through much of the product.<\/p>\n\n\n\n The telephone matters because it is familiar and widely accessible.<\/p>\n\n\n\n But Margelis does not see the telephone itself as the defining feature.<\/p>\n\n\n\n Voice is simply the interface.<\/p>\n\n\n\n The more important question is what the service can actually help someone accomplish.<\/p>\n\n\n\n R\u016bpest\u0117lis is designed to help with ordinary everyday needs.<\/p>\n\n\n\n The assistant can hold conversations, find information, manage reminders, help organise requests and support communication with services or institutions through available and authorised channels.<\/p>\n\n\n\n A user might want information about a local service.<\/p>\n\n\n\n Help preparing an enquiry to an institution.<\/p>\n\n\n\n A reminder about something important.<\/p>\n\n\n\n Or assistance arranging contact with another person or organisation.<\/p>\n\n\n\n The distinction between those tasks matters.<\/p>\n\n\n\n R\u016bpest\u0117lis does not treat every request as something the AI can automatically complete.<\/p>\n\n\n\n The company separates three different states:<\/p>\n\n\n\n a request;<\/strong><\/p>\n\n\n\n an action that has started;<\/strong><\/p>\n\n\n\n and<\/p>\n\n\n\n a confirmed result.<\/strong><\/p>\n\n\n\n If a user asks the assistant to help with something, that does not automatically mean the task has been completed.<\/p>\n\n\n\n For Margelis, communicating that distinction clearly is an important part of trustworthy AI.<\/p>\n\n\n\n Voice assistants face a difficult temptation.<\/p>\n\n\n\n They are designed to keep conversations flowing.<\/p>\n\n\n\n When something is unclear, a language model can often produce a plausible interpretation and continue.<\/p>\n\n\n\n That may make a demonstration look smooth.<\/p>\n\n\n\n For R\u016bpest\u0117lis, it can be exactly the wrong behaviour.<\/p>\n\n\n\n \u201cIf the assistant is uncertain, it should clarify,\u201d Margelis says.<\/p>\n\n\n\n \u201cIf an action cannot be completed, the person should not be told that it succeeded.\u201d<\/p>\n\n\n\n That sounds obvious.<\/p>\n\n\n\n In practice, it touches one of the most important problems in applied AI.<\/p>\n\n\n\n A system that speaks confidently can create the impression that something happened even when it did not.<\/p>\n\n\n\n A message may have been prepared but not sent.<\/p>\n\n\n\n A notification may have been sent but never read.<\/p>\n\n\n\n A request may have been initiated but not confirmed.<\/p>\n\n\n\n R\u016bpest\u0117lis is therefore being refined<\/strong> around what Margelis calls a principle of honesty.<\/p>\n\n\n\n When the system is uncertain, it should ask.<\/p>\n\n\n\n When something fails, it should say so.<\/p>\n\n\n\n And when a result has not been confirmed, the AI should not turn uncertainty into certainty simply to keep the conversation moving.<\/p>\n\n\n\n Building a Lithuanian-speaking voice assistant also creates a very practical engineering problem.<\/p>\n\n\n\n Real people do not speak like written prompts.<\/p>\n\n\n\n They pause.<\/p>\n\n\n\n Interrupt themselves.<\/p>\n\n\n\n Change subjects.<\/p>\n\n\n\n Use regional expressions.<\/p>\n\n\n\n Speak quickly.<\/p>\n\n\n\n Speak slowly.<\/p>\n\n\n\n Leave thoughts unfinished.<\/p>\n\n\n\n Repeat themselves.<\/p>\n\n\n\n For an older user, the experience also has to remain understandable and comfortable.<\/p>\n\n\n\n R\u016bpest\u0117lis therefore pays particular attention to clear language and conversational pacing.<\/p>\n\n\n\n If the system does not understand, the preferred behaviour is to ask for clarification rather than guess.<\/p>\n\n\n\n Margelis does not disclose the underlying engineering implementation, describing it as proprietary.<\/p>\n\n\n\n But the design target is straightforward:<\/p>\n\n\n\n the conversation should adapt to the person rather than requiring the person to learn how to speak to a machine.<\/p>\n\n\n\n This becomes especially important once a voice assistant can do more than answer questions.<\/p>\n\n\n\n A conversational AI that can coordinate with relatives, care workers, institutions or services has access to increasingly sensitive parts of somebody’s life.<\/p>\n\n\n\n R\u016bpest\u0117lis approaches that problem through a clear principle:<\/p>\n\n\n\n the older person remains the decision-maker.<\/strong><\/p>\n\n\n\n A conversation with the assistant is not treated as unlimited authorisation.<\/p>\n\n\n\n Different activities can require different permissions.<\/p>\n\n\n\n The individual should decide what information can be shared.<\/p>\n\n\n\n What can be remembered.<\/p>\n\n\n\n Which actions may be performed.<\/p>\n\n\n\n Who may be contacted.<\/p>\n\n\n\n And when the service should stop.<\/p>\n\n\n\n Family members can be involved.<\/p>\n\n\n\n Care workers can be involved.<\/p>\n\n\n\n But paying for a service does not automatically transfer control over the person receiving it.<\/p>\n\n\n\n \u201cThe person who receives the assistance must remain central to the consent and service relationship, regardless of who pays,\u201d Margelis says.<\/p>\n\n\n\n That may become one of the most important parts of the product.<\/p>\n\n\n\n Technology designed for older adults can easily slide into surveillance.<\/p>\n\n\n\n A family member may have good intentions.<\/p>\n\n\n\n They want to know that a parent or grandparent is doing well.<\/p>\n\n\n\n They want reminders to work.<\/p>\n\n\n\n They want to be notified if something needs attention.<\/p>\n\n\n\n But there is a fundamental difference between helping someone and monitoring them.<\/p>\n\n\n\n R\u016bpest\u0117lis explicitly tries to preserve that distinction.<\/p>\n\n\n\n \u201cAn older person\u2019s dignity and independence should not diminish simply because they are receiving assistance,\u201d Margelis says.<\/p>\n\n\n\n Family involvement can be useful.<\/p>\n\n\n\n But it should not automatically override the older person’s choices.<\/p>\n\n\n\n For R\u016bpest\u0117lis, consent is therefore not simply a legal checkbox.<\/p>\n\n\n\n It is part of product design.<\/p>\n\n\n\n One of the more difficult questions for any AI assistant is what happens when the machine should no longer be the one handling the situation.<\/p>\n\n\n\n R\u016bpest\u0117lis can support coordination or escalation where a particular service includes an authorised human contact or operator.<\/p>\n\n\n\n But Margelis deliberately avoids making vague promises around \u201chuman handoff.\u201d<\/p>\n\n\n\n There is an important difference between sending a notification<\/strong> and confirming that someone has actually provided assistance<\/strong>.<\/p>\n\n\n\n A message to a family member does not guarantee that the family member saw it.<\/p>\n\n\n\n An escalation does not mean the issue has been resolved.<\/p>\n\n\n\n The operational arrangement therefore depends on the particular service and workflow.<\/p>\n\n\n\n And, according to Margelis, those arrangements have to be verified before they are promised to users.<\/p>\n\n\n\n That is a notably conservative approach for an AI product.<\/p>\n\n\n\n But for a service aimed at older adults, ambiguity about whether a human has actually taken over could be much more dangerous than simply admitting that the system does not know.<\/p>\n\n\n\n The same caution applies to health and emergencies.<\/p>\n\n\n\n R\u016bpest\u0117lis is not a medical device<\/strong>.<\/p>\n\n\n\n It is not an emergency dispatch service<\/strong>.<\/p>\n\n\n\n And it is not a replacement for healthcare professionals<\/strong>.<\/p>\n\n\n\n A person who needs urgent emergency assistance should contact 112<\/strong>.<\/p>\n\n\n\n The company is working on appropriate safety guidance and possible human escalation, but Margelis does not claim that an AI conversation can guarantee an emergency response.<\/p>\n\n\n\n That boundary is important because users may not always think in product categories.<\/p>\n\n\n\n If an assistant is available every day and speaks naturally, somebody may instinctively begin to treat it as a general source of help.<\/p>\n\n\n\n The responsibility of the product is therefore not only to explain what it can do.<\/p>\n\n\n\n It must also make clear what it cannot<\/strong> do.<\/p>\n\n\n\n R\u016bpest\u0117lis sees several possible customer groups.<\/p>\n\n\n\n Older people may subscribe themselves.<\/p>\n\n\n\n Family members may purchase the service for a relative.<\/p>\n\n\n\n Care organisations or community-support providers may use it as part of a wider service.<\/p>\n\n\n\n But those different payment models do not change the central principle.<\/p>\n\n\n\n The older person receiving the assistance remains central to consent.<\/p>\n\n\n\n That creates an interesting business-model challenge.<\/p>\n\n\n\n In many consumer products, the customer and user are the same person.<\/p>\n\n\n\n With R\u016bpest\u0117lis, they may not be.<\/p>\n\n\n\n A son or daughter may pay.<\/p>\n\n\n\n A care organisation may procure the service.<\/p>\n\n\n\n But the person answering the phone is the one whose autonomy the product has to protect.<\/p>\n\n\n\n R\u016bpest\u0117lis currently publishes a private consumer price of \u20ac29.99 per month<\/strong>.<\/p>\n\n\n\n The objective is to make routine assistance affordable enough for ordinary use while using automation to keep the service efficient.<\/p>\n\n\n\n But the economics are connected directly to the safety model.<\/p>\n\n\n\n The more situations that require human intervention, the more operational cost enters the service.<\/p>\n\n\n\n The more tasks the AI performs autonomously, the more carefully those boundaries have to be defined.<\/p>\n\n\n\n Margelis does not disclose the company’s proprietary cost structure or supplier arrangements.<\/p>\n\n\n\n Instead, he frames sustainable pricing around three things:<\/p>\n\n\n\n operational performance;<\/p>\n\n\n\n responsible service scope;<\/p>\n\n\n\n and continuous quality improvement.<\/p>\n\n\n\n The challenge is not simply making an AI inexpensive.<\/p>\n\n\n\n It is making the entire service dependable at a price people can actually pay.<\/p>\n\n\n\n R\u016bpest\u0117lis is already built, tested and operating in Lithuania<\/strong>, with payments integrated through Stripe<\/strong>.<\/p>\n\n\n\n According to Margelis, as of 27 October 2026<\/strong>, the assistant performs 42 functions<\/strong>.<\/p>\n\n\n\n A total of 196 functions have been developed<\/strong>, including capabilities that are still awaiting connection or activation.<\/p>\n\n\n\n That distinction matters.<\/p>\n\n\n\n The company separates what the service can perform today from functions that have already been developed technically but are not yet part of the active service.<\/p>\n\n\n\n The next stage is therefore not simply building more features.<\/p>\n\n\n\n It is connecting and activating additional capabilities while expanding real-world use of the existing product.<\/p>\n\n\n\n R\u016bpest\u0117lis has also presented its complete municipal service package.<\/p>\n\n\n\n Margelis reports interest from 43 of Lithuania\u2019s 60 municipalities<\/strong>.<\/p>\n\n\n\n That number represents expressed interest rather than 43 signed contracts, but it points toward a potentially important second market for the company beyond private consumers and families.<\/p>\n\n\n\n And some of that municipal interest is beginning to translate into concrete referrals.<\/p>\n\n\n\n By 11:25 a.m. on 27 October<\/strong>, according to Margelis, Neringa Municipality had submitted eight people wishing to use R\u016bpest\u0117lis<\/strong>.<\/p>\n\n\n\n Again, the distinction is important.<\/p>\n\n\n\n A referral is a step toward activation.<\/p>\n\n\n\n It does not mean that all eight people are already receiving calls from the service.<\/p>\n\n\n\n That careful separation between interest, referral, activation and actual service delivery follows the same philosophy R\u016bpest\u0117lis applies inside the product itself:<\/p>\n\n\n\n do not describe something as completed until it actually is.<\/strong><\/p>\n\n\n\n The company has been demonstrating R\u016bpest\u0117lis to potential users as well as public-sector organisations.<\/p>\n\n\n\n Its immediate challenge is now increasingly one of deployment rather than proving that a voice assistant can technically exist.<\/p>\n\n\n\n The service already operates.<\/p>\n\n\n\n Payments already work.<\/p>\n\n\n\n Dozens of functions are active.<\/p>\n\n\n\n Many more have been developed.<\/p>\n\n\n\n Municipalities are showing interest.<\/p>\n\n\n\n Potential users are beginning to be referred.<\/p>\n\n\n\n The next stage is expanding use of the operating service while connecting additional capabilities that have already been built.<\/p>\n\n\n\n But Margelis remains careful about the distinction between live functionality and future integrations.<\/p>\n\n\n\n Optional integrations and features still undergoing verification are not presented as if they were already active.<\/p>\n\n\n\n That reflects the same principle seen elsewhere in the product.<\/p>\n\n\n\n Do not present something as completed before it actually is.<\/p>\n\n\n\n Over the coming year, the company wants to deepen the practical value of the service and build trusted relationships with both individual users and organisations.<\/p>\n\n\n\n There is no stated ambition to build a giant general-purpose language model.<\/p>\n\n\n\n Margelis sees that as somebody else’s problem.<\/p>\n\n\n\n The world’s largest AI companies are spending billions building increasingly capable general-purpose models.<\/p>\n\n\n\n A small Lithuanian startup has little reason to compete with them directly.<\/p>\n\n\n\n R\u016bpest\u0117lis is taking the opposite approach.<\/p>\n\n\n\n Use increasingly powerful underlying AI technology.<\/p>\n\n\n\n Then specialise around a very specific user.<\/p>\n\n\n\n A specific language.<\/p>\n\n\n\n A specific interface.<\/p>\n\n\n\n A specific set of needs.<\/p>\n\n\n\n And, perhaps most importantly, a specific set of boundaries.<\/p>\n\n\n\n \u201cOur aim is not to compete with the world\u2019s largest AI companies in building general-purpose foundation models,\u201d Margelis says.<\/p>\n\n\n\n \u201cIt is to offer a specialised service that helps real people with real everyday needs.\u201d<\/p>\n\n\n\n This is becoming an increasingly important model for AI startups.<\/p>\n\n\n\n The foundation model may become infrastructure.<\/p>\n\n\n\n The actual product is what gets built around it.<\/p>\n\n\n\n R\u016bpest\u0117lis is easy to describe as a voice AI assistant for older adults.<\/p>\n\n\n\n But the founder interview suggests that the more interesting product problem lies elsewhere.<\/p>\n\n\n\n The difficult questions are not simply:<\/p>\n\n\n\n Can the AI speak Lithuanian?<\/p>\n\n\n\n Can it answer a question?<\/p>\n\n\n\n Can it make a phone call?<\/p>\n\n\n\n The harder questions are:<\/p>\n\n\n\n When should it ask instead of assume?<\/p>\n\n\n\n When is it authorised to act?<\/p>\n\n\n\n How does it prove that an action actually succeeded?<\/p>\n\n\n\n When should a human become involved?<\/p>\n\n\n\n How does it preserve a person’s independence when a relative is paying?<\/p>\n\n\n\n What happens when somebody asks the system for help it cannot safely provide?<\/p>\n\n\n\n And how do you explain those boundaries clearly enough that a user understands them?<\/p>\n\n\n\n Those questions are less spectacular than a polished AI demonstration.<\/p>\n\n\n\n They are also much closer to what determines whether such a product can be trusted.<\/p>\n\n\n\n The central idea behind R\u016bpest\u0117lis is not that older people need more technology.<\/p>\n\n\n\n It is almost the opposite.<\/p>\n\n\n\n They should not have to become technology experts just to benefit from what modern software can do.<\/p>\n\n\n\n The interface can be something they already understand.<\/p>\n\n\n\n A telephone.<\/p>\n\n\n\n A conversation.<\/p>\n\n\n\n A familiar language.<\/p>\n\n\n\n Underneath that simplicity, however, sits a much more difficult product challenge: deciding how far the technology should be allowed to go.<\/p>\n\n\n\n R\u016bpest\u0117lis is trying to answer that question by keeping the user at the centre.<\/p>\n\n\n\n Not the family member.<\/p>\n\n\n\n Not the care organisation.<\/p>\n\n\n\n Not the AI.<\/p>\n\n\n\n The person.<\/p>\n\n\n\n As the company summarises its own philosophy:<\/p>\n\n\n\nStart with the person, not the interface<\/h3>\n\n\n\n
More than a conversation<\/h3>\n\n\n\n
An AI should be able to say \u201cI don\u2019t know\u201d<\/h3>\n\n\n\n
Lithuanian conversation is not a clean transcript<\/h3>\n\n\n\n
The older person remains the decision-maker<\/h3>\n\n\n\n
Help from family should not mean loss of autonomy<\/h3>\n\n\n\n
Human handoff is not the same as human rescue<\/h3>\n\n\n\n
R\u016bpest\u0117lis is not an emergency service<\/h3>\n\n\n\n
Who pays is not the same as who controls the service<\/h3>\n\n\n\n
\u20ac29.99 per month<\/h3>\n\n\n\n
Already operating \u2014 and expanding what the service can do<\/h3>\n\n\n\n
Already operating, but deliberate about what comes next<\/h3>\n\n\n\n
Don’t compete with the foundation models<\/h3>\n\n\n\n
The difficult part may not be the AI<\/h3>\n\n\n\n
Assistance on your terms<\/h3>\n\n\n\n