How Jedro uses AI
Jedro drafts. The agency decides.
Jedro uses an AI model to read property enquiries, draft the agency’s replies, and answer the calls an agent misses. This page says what the AI does, what it is never allowed to do, who approves what, and how Jedro meets the EU AI Act and the GDPR, in every EU country. Jedro is being built and opens to agencies in November 2026; where something below is not built yet, the line says so.
In short
Jedro is an AI system that drafts replies to property enquiries from buyers and sellers, asks the questions that qualify them, suggests viewing times, keeps an agency’s pipeline up to date, and answers by voice the calls an agent misses. It does not name a price other than the advertised one, write contract text, send a property’s exact address on its own, or make decisions about people.
Every message it drafted says so, in the recipient’s language, and its voice says so in its first sentence. What it may send without a person approving each message is decided by the agency, one action at a time — and nothing is on that list by default.
What the AI does
Only work an agent would otherwise do by hand, and only on the agency’s own enquiries.
Reads each enquiry
The language it is written in, the property it is about and what the buyer is asking, so it lands in the right queue with the right listing.
Drafts the reply
In the buyer’s language and the agency’s voice, from the listing’s facts and the agency’s own rules. A person at the agency reads it, changes it if they want, and sends it — or it goes out under a switch the owner has turned on.
Suggests viewing times
From the agent’s calendar. A slot is offered to a buyer only once the draft that carries it has been approved, or under a switch the owner has turned on.
Answers the calls an agent misses
Each agent chooses for their own phone: the calls they miss, which is the default, or every call. On missed calls only, if the agent picks up, Jedro is not on the call. It never calls anyone. Its voice says it is the agency’s AI assistant, asks before it notes anything down, finds out what the caller needs, asks the first questions, and agrees the next step: a message on WhatsApp if the caller says yes, a time held, or a call from the agent. Being rolled out one language at a time; until the voice answers in a caller’s language, they hear the agent’s recorded greeting.
Summarises the day for the agent
What came in, what is waiting and what is overdue. The summary is for the agent; the buyer never sees it.
Drafts listing posts
Text for a listing’s posts, from its facts and photos. Nothing is published by Jedro; an agent publishes it, or does not.
What it is never allowed to do
These are not switches. No setting turns them on, for any plan.
- name a price other than the advertised one, offer a discount or negotiate
- write contract text — contracts are the agency lawyer’s paper, filled in, never written
- send a property’s exact address on its own — an agent sends it, with a confirmed viewing
- answer a legal question from anything but the lawyer’s own rules — it refuses the rest and hands it to a person
- decide whether a buyer is served, refused or put first — it suggests, a person decides
- evaluate, rank or make decisions about the agency’s staff — Performance shows counts of what happened, not a model’s judgement of anyone
- read emotions, in writing or in a voice, or sort people by ethnicity, religion, health, sexual orientation or any other sensitive characteristic
- recognise a person by their voice, or keep the sound of a call its voice answered — only the words are kept
- use clients’ messages to train a model — ours, Anthropic’s or anyone’s
- say it is a person when someone asks
Who approves what
The agency decides what Jedro may send on its own, one action at a time. Nothing is switched on by default: until the owner turns a switch on, a person at the agency approves each message.
Sending alone works through switches, one per action: alone, asks first, or never. Every switch starts at asks first. The agency owner decides what may go alone; an agent can tighten a switch, never loosen it.
A new rule’s first week lands in Review, so a person sees what it would have sent before it sends anything. Anything sent alone has six seconds of undo and leaves a receipt: what was sent, under which rule, and when.
Any switch can be set back to asks first at any time. It applies from the next message.
The voice is not one of those switches: each agent chooses, for their own phone, whether it answers the calls they miss or every call, because a call cannot wait for approval. On a call it does only what the agency’s other switches already allow, and anything else it can only promise in writing, where those switches decide. It writes to a caller afterwards only if they said yes.
Telling people it is AI
Article 50 of the EU AI Act says people must be told when they are dealing with an AI system, and that AI-generated text, audio and images must be marked so software can detect them. It has applied since 2 August 2026.
A line on every message
Every message Jedro drafted carries a line saying it was written with AI and sent by the agency, in the recipient’s language. The agency chooses among fixed wordings; it cannot remove the line.
On a call
Jedro’s voice says it is the agency’s AI assistant in its first sentence, tells the caller it is noting down what they say, and waits for a yes before going on. A caller who says no, or who asks for a person, gets a call from the agent.
In the agency’s name
A reply a person has read and sent goes out under that person’s name. A reply Jedro sends on its own opens by saying it comes from the agency’s AI assistant, and names the person who can step in. Neither advertises Jedro. The buyer’s relationship is with the agency.
If someone asks
If a buyer asks whether they are talking to a person, in writing or on a call, the answer is the truth, and a person at the agency takes the conversation over.
A record behind every message
Each message is stored with where it came from: drafted by AI or not, changed by a person or not, approved by whom, or sent alone under which rule. The agency can see this for any message, for as long as the message is kept.
A machine-readable mark
Where a channel lets a message carry metadata, as email does, AI-drafted text will be marked there too, so software can tell as well as people. This is being built and will be in place before Jedro sends its first message for an agency. The Commission’s guidance does not yet say how live speech on a telephone line carries such a mark. The voice says what it is aloud on every call, and no audio of the call is kept.
The model underneath
Drafts are written by Claude, a general-purpose AI model made by Anthropic, through its API. Anthropic is the provider of that model and carries its own obligations under the AI Act, which have applied to general-purpose models since 2 August 2025.
Anthropic processes a message to draft the reply and does not use it to train its models. The message is sent to the United States; how that transfer is safeguarded is on the privacy page.
The voice turns speech into text and text into speech with further providers. They are not chosen yet. Before the voice answers a single call, this page and the privacy page name them, with where they process the call.
If we change the model, or the company behind it, this page and the privacy page change first.
Where Jedro sits under the AI Act
The AI Act — Regulation (EU) 2024/1689, as amended by the Digital Omnibus on AI, Regulation (EU) 2026/1744, in force since 27 July 2026 — gives each party a role. Vila Markotić d.o.o. is the provider of Jedro, the AI system. The agency that uses it is the deployer. Anthropic is the provider of the general-purpose model Jedro is built on.
Jedro is not a high-risk AI system. It is not used to assess creditworthiness, to recruit, evaluate or make decisions about workers, to decide access to essential services, or for anything else listed in Annex III. It is built to stay that way: whatever would put it there is on the list of what it is never allowed to do.
None of the practices Article 5 prohibits is part of Jedro — among them manipulation, exploiting people’s vulnerabilities, social scoring, emotion recognition at work and biometric categorisation.
What does apply to Jedro are the transparency duties of Article 50 and the AI literacy duty of Article 4. Both are on this page.
The dates, and what we do by each
- Prohibited practices and AI literacy apply. Jedro uses none of the first; the second is below.
- Duties for general-purpose AI models apply — Anthropic’s, for the model Jedro is built on.
- Transparency duties apply. Every message Jedro drafts says so, and its voice says so on every call.
- Machine-readable marking has applied since 2 August 2026. Systems that were already on the market before that day have until this date. Jedro was not, so its marking is in place from the first agency’s first day. Two prohibitions added in July 2026, on child sexual abuse material and on intimate content made without consent, apply from this date too. Jedro makes no images, audio or video of people.
- Rules for the high-risk systems in Annex III apply, postponed from 2 August 2026. Jedro is not one; we check that again before this date.
Dates as set by the AI Act as amended in July 2026. If they move again, this list moves with them.
Automated decisions and your data
Jedro makes no decision that has a legal or similarly significant effect on anyone by automated means alone, in the words of Article 22 of the GDPR. It suggests a reply, a slot or a next step; no one is turned away and no one is ranked, and a person at the agency can be reached at the first asking.
What personal data Jedro handles, where it is kept, for how long and how to have it deleted is on the privacy page.
Knowing how it works
Article 4 of the AI Act, as amended in July 2026, asks everyone who provides or uses an AI system to take measures that support the AI literacy of the people working with it; it does not set a level anyone must reach. Everyone at Jedro who builds or supports it is trained on what the model can and cannot do, and where it tends to be wrong.
Every agency is onboarded with the same: how to read a draft, when to change it, what each switch does, what its voice says on a call and what Jedro will never do. An agency that wants it in writing for its own staff can ask for it.
Country by country
The AI Act is an EU regulation, so it applies directly in every member state, with no national law needed to bring it in. Each country names its own authorities; these are the five we have checked so far: Spain, France, Italy, Portugal and Croatia.
Spain has set up AESIA, the Spanish agency that supervises artificial intelligence, and the AEPD protects personal data. The bill that would have made AESIA the main market surveillance authority and set penalties reached parliament on 28 May 2026 and lapsed when parliament was dissolved on 6 October 2026, ahead of the general election of 29 November. As far as we can establish on 5 November 2026, no law yet designates Spain’s market surveillance authority.
France has not yet adopted the law that names its authorities for the AI Act. The government’s bill adapting French law to EU law was adopted by the Senate on 18 February 2026 and has been before the National Assembly since 20 February 2026. In the Senate’s text, the DGCCRF, the consumer protection authority, and Arcom, the regulator for audiovisual and digital communication, would both check that people are told when they are dealing with an AI; the CNIL, which protects personal data, would be one of the supervising authorities; and fines would follow the AI Act’s own ceilings. The Défenseur des droits is among the bodies France has named to protect fundamental rights under Article 77. As far as we can establish on 5 November 2026, the Assembly has not debated the bill, and the European Commission’s list shows the DGCCRF as France’s contact point, with the national decision still pending.
In Italy, Law 132/2025, in force since 10 October 2025, names AgID and ACN, the national cybersecurity agency, as the authorities for artificial intelligence: AgID notifies the bodies that check AI systems, and ACN supervises the market. Legislative Decree 179/2026, in force from 23 October 2026, sets out their powers and the penalties, and makes the Garante, which protects personal data, the supervisor of high-risk AI used in policing, migration, border control and justice.
Portugal has named the bodies that protect fundamental rights under Article 77, among them ANACOM, the communications regulator. In September 2025 the government said it intends to make ANACOM the market surveillance authority, and the European Commission’s list shows it, with the national decision still pending. The government’s national AI agenda (Council of Ministers Resolution 2/2026, of 8 January 2026) scheduled the choice of authorities and the penalty rules for the first half of 2026. As far as we can establish on 5 November 2026, no law doing so has been published, so Portugal has no national penalty rules under the AI Act yet.
Croatia has named the bodies that protect fundamental rights under Article 77, among them AZOP, the data protection authority, and the Ombudswoman. As far as we can establish on 5 November 2026, the national law that designates the market surveillance authority and sets penalties has not been adopted, and no draft has been put to public consultation. When it is adopted, this page names the authority.
Jedro works for agencies across Europe, starting in Spain, France, Italy, Portugal and Croatia; buyers and sellers write from everywhere. The disclosure goes out in each person’s own language, and this page is in English, French, Spanish, Italian and Croatian.
When it gets something wrong
Every draft a person changes or rejects is counted, and the agency sees how often Jedro was overridden, and on what.
If Jedro sent something it should not have, or you think it treated you unfairly, write to hello@jedro.ai. A person reads it, and we answer within one month.
If you wrote to an agency, or called it, and the answer looked wrong, tell the agency — the answer was given in its name and it can put it right. You can write to us as well.
If it concerns your personal data, you can also complain to AZOP — Agencija za zaštitu osobnih podataka, Ulica Metela Ožegovića 16, 10000 Zagreb, azop.hr — or to the data protection authority where you live: the AEPD in Spain, the CNIL in France, the Garante in Italy, the CNPD in Portugal.
Changes to this page
If this changes, the date at the top changes with it. We review the page before each date above, and before Jedro may do anything new on its own.