The most interesting aspect of Julia is not that Roma Capitale built a chatbot. In 2026, a chatbot on its own is no longer news. The interesting part is the ambition the project has taken on over time: transforming artificial intelligence into a new gateway to the city and, progressively, to the administrative machinery.

Julia was created with a relatively narrow goal: helping those who live in, visit, or pass through Roma navigate mobility, culture, events, tourism, and essential services. In just over a year, however, the scale of the project changed. From its first release in March 2025, it moved to version 2.0, and then to version 2.5 in June 2026, integrating directly into Roma Capitale's institutional portal. It is here that Julia stopped being just a digital guide and became a more significant experiment: determining whether natural language can replace at least part of the traditional navigation of the Pubblica Amministrazione.

It is a transition worth examining. Roma Capitale's portal logs around 18 million visits a year, hosting more than 160,000 informational entries and around 90 online services. For citizens, the problem is rarely a lack of information: it is finding it, figuring out which page is up to date, which office is responsible, what documents are required, and what the next step is. Julia aims to tackle precisely this friction.

From tourist guide to city interface

The project was announced in October 2024, during Microsoft's AI Tour in Roma, as a virtual tourist guide developed with the contribution of Microsoft and OpenAI technologies. The initial idea was straightforward: provide tourists and citizens with a multilingual assistant capable of querying data from local operators and accredited sources, suggesting itineraries, and providing information on transport, museums, restaurants, events, and lesser-known sights.

The first public release arrived on March 7, 2025. Roma Capitale introduced Julia as its first urban generative artificial intelligence: an assistant available on WhatsApp, Telegram, Messenger, and web chat, capable of communicating in over 80 languages. The technology powering the initial version used GPT-4o, engineered for the project with Microsoft's support, while NTT DATA worked on developing the first release alongside the other partners.

Even in version 1.0, the most interesting architectural feature emerges: Julia is not designed to freely search for any information across the web. Instead, responses are built from a certified knowledge base made up of official Roma Capitale data and other validated sources. It is a less spectacular choice than the promise of an assistant that "knows everything," but far more sensible for a public service.

The real value: not knowing everything, but knowing where the answer comes from

Generative artificial intelligence has a structural problem: it is very good at producing linguistically convincing answers even when the information is incomplete or wrong. In a creative engine, this flaw may be tolerable. In an official city assistant, it is much less so.

If a municipal system gets an exhibition's schedule wrong, the damage is limited. If it gets an administrative procedure, a deadline, a mobility regulation, or information tied to a public service wrong, the response carries a different weight precisely because it comes from an institutional body. Users naturally tend to place greater trust in it.

This is why the idea of a certified knowledge base is probably Julia's strongest starting point. The goal should not be turning the municipality into a local version of ChatGPT, but building a system that thoroughly understands a defined set of information, cites its sources, and recognizes its own limits.

The case of New York shows just how concrete this issue is. The MyCity chatbot, launched to help businesses navigate rules and services, was found in 2024 to be providing incorrect guidance even on labor laws, housing, and commercial obligations. In some tests, it went as far as suggesting unlawful behavior. The experiment became a textbook case: when a generative response carries the seal of a public administration, a hallucination is not merely a technical error, but an issue of institutional trust.

Julia starts from a more prudent philosophy. That does not mean it is immune to errors: Roma Capitale itself warns that, like any AI-powered chatbot, it can make mistakes. It does mean, however, that the project attempts to reduce risk through the quality and validation of its sources rather than relying solely on the model's general capability.

Julia 2.0: memory, geolocation, and the first agents

With version 2.0, unveiled on July 31, 2025, the project picks up the pace. The model transitions from the GPT-4o family to GPT-4.1, the system prompt is revamped, and new specialized agents appear. Roma Capitale announces three additional agents integrated into a new multi-agent architecture.

The concept is an important one. Instead of assigning every question to a single general-purpose system, the architecture distributes the workload among domain-specialized components. It is an increasingly common logic in agentic AI: an orchestrator understands the request, identifies the required skills, and coordinates different modules before returning a single response.

Version 2.0 also introduces geolocation on WhatsApp, Telegram, and web chat. A user can share their location and receive contextual information about what is nearby. New information domains are added, including public green spaces, places of worship, sports and leisure, news, and updates on the city and mobility. Existing information is also improved, such as shared mobility, ZTL, study rooms, Roma Pass, and MIC Card.

This is where Julia begins to look less like a “tourism bot” and more like a conversational urban infrastructure. A question like “where can I study near me until late?” combines location, hours, services, and context. It is precisely the kind of request that a traditional website handles poorly and that a conversational interface can make much simpler.

The leap to 2.5: Julia enters Roma Capitale

Version 2.5, presented in June 2026, is probably the most significant step so far. Julia is integrated directly into the institutional portal. The widget remains available while browsing, retains conversation context, and can return direct links to official pages, documents, and sources.

The point is not cosmetic. A portal with 160,000 pieces of content cannot be genuinely simple through better menus alone. The complexity mirrors that of the administration: departments, boroughs, regulations, procedures, services, public tenders, forms, and contacts. A conversational layer can instead allow the user to start from the problem rather than the bureaucratic structure.

The difference is substantial. A citizen shouldn't necessarily have to know whether a procedure falls under a department, a municipality, or a specific office. They should be able to ask: “I need to do this, what documents do I need, and where do I have to go?”. If Julia can translate organizational complexity into an understandable journey, its value is not technological: it is administrative.

In 2.5, the architecture evolves further toward specialized agents coordinated by an orchestrating Super Agent. Official sources also point to new information areas, ranging from cinema schedules to post offices, public water dispensers, waste bins, Punti Roma Facile, and anti-violence centers. The project is being developed with an ecosystem of partners that includes, across various releases, Microsoft, Intellera, NTT DATA, Digitalia, BIP, Fastweb, and 01 Sistemi.

This is where Public Administration can truly change

To understand Julia's potential, we need to take the words “artificial intelligence” off the table for a moment. The problem it is trying to solve has existed for decades: administrations possess vast amounts of information but present it according to internal logic, whereas citizens think in terms of needs.

“I got a fine”, “I need to change my residence”, “I want to know if the ZTL is active”, “I'm looking for an anti-violence center”, “where can I practice sports with a voucher?”, “what documents do I need for this procedure?”. These are questions that cut vertically across different databases and organizational structures.

A well-designed conversational system can become a translator between these two worlds. It does not decide on behalf of the administration and does not replace civil servants; it reduces the cognitive cost required to obtain the correct information.

This is perhaps the most critical distinction to uphold. Public AI is at its best when it helps citizens understand, search, and prepare. It becomes far more delicate when it shifts from informational support to decision-making: granting or denying a benefit, classifying an individual, establishing priorities, or producing administrative effects. Today, Julia operates primarily on the former ground, and that is an advantage.

Privacy: a serious approach, but communication must be crystal clear

Julia’s privacy policy contains several notable choices. Roma Capitale states that user data is not used to train the chatbot and that, under normal use, personal data is retained for a maximum of 24 hours, strictly limited to the time needed to retrieve the conversation. Filing a report with 060606 or 060608 operators, however, involves a separate transfer to the CRM, subject to user consent.

There is also an important distinction between the official web chat and third-party channels. If Julia is used via WhatsApp, Messenger, or Telegram, the respective platforms’ policies also come into play. This point should be highlighted prominently, especially since one of Julia’s key strengths is precisely its availability on channels users already use.

One element still warrants a clearer explanation. The presentation of Julia 2.0 mentioned a conversation history extended to seven days; the current privacy policy, however, sets a maximum of 24 hours for retaining personal data during normal chatbot use. The two statements may be technically compatible—for example, through a distinction between conversational state and personal data—but for a non-technical user, this distinction is not immediately obvious. A public project should make even this layer transparent.

The cost issue: metrics are needed, not just releases

A public AI infrastructure cannot be evaluated solely on the quality of its demo. It must also be assessed on cost and, above all, on the results it delivers.

Roma Capitale’s PNRR pages outline, within the broader digital tourism transformation ecosystem, a 1.7 million euro measure for “The Roman Land” app, designed to integrate with the Tourist Virtual Assistant platform. It is inaccurate to use that figure as the “cost of Julia”: it is a broader, linked project, not a declared total for the assistant. Yet this very distinction highlights why publishing an easily readable financial breakdown of the service over time would be valuable.

How much does it cost to handle a thousand conversations? What is the impact of the model, cloud, database maintenance, testing, and integrations? How much work is saved for the contact center? How many people manage to complete a procedure without calling or physically visiting a service desk? How many answers are deemed incorrect? How many are escalated to a human operator?

These are the metrics that separate a demonstration project from true public digital infrastructure.

The benchmark should not be ChatGPT, but Singapore

To compare Julia with other public systems, looking at commercial chatbots is not enough. An interesting reference point is Singapore, where GovTech manages VICA, a conversational platform shared by more than 60 government agencies and used across over 100 chatbots. GovTech reports an average of over 800,000 monthly queries.

The figure is not meant to rank cities. It serves to show how evaluation changes when AI becomes infrastructure: number of entities served, request volume, operational continuity, satisfaction, resolution rate, reduction of manual labor. Singapore measures the system as a digital public service, not as a tech showcase.

This is the transition that Roma should make with Julia. After the release phase, the focus should turn to public, comparable results.

A smart choice: no advertising, no opaque commercial ranking

Another feature of Julia deserves protection. Roma Capitale presented the service as free of advertising, whether explicit or covert. Businesses can register their activity for free on the 060608 portal so that it enters the assistant’s knowledge base.

It is an important choice because an urban assistant can very quickly become an economic intermediary. If we ask “where can I eat nearby?” or “which hotel do you recommend?”, the system determines visibility and potentially revenue. The neutrality of response criteria must therefore be treated as part of the service’s governance.

In the future, it would be useful to make public the principles by which Julia selects or ranks equivalent businesses: distance, relevance, opening hours, accessibility, data completeness, rotation of results. The more widely the assistant is used, the more economic significance these rules acquire.

The biggest risk is not that Julia makes mistakes. It is that we stop verifying it

Every generative system will make mistakes, sooner or later. Expecting zero errors is unrealistic. The quality of a public project is measured by how errors are prevented, identified, and corrected.

For Julia, this means building an ongoing evaluation system: critical test batteries, regression tests with every release, source monitoring, user feedback, audits on the most sensitive domains, and a straightforward mechanism for handoff to a human when the answer is not sufficiently certain.

Including links and official sources in responses is already a positive step, as it allows users to verify information. But the next step could be making the confidence level explicit, or at least clearly distinguishing between verified facts, recommendations, and content that requires confirmation from an administrative office.

The next challenge: from an assistant that informs to an assistant that guides

The real transformation will come if Julia can guide citizens through a process without becoming the decision-maker. Today, it can explain which documents are needed. Tomorrow, it could verify that the user has understood the steps, directly open the correct service, prefill non-sensitive information, send deadline reminders, or pass the context to a human operator.

This is the compelling frontier of AI in the PA: not replacing the administration, but making part of its complexity invisible.

For a city like Roma, the potential is particularly significant. The capital combines nearly three million residents, millions of city users, a global tourist influx, an enormous cultural heritage, and a highly complex administrative machine. It is precisely the kind of environment where a conversational interface can deliver real value if properly governed.

Why Julia is a project worth watching

Julia is still a young project. It would be premature to describe it as a completed revolution of the Pubblica Amministrazione. But it would be just as reductive to dismiss it as yet another institutional chatbot.

The trajectory is interesting: certified sources rather than the indiscriminate web, progressive specialization of agents, integration with channels citizens already use, geolocation, direct access to the institutional portal, and a gradual expansion from tourism content to administrative services.

Roma is doing something that many administrations announce and few actually manage to put into production: using a generative model as an operational interface for a real city, with real data and real citizens.

Now comes the hardest part. The next release of Julia will matter less than the numbers Roma will be able to publish regarding its usefulness. Accuracy, resolution rate, time saved, accessibility, escalation to operators, satisfaction, and costs should become as much a part of the narrative as the model used or the number of agents.

If this step is achieved, Julia could be something more than a good urban innovation project. It could become a replicable model of how artificial intelligence enters the Pubblica Amministrazione without turning the administration into a black box: technology out front, sources and public accountability behind it.

Sources