Skip to content
Giallo

Local government chatbot: a help desk that answers even when the public information office is closed.

This page is for mayors, municipal secretaries and the heads of the public information office (URP), registry services and digital transition. We walk through a day at the help desk as seen by the agents we build: what they do at each hour, which model we use and where they stop to make way for the offices. Internal records management and case processing are covered on the public administration page.

A circular town square with a white bell tower connects the municipality's requests and services. Small yellow dots converge from the edges toward the center

What can a chatbot do for a municipality?

A local government chatbot answers citizens about services, documents, opening hours and procedures using only the pages the authority has approved, collects reports of local issues and prepares appointment requests. Giallo Studio custom-designs it with Claude, GPT and Gemini, connects it to the website and the offices, and hands every case that needs an assessment to a person.

A working day, agent by agent.

The local issue reporting agent on the town map: the pothole photo arrives with the license plate already covered, the second report at the same spot is linked to the first instead of duplicating it, the broken streetlight goes to the lighting operator and the citizen receives a reference to follow the case.
  1. From early morning, before the URP opens its doors, and all day long.

    Website help desk agent

    Answers citizens' questions about municipal services using only the pages and documents published by the authority, with a link to the source.

    1. Introduces itself as the municipality's automated assistant and says when it's better to call or go to the desk.
    2. Looks for the answer in the pages of the official website, the regulations and the service sheets approved by the URP.
    3. Answers in a few lines and attaches the link to the official page the information came from.
    4. If the answer isn't in the sources, it says so and offers the contact of the right office instead of improvising.
    5. Hands over to a staff member any conversation about personal situations, such as evictions, benefits or social services.

    Connects to: Municipality's official website · URP service sheets · Online forms

    Claude Sonnet
    Works well with long instructions and refusal rules: it answers only with what's in the authority's sources and keeps an institutional tone even with an angry citizen.
    Claude Haiku
    Sorts incoming questions by service and immediately spots personal ones to pass to a person, so the more capable model works only on the answers.

    Who approves. The URP approves the sources the agent can answer from and the list of topics that always go to a staff member; every new source goes through the office manager.

  2. When the desks open and requests start coming in for ID cards, certificates and changes of residence.

    Registry services appointment agent

    Guides citizens to the right channel and collects appointment requests with the documents already checked.

    1. Works out what's really needed: a certificate the citizen can download on their own from the national resident registry, a desk appointment or a procedure to start online.
    2. For the electronic ID card, it points to the booking procedure the office uses, with the list of documents to bring.
    3. For other procedures it offers free slots in the desk calendar, if the municipality has one that can be connected.
    4. Sends a reminder the day before and handles cancellations and changes written in free form.

    Connects to: Desk calendar · Links to the national resident registry's online services · WhatsApp Business Platform

    Claude Haiku
    Requests are short and numerous, like “can I move it to Thursday?”: you need a fast, low-cost model that understands dates and intentions without hesitation.
    Codex
    It doesn't see citizens' data: we use it to write and maintain the connector to the desk calendar, with code reviewed by our team.

    Who approves. Registry services decide which appointments the agent can book on its own; residence changes, civil status procedures and special cases remain a proposal that the office confirms.

  3. In the afternoon and on weekends, when citizens are out and about in town and notice potholes, broken streetlights or dumped waste.

    Local issue reporting agent

    Collects text, photo and location, fills in missing information and opens the report for the relevant office.

    1. Receives the report from the website or WhatsApp and asks for the location if it's missing.
    2. Looks at the photo to work out the type of job: road maintenance, street lighting, green spaces, waste.
    3. Blurs faces and license plates in the photo before saving it to the report file.
    4. Checks whether an open report already exists at the same spot and links to it instead of duplicating it.
    5. Creates the ticket for the technical office or the service operator and gives the citizen a reference to follow it.

    Connects to: Municipal reporting system · Area map · WhatsApp Business Platform

    Gemini
    Reads photo, text and coordinates together and recognizes the type of job from the image: it's the model we choose when a report starts from a photo.
    GPT
    Fills in the report form in a fixed format, field by field, which the technical office's software can import without touch-ups.

    Who approves. Priority, crew and response times are decided by the technical office; the agent doesn't promise citizens repair dates.

  4. In the evening, after the offices close, when families have time to write calmly.

    Multilingual reply agent

    Answers in the citizen's language starting from the same official sources in Italian and flags to the office the people who need an appointment.

    1. Recognizes the language of the message and replies in the same language.
    2. Uses only approved service sheets, translated on the fly without adding information.
    3. Attaches the link to the official page in Italian and says so, so nobody mistakes the translation for an official act.
    4. When a desk visit is needed, forwards the request to the office with the language spoken and a summary in Italian.

    Connects to: WhatsApp Business Platform · Website chat · Service sheets

    GPT
    We use it for the languages most spoken by local communities, after comparing it in testing with Claude Sonnet on the same questions translated by the municipality's mediators.
    Gemini Flash
    Recognizes the language and routes the many short messages to the right service before the actual reply starts.

    Who approves. Translations of the most-used sheets are reviewed by cultural mediators or the offices before going live; texts with legal value stay in Italian only.

  5. Every Monday morning, before the meeting with service managers.

    URP report agent

    Reads the week's conversations in anonymized form and suggests which website pages to update and which answers are missing.

    1. Gathers the week's conversations, already stripped of names and contact details.
    2. Groups the questions left unanswered and those most often passed to a staff member.
    3. Points out the website pages that caused doubts or contradictory information.
    4. Prepares a short report for the URP with proposals for new service sheets to write.

    Connects to: Conversation log · Official website · URP mailbox

    Claude Opus
    Works once a week on many conversations at a time and must tell a recurring doubt from an isolated case: here reasoning matters, not speed.

    Who approves. The head of the URP decides which proposals become website changes; the agent publishes nothing.

Where time gets lost today.

These are the points we hear about most often. Each one is a possible first agent.

On the phone they always ask the same things

Registry office hours, documents for an ID card, local tax deadlines, the recycling collection calendar: the URP answers the same questions all morning, while people with a real problem are left waiting.

Informational questions get answered on the website and on WhatsApp, with a link to the official page each piece of information comes from; the phone line and the help desk get the cases that need a person.

The information is on the website, but nobody finds it

A citizen searches for “change of residence” and finds a regulation, outdated forms and the office's page. In the end they call, or show up at the desk without the right document.

The agent understands questions written in everyday words, answers in a few lines with the steps to take and leads the citizen to the right page and form on the official website.

Reports come in from everywhere

A pothole on the provincial road arrives by email, a broken streetlight by private message to a council member, dumped waste as a photo on social media. The technical office pieces them together by hand, often without a precise address.

A single channel collects text, photo and location; the agent asks for what's missing, classifies the type of job and creates the report for the relevant office, which decides priority and crew.

People who don't speak Italian well get left out

Recently arrived families look for information on residence, school enrollment and services, but the website is only in Italian and at the desk they often need a relative to translate.

The agent answers in the citizen's language, starting from the same official pages, and when an appointment is needed it flags it to the office along with the language spoken.

Which AI model we use, and for which task.

We don't tie you to a single provider: we work with Claude by Anthropic, GPT and Codex by OpenAI, and Gemini by Google, and we choose the version during testing, on your own cases.

Claude Sonnet

  • Website help desk agent. Works well with long instructions and refusal rules: it answers only with what's in the authority's sources and keeps an institutional tone even with an angry citizen.

Claude Haiku

  • Website help desk agent. Sorts incoming questions by service and immediately spots personal ones to pass to a person, so the more capable model works only on the answers.
  • Registry services appointment agent. Requests are short and numerous, like “can I move it to Thursday?”: you need a fast, low-cost model that understands dates and intentions without hesitation.

Codex

  • Registry services appointment agent. It doesn't see citizens' data: we use it to write and maintain the connector to the desk calendar, with code reviewed by our team.

Gemini

  • Local issue reporting agent. Reads photo, text and coordinates together and recognizes the type of job from the image: it's the model we choose when a report starts from a photo.

GPT

  • Local issue reporting agent. Fills in the report form in a fixed format, field by field, which the technical office's software can import without touch-ups.
  • Multilingual reply agent. We use it for the languages most spoken by local communities, after comparing it in testing with Claude Sonnet on the same questions translated by the municipality's mediators.

Gemini Flash

  • Multilingual reply agent. Recognizes the language and routes the many short messages to the right service before the actual reply starts.

Claude Opus

  • URP report agent. Works once a week on many conversations at a time and must tell a recurring doubt from an isolated case: here reasoning matters, not speed.

How we choose between models: the best AI model for agents.

Your municipality's help desk, the first agent

Send us the questions your desk gets most often and the link to the municipality's website: we'll tell you which can be answered automatically, from which sources, and what must stay with the offices.

Prefer the full form? Go to the contact page.

No spam or sales pitches. Response within 1 business day.

We use this data to contact you and, if we work together, to manage the project. We store it in Giallo's internal management system, on servers in the European Union, where only our team can see it; by email we only receive a notice, without your data. If you do not become a client, we delete it 3 months after our last contact. Details and your rights: privacy policy.

Do you prefer to write or call us?

How we work with you.

  1. A morning behind the desk

    We listen to calls and emails at the URP and registry services to see which questions recur, which need a person and where the website doesn't help.

  2. Sources before the chatbot

    Together with the offices we choose the pages and documents the agent can answer from; if a sheet is outdated or contradictory, we fix it before switching the channel on.

  3. Accessible and recognizable

    The chat is built into the official website following the accessibility rules that apply to public websites, and always states that it is an automated assistant.

  4. Testing with staff, then with citizens

    First URP staff use it with real questions from the previous week; only when the answers hold up do we open the channel to the public.

Frequently asked questions, straight answers.

No. A municipal chatbot explains how to get them, for example by downloading them from the national resident registry with a digital identity or by booking a desk appointment, but it doesn't issue official documents or access the registry records.

Yes, often more than for large ones. Where the URP is a single person, an agent that answers recurring questions frees up time for the people who come to the desk; you start with one channel and a few well-written service sheets.

Yes, if the municipality decides so. It answers in the citizen's language starting from the official pages in Italian and says so; texts with legal value stay in Italian and the most-used translations are reviewed by mediators.

You trace it straight back to the source. Every answer includes the link to the page it comes from: if the page was outdated, it gets corrected there, and the agent uses the updated version from the next conversation on.

Yes, through WhatsApp Business Platform with a dedicated municipal number. The channel is practical for reports with photos; informational messages sent by the municipality follow the rules for templates approved by Meta.

Very little, and almost all of it public. You need the website pages, the service sheets and the list of offices with their contacts; citizens' data comes in only for appointments and reports, with the fields the municipality decides.

Sources of answers: Can the municipal chatbot issue certificates?Can the chatbot also run on WhatsApp?

AI agents for municipalities and local authorities

Which citizen questions do you want to take off the URP's phone?

Send us the questions your desk gets most often and the link to the municipality's website: we'll tell you which can be answered automatically, from which sources, and what must stay with the offices.