Put a city's public services in one WhatsApp chat
Boti, the City of Buenos Aires' assistant on WhatsApp, books appointments, checks fines and answers residents' questions, and the city has piloted an AI agent to guide residents through more than 1,300 city procedures in local Spanish.
Built withChatGPT / OpenAI models
Try it withClaudeChatGPT / OpenAI modelsGeminiGrok
Deployed; Boti has run on WhatsApp since February 2019, with GPT-4o features since 2024; the procedures agent was a pilot when AWS described it in August 2025, Government of the City of Buenos Aires, with Microsoft (GPT-4o upgrade) and AWS (procedures agent pilot)
Why it matters
Boti puts city services inside WhatsApp, so residents can start a procedure in a chat instead of finding the right office or website. The city says Boti answers questions from more than 6 million people, and AWS, which worked with the city on the newest version, reports more than 3 million conversations a month. During the first months of the pandemic it triaged 120,000 queries about symptoms, and today it is also the way many people book a new identity card, a driving licence or a health appointment, check fines, or report a badly parked car with a photo. It shows a city treating a chat assistant as a front door to public services, with a person behind it when needed.
How it works
Residents save the city's number in WhatsApp and send a message. Boti has run since February 2019. In 2024 the city added OpenAI's GPT-4o, working with Microsoft, so Boti can hold more natural conversations and build tourism plans from the city's official data. In 2025 the city and AWS's Generative AI Innovation Center built a pilot agent for government procedures: a guardrail model first screens each message for abuse or prompt injection, then an agent built with LangGraph on Amazon Bedrock finds the right procedure among more than 1,300 and writes the answer in Rioplatense Spanish. Some services need identity checks through the city's miBA account, and Boti hands the conversation to a human operator or the 147 call centre when the user asks or after three failed attempts to understand a message.
Who it's for
City councils, government service centres and public agencies that answer large volumes of routine questions and want to meet residents on the messaging apps they already use.
The prompt
You are helping me design a WhatsApp assistant that guides residents through our public services and hands them to a person when it should. Do not ask me for the names, phone numbers or identity numbers of any real resident. Our organisation: [for example, a city council or a government service centre] The services to cover first: [for example, ten of the most requested procedures, with links to their official pages] Languages and tone: [for example, the local language as people actually speak it, plus English] 1. For each service, write what the assistant needs to know to guide someone: who is eligible, documents needed, fees, where and how to apply, and the official link. 2. Write the rules the assistant must follow: answer only from our official pages, say which page it used, say plainly when it does not know, and never promise an outcome or a date. 3. List the messages it must screen out or answer with care (abuse, attempts to make it ignore its rules, emergencies, distress), and write the reply for each. 4. Define when it hands over to a person (the resident asks, it fails to understand three times, or the topic is sensitive) and write the exact handover message, including what the resident sees while waiting. 5. Draft five sample conversations in our local tone, each under 100 words, ending with the next step the resident should take. 6. Draft a plain notice explaining that the assistant uses AI, what it stores, and how to reach a person. Do not invent procedures, fees, phone numbers or links; leave placeholders. Mark every rule that our legal or data-protection officer must approve before use.
Open inClaude (opens in a new tab)ChatGPT (opens in a new tab)Gemini (opens in a new tab)Grok (opens in a new tab)
Sources
The case study is the city's own and the AWS post was written by the partner and city staff who built the pilot. La Nación reported the GPT-4o upgrade independently but gave no usage figures. No independent evaluation of outcomes was found.
1. Government of the City of Buenos Aires: "más de 6 millones de personas"; "120.000 consultas"; "solo funciona con humanos detrás". Caso Boti (case study) (opens in a new tab)
2. AWS Machine Learning Blog: "facilitates more than 3 million conversations each month"; "includes over 1,300 government procedures". Meet Boti: the AI assistant transforming how the citizens of Buenos Aires access government information with Amazon Bedrock (opens in a new tab)
3. La Nación: "Boti integró GPT-4o". Boti, el chatbot porteño, cambia: integra a GPT-4o y hace recomendaciones turísticas (opens in a new tab)