Find out which process to hand to an agent first, what it costs in real money and whether it is safe to switch on.

This is a free kit for the owner of a small or mid-sized business to move past casual ChatGPT use and put an AI agent to work on a real process. There are three tools that run right here on the page, no sign-up: the Process X-ray shows where to start, the calculator shows what it costs and what it pays back, and the checklist tells you whether the agent is safe to switch on. Then the track leads to INEMA courses and kits to build each agent. All you need is a browser; nothing leaves your computer.
In Berlin, in October 2026, Jensen Huang (Nvidia) said every company will become an artificial intelligence company and that the way forward is to produce AI, not just consume it. He explained it with a five-layer cake. Here is the same idea translated for the owner of a small business.
In practice: an agent that answers opening hours and prices on the clinic's WhatsApp on its own, and hands what it doesn't know to the front desk. That is layer 5 running on top of the other four.
It is the number one barrier for small businesses, ahead of cost (31%). The X-ray tackles exactly that.
Serasa Experian, survey of SMEs, 2026Of the companies that use AI, 80% buy ready-made software and only 26% build their own solution. That is Jensen's "consume".
Cetic.br, TIC Empresas 2025Forecast through 2027, due to cost, unclear value and uncontrolled risk. The calculator and the checklist exist so you don't end up in that number.
Gartner, forecast of June 2025One process at a time. You pick, calculate, check safety, run a 30-day pilot and measure. If it works, move on to the next in line.
The tools run on this page. Your numbers are stored only in your browser; nothing is sent to anyone.
The tasks that repeat every week in the business. If you don't know them by heart, ask the team to write them down for three days.
How many times a week each task happens and how many minutes it takes. An estimate is enough to get started.
On GitHub, the kit/ folder has the X-ray and the calculator as CSV to open in Excel or Google Sheets, and the checklist as text (in Portuguese).
The page comes pre-filled with an example from a small clinic. Replace it with your own business data.
List the tasks that repeat. For each one, say how many times a week it happens, how many minutes it takes, whether the rule is clear, whether the information is already on the computer and how much damage a mistake would do. The X-ray returns the queue of your first three processes and the INEMA kit for each one.
How the score works: hours per month × rule clarity × data availability ÷ risk. It is a ruler for ranking, not the truth. A high-risk task can go in the queue, but only with a person approving every answer.
Take the first one in the queue. Compare today's cost (people doing it) with the agent's cost (the AI bill plus someone's time reviewing). The metric that matters is cost per task, counting the review and the tasks that go back to a person.
The pre-filled values are an example, not any vendor's price. Measure your baseline before the pilot (how many tasks, how much time) and redo the math with the real numbers after 30 days.
The agent works on its own and touches files, accounts and messages. The rule from the keynote is least privilege: it only accesses what the task needs, and someone keeps an eye on it. Tick what is already guaranteed.
Items 2, 4 and 10 are required: without any of them the result stays at supervised pilot.
Based on the principle of least privilege and on the risks in the OWASP Top 10 for Agentic Applications (2026), translated for non-technical readers. The text version is in kit/checklist-agente-seguro.md (in Portuguese).
Build the agent for the first process with the kit the X-ray recommends. Over the 30 days, write down every week: how many tasks it did, how many went back to a person, how much the AI bill was and how many complaints there were. At the end, redo the calculator with those numbers. If the difference is positive and the checklist is still green, move on to the second in line.
All open and free. The X-ray already points to one of these for each task in your queue.
A WhatsApp assistant that never makes up an answer (in Portuguese).
Service by industryClinic, gym, physiotherapy and nutrition, ready to adapt (in Portuguese).
Scheduling and bookingsRestaurant and hotel, with bookings handled by the agent (in Portuguese).
DiagnosisHow to map a company's processes before automating (in Portuguese).
ImplementationAI implementer for SMEs: processes, ROI and delivery (in Portuguese).
People and processesHow people, processes and agents reorganize (in Portuguese).
SafetyUse AI agents without getting burned (in Portuguese).
Business contextWhat the agent needs to know about your business, in one place (in Portuguese).
ToolsGet AI to use the tools you already have, with rules (in Portuguese).
ManagementHow to manage processes with agents day to day (in Portuguese).
AI factoryAI on your own machine: privacy and 24-hour agents (in Portuguese).
First stepsFrom ChatGPT to your first agent, in four steps (in Portuguese).
This is the first release. The next pieces complete the cake.