An agent is an assistant you assemble: it queries your data, reads your documents and answers in Portuguese, always stating where each number came from.
What separates an agent from a regular chat:
- It doesn't hallucinate. When the data doesn't provide an answer, it says so.
- It doesn't act on its own. If it needs to do something — reprocess a pipeline, generate a report — it proposes the action and waits for you to authorize with a click.
- It shows the path. Every answer comes with the steps it took and how much it cost.
The screen, label by label#
Where it is: menu IA → Agentes.
The top-right button is Criar agente — with no agents yet, the screen shows Criar meu primeiro agente and Ver exemplos. Each card in the list has three actions, and a duplicate icon:
| Button | What it opens |
|---|---|
| Conversar | ask the agent |
| Configurar | go back to the draft, adjust and republish |
| Atividade | each run, with steps, cost and what it proposed |
Inside the agent, everything lives on a single screen, in tabs: Desenho (where it is assembled, connecting boxes), Testar, Atividade, Agenda and Memória. The result can be sent by email or WhatsApp to the people you register — see Criar um agente.
The card badge shows the state: Ativo, Pausado, Rascunho or Com erro, followed by the published version. When a recent run failed, the card warns: "Uma execução recente falhou — veja a atividade."
Next to Agentes, the menu has Como funciona — first the three ways to use AI (ask an agent, let it run on its own and build a sequence, with the shortcut Ir para Fluxos) and, inside, the diagram of an execution path, from the Gold table to the Porta dos modelos de IA. It's worth a visit before creating the first one: that's where it becomes clear that the Agent Engine is ours and the model provider (OpenAI, Anthropic or Google, with your company's key) is a swappable piece.
Where to start#
Before the first agent: the AI key
Agents run with your chave de IA da sua empresa — OpenAI, Anthropic or Google. The workspace owner registers it in Configurações → IA: paste the key, click Salvar e testar and check the Funcionando badge. The model cost is charged to your company's account at the provider; ingestia.io credits pay only for data usage. Without a key, the agents screen shows the notice Cadastrar chave de IA →.
When an agent is not enough#
- Flows: sequential tasks, box by box
- Teams: a coordinator and specialists
- What the agent remembers
- Connect external systems
Operate#
- Where the intelligence comes from
- Monitor: cost, failures and what's stalled
- Who can do what
- Take the agent elsewhere
What it can do#
| Capability | What it means |
|---|---|
| Query data | Reads the tables you grant and answers with numbers and their origin |
| See the catalog | Discovers on its own which tables and columns exist |
| Query documents | Searches the documents you uploaded, citing the excerpt |
| View dashboards and runs | Tracks what's happening in the operation |
| Propose actions | Reprocess a pipeline, generate a report — always asking for confirmation |
| Consult specialists | A coordinator agent asks other agents |
| External tools | External systems, via an open standard |
How much it costs#
Each question consumes AI and is charged to your company's monthly account, on the line "Agente do cliente". Each agent's card shows how many runs it made today and how much it cost. You set a spending cap per run in the agent's configuration.