Interactive lab

Explore a workflow execution

Choose a scenario — Support, Documents or Operations — and follow what happens at each step of the workflow. An interactive demonstration using simulated data: nothing here is connected to real systems.

Interactive demonstration · simulated data

Support

step 1 · Intake
A request arrives
An employee of Empresa Exemplo S.A. (fictitious) opens a ticket asking for guidance about the shift change.
step 2 · Knowledge
The agent searches the right sources
The question is compared with the company's knowledge base — policies, procedures and announcements — and the relevant sources are retrieved with references.
step 3 · Agent
The answer is built with citations
The agent drafts the answer from the retrieved sources, cites every excerpt used and classifies the ticket by topic and urgency.
step 4 · Approval
A human approves before sending
Because the customer defined human approval for replies external to the team, the agent prepares everything and WAITS. Nothing goes out without the approver's sign-off.
step 5 · Result
Answer sent and recorded
With the approval, the answer goes to the employee and the whole flow is recorded — who asked, what was consulted, who approved.

Documents

step 1 · Intake
A batch of documents arrives
The finance team of Empresa Exemplo S.A. (fictitious) receives contracts and invoices in PDF that must become organized data.
step 2 · Knowledge
The agent understands each document
Each file is read and compared with the models the company uses — the agent knows what to look for in a contract and what to look for in an invoice.
step 3 · Agent
Extraction and organization
The agent extracts the fields, cross-checks them with the control spreadsheet and flags discrepancies — such as a term that does not match the amendment.
step 4 · Approval
The discrepancy waits for a human decision
The 11 consistent documents move forward; the discrepancy becomes a clear pending item for the manager to decide. The agent does not 'resolve' ambiguity on its own.
step 5 · Result
Organized base, exception decided
With the decision recorded, the base is complete and auditable: every data point points back to the document and page it came from.

Operations

step 1 · Intake
An indicator leaves the range
On the operations panel (simulated), the queue time of a stage passes the agreed limit — the kind of signal that usually gets lost mid-shift.
step 2 · Knowledge
Context before opinion
The agent gathers the context: recent stage history, occurrences recorded in the shift and the operating procedure that covers this deviation.
step 3 · Agent
Diagnosis and proposal
Cross-checking the data, the agent builds a summary of the likely bottleneck and proposes the action foreseen in the procedure — with the numbers that justify it.
step 4 · Approval
The coordinator decides
Changing the operation's allocation is a human decision: the agent delivers the ready diagnosis and waits for the coordinator to approve, adjust or reject.
step 5 · Result
Action recorded, cycle closed
Once the action is approved, the agent records the outcome and follows the indicator — if the queue does not return to range, the cycle reopens with the new context.

Support · step 1 of 5

A request arrives

An employee of Empresa Exemplo S.A. (fictitious) opens a ticket asking for guidance about the shift change.

channel: internal ticket #4821
from: employee (simulated)
subject: Question about the shift-change procedure

Want this workflow running in your process?

The lab shows the mechanics; designing the real agent starts with your context, your systems and your approval limits.

Design my agent