AI agents
The month closes itself
into drafts you review.
An agent is a flow on a schedule. The first one shipped is the month-close: it takes the work approved in the month, builds the avanzamenti, prepares the draft invoices grouped the way each customer is billed, and sends you a report of what it did. Everything it writes is a draft — issuing stays yours.
It prepares. It does not issue.
The month-close agent writes avanzamenti and draft invoices, and stops there: nothing is emitted, nothing reaches the SdI, no fiscal act happens without a person. The drafts land in the same lists you already work, ready to be checked and issued.
Drafts prepared by the agent, in the same table where drafts are issued by hand.
The report's figures are counted by the server.
After every run you get a report: how many avanzamenti, how many drafts, what was skipped and why. Those counters are written by the server from what actually happened — the model never authors a number in the report.
The run report: advanced, drafted, skipped — counted by the system.
It runs on your schedule, within your budget.
Each company sets the schedule, the execution policy — what the agent may create without asking, and above which threshold it must leave things pending — and a monthly token budget with a cap on writes per run. A run that would exceed either stops and says so.
The agent's configuration: schedule, policy, thresholds, budget.
Every run leaves a transcript.
The run history holds what each run did, the records it created, the tokens it spent, and the full transcript of its steps. An agent you can audit is an agent you can trust with the boring part of the month.
A run's transcript and the records it created, kept on the record.
Everything it does
- Automated monthly closeMonthly hours and milestones become draft advancements and per-customer invoices, expenses and backlog too.
- Scheduled or on-demand runsAutomatic monthly execution or on-demand for one or more months, never closing the same period twice.
- Amount and approval controlsAmounts are always computed by the system; above a per-customer threshold the invoice waits for approval.
- Reversible drafts, never a direct sendThe agent only creates drafts within caps on records written and reasoning; issuing stays a human act.
- Execution transparencyLive flow, full reasoning transcript, what was created, and a final report to chosen recipients.
- AI cost and token monitoringCost and tokens per run, monthly consumption by feature, model and user, and a token budget per agent.
Questions we actually get asked.
- Can it send invoices to the SdI?
- No, and it is built so it cannot: the agent's writes stop at drafts. Transmission remains a human act, in the invoicing module, with the same checks as always.
- What if it runs out of budget mid-run?
- It stops and reports budget exhausted, with the counters of what it had done up to there. Nothing keeps spending silently, and the next run starts within whatever budget remains.
- Is there more than one agent?
- Today the framework and the month-close agent ship, and that is what this page sells. Further agents will run on the same rails — schedule, policy, budget, report — as they arrive.
Works with
- AIAn assistant that answers from your data, and recurring tasks that run on the schedule you set.
- Reports and exportsAny table becomes an XLSX with the filters it was showing, from templates you keep for the exports you rerun.
- Public APIPer-company API keys with scoped access, a REST API over your records, and signed webhooks when they change.
Tell us how you work today.
There is nothing to prepare. Half an hour on how your company runs — who tracks the hours, how an invoice comes together, when you find out whether a commessa made money — and we tell you which modules would fix it and what a quote looks like.