Skip to content
ERXIt doesn’t just plan, it executes

The AI-employee ERP

Hire your AI workforce.

AI inside the ledger, not layered on top of it. Eleven AI employees run hundreds of checks across your books, your people and your stock, and agentic workflows act on what they find. You review, they work.

  • CFO
  • HR
  • Tax Adviser
  • Compliance
  • Bookkeeper
  • Payables
  • Collections
  • Sales
  • Operations Director
  • Shift Lead
  • Management

For businesses anywhere, launching first in Spain — where VERI*FACTU, 15 AEAT modelos and Ley 2/2023 are in the core rather than in a plugin.

New release · September 2026

VIDA26

the joy of living

Less time fighting paperwork. More time to live.

11
AI employees
Hundreds
of scheduled checks
162
typed tools the AI can call
1,282
catalogued procedures
Granular
security, set per role, company and record

Inside the system of record

AI inside the ledger, not a layer on top.

Plenty of ERPs are adding AI. What is worth asking each of them is what the AI can reach, what it is allowed to do, and what stops it. Here, each question has an answer you can check.

It works from the posted books

The crew’s findings and the assistant’s answers come from real entries and the documents behind them — not from a summary written overnight.

Meet the crew →

You extend it by describing it

Describe a workflow or a CRM automation in a sentence and read back real steps over your own data. Nothing runs until a person publishes it.

Agentic workflows →

It rehearses before it counts

Anything irreversible the assistant is asked to do — post, void, file — comes back as a preview with a one-time confirmation before it commits.

L.U.C.A. →

It follows your staff’s rules

Every AI employee, agent and flow runs with a real identity, so row-level security decides what it can see and do — in the database, not in a prompt.

How access is decided →

AI Crew

Eleven specialists. Hundreds of checks. One feed.

Each employee watches one domain under a real, non-administrative role, and files what it finds into a single per-company feed. Nothing is a chatbot; every card names a document, a figure and a decision.

11 on duty · hundreds of checks

CFO

Working capital, runway, leverage and the covenants nobody re-reads.

HR

Contracts, absences, working time and the signatures about to expire.

Tax Adviser

AEAT deadlines, VERI*FACTU submissions and every modelo before it is filed.

Compliance

Evidence and process — signatures, lifecycle steps, workflow bypasses.

Bookkeeper

Whether the ledger can be trusted: unposted entries, unmatched bank lines.

Payables

Bills falling due, duplicate payments and supplier concentration.

Collections

Overdue money, named customers, and a decision for each one.

Sales

Untriaged leads, stalled deals and conversations nobody answered.

Operations Director

What the company owns and moves — assets, stock, slow movers.

Shift Lead

The routines: was the checklist walked, signed off, and what keeps failing.

Management

The feed itself — whether anyone is acting on what the other ten say.

The crew feed: a critical finding, its evidence, and the document behind it.Watch on YouTube ↗

They do not all run nightly

Some checks run every four hours — the VERI*FACTU ones among them, because a rejected submission you learn about tomorrow is one that sat live all day.

How the crew works →

Guardrails

They read and propose. They never pay, sign or send.

Every AI employee runs as a real system role with one visibility permission — never as an administrator. What it may see is what a human in that role may see, enforced by the same row-level policies, in the database, not in the interface.

Watched by

Compliance

A real role, never admin

Each employee is bound to a shipped non-admin role. Widening one to make a card work would widen it for people too.

Read and propose only

No AI employee pays an invoice, signs a document or sends a message on your behalf. Actions need a person.

Refusals are logged

When the permission layer refuses a read, the attempt is written to the audit trail rather than silently dropped.

L.U.C.A. · AI assistant

Ask for it in a sentence. L.U.C.A. does the work.

The crew only reads and proposes. L.U.C.A. is the one you talk to, and it acts: it answers from your posted books and creates the invoice, the loan or the asset you describe — as you, with your permissions, and with a confirmation before anything that cannot be undone.

An invoice, a loan, an asset and a report, each asked for in a sentence, and the confirmation it asks for before voiding anything.Watch on YouTube ↗
  • Answers from the real ledger

    Customers, suppliers, balances and reports, read from posted entries — and only what your own permissions allow.

  • Does it, not just drafts it

    Invoices, bills, payments, loans and fixed assets, described in a sentence or read from the document you attach.

  • Asks before anything irreversible

    Post, void, file or delete comes back as a preview you confirm. Without your OK, nothing changes.

  • Shows its work

    It streams as it goes, and every tool it called is listed for when you want to check.

Agentic Workflows

Agents that read the paperwork and act on it.

The work that resists automation is the work that arrives as a document — a supplier PDF, a photographed delivery note, a ticket in a shared folder. A flow puts a language model inside the process to read it, decide what it means, and hand structured fields to the next step. Every write stays a step you can see, and nothing runs beyond the permissions the flow itself has.

Building a flow, wiring a node, and watching a dry run light up green.Watch on YouTube ↗
  • A model inside the process

    One step in the flow is a language model with a job: a prompt, the exact set of tools you allow it, and a result whose fields you declare — severity, reason, amount, whatever the next step needs.

  • Paperwork read for you

    A supplier invoice as a PDF.

  • Documents where they already live

    Suppliers do not send things the way your process wants them.

  • Describe it, and it drafts it

    Say what you want to happen in your own words and get a working draft back — the steps wired, the conditions filled in, pointed at your actual data.

How the agents work →

How it works

Connect, watch, act.

1

Connect

Banks, payment rails, mailboxes and messaging channels feed one reconciled ledger.

2

The crew watches

Eleven employees run their checks on their own cadence and file what they find.

3

Workflows act

A finding triggers a flow: route it, approve it, email it, or draft the correcting document.

Spanish compliance

VERI*FACTU, submitted. 15 modelos, generated.

A full VERI*FACTU pipeline with hash-chained records, QR and cancellation. 15 AEAT modelos derived from the ledger, 13 of them exported in the BOE’s own fichero layout. Facturae 3.2.2, qualified signatures to PAdES B-LTA, and Ley 2/2023 with its deadlines built in.

Watched by

Tax Adviser

303390347349369420200202111115123180190193232

What aicroo generates is a draft, meant to be checked with your gestoría or tax adviser before filing. Review, submission and ultimate responsibility for what is declared remain the user’s.

See what is covered →

Payments and portals

Take the payment. Reconcile the payout.

Cards through Redsys, Revolut or Stripe, PayPal for the customers who would rather pay from their own account, and direct debit that charges without the customer present. Your customer pays from a portal that also holds their invoices and the documents waiting for their signature — and your suppliers get the mirror image, where they submit invoices instead of emailing them into a shared mailbox.

Watched by

Collections

  • RedsysRedsys
  • RevolutRevolut
  • StripeStripe
  • PayPalPayPal

When the processor pays one lump for forty charges, three refunds and its own fees, each line finds its invoice and the net drains against the bank — to zero, or it is not reconciled.

How getting paid works →
An invoice sent to Verifactu and paid from the customer portal, and a supplier invoice uploaded from theirs and accepted.Watch on YouTube ↗

Electronic signature

Employees, customers and suppliers sign in one place.

Contracts, quotes, purchase orders and NDAs go out for signature from the record they belong to, come back sealed to PAdES with an independent timestamp, and keep an evidence trail that is chained entry by entry. Advanced signatures are sealed in-house; qualified ones run through AutoFirma with the signer’s own certificate.

Watched by

HR

An employment contract signed by the employee and their manager, and a client contract signed from a link, both kept in the vault.Watch on YouTube ↗
  • Signed inside onboarding, the portal or the deal — never as a separate errand
  • Identity checked at the moment of signing, and recorded with it
  • Long-term validity by default, so it still verifies years later
  • The AI Crew watches what is expiring, stuck or unverifiable
SESAESQES
How signing works →

Integrations

It plugs into what you already run.

Your bank, the rails you take money through, the channels your customers write from and the tools your team already has open — connected to the ledger itself, not to a plugin somebody has to keep alive.

Banks

Pontoopen-banking.ioRevolutRevolut BusinessCAMT.053Norma 43

Payments

StripeStripeStripeStripe ConnectRedsysRedsysRevolutRevolutPayPalPayPalAdyen payouts

Messaging

WhatsAppWhatsAppInstagramInstagramGmailGmailOutlookIMAP

Workplace

SlackMicrosoft TeamsSharePointOneDriveGoogle DriveGoogle Drive

Spreadsheets

Google SheetsGoogle SheetsExcel Online

Then the half nobody demos: making it all agree.

A connection is only worth having if it closes the loop. What arrives from the bank and from the processor is matched against what the books already expect — and what does not match is a short list, not a silence.

The bank

Connect the account and the statements arrive on their own; or import CAMT.053, Norma 43 or a CSV. Either way the lines are matched against what the books already expect, and what is left over is the short list worth a person’s attention.

The payouts

The lump a processor pays you on Tuesday is not one payment — it is forty of them, minus three refunds and its own fees. Each line finds the invoice it belongs to, the fees are posted, and the batch is locked: if it was wrong it is voided and reversed as one act, not by hand-editing forty settled invoices.

An imported statement: each line arrives with the invoice it pays and how sure the match is. What has no document stays in view.

Developers

162 typed tools. 1282 procedures. One API key.

Every procedure in the catalogue declares whether it mutates, whether it is reversible, and which permission it needs — so an agent can be given exactly as much authority as you intend and no more.

An AI agent asks who owes you and answers from the real ledger, seeing only what its MCP client was granted.Watch on YouTube ↗

Security

Enforced in the database, not in the interface.

Anyone can call the API directly, so hiding a field in a screen is not security. Sensitive data is gated by row-level policies and permission checks at every layer that can return it.

Roles you can audit

87 areas at four levels — a grid somebody can read, not hundreds of loose switches.

Hundreds of row-level policies

Tenancy and authority are enforced by the database, on every read and write.

Passkeys and required MFA

WebAuthn sign-in, and a second factor an organisation can require of everyone.

Full audit trail

Who saw what, who changed what, and what was refused.

Pricing

Pay for the modules you switch on.

Per-user where seats are the cost, flat where they are not. No mandatory platform fee, no annual lock-in to see a price.

Closer AI Factura

€9/month

Humanus

€7/user/month

min €25/month

AI Crew

€49/month

Ready to hire?

Book a demo and see the crew run against your own numbers.