Product
Autonomy you can inspect
mozr's agents are useful because they are constrained. Five surfaces make every action visible before, during and after it happens.
1 · Projection preview
Before anything posts, the exact journal entry is shown — debits, credits, period, tax — computed by the server, never by the screen. What you see is what lands.
2 · Approval queue
Your policy names the amounts, accounts and counterparties that need a human. Everything else the agents commit themselves. Approvals are one keyboard away, with the rationale.
3 · Replayable runs
Every agent run is a record: inputs, tools called, proposals, outcomes. Open any past run, see what it saw, reverse what it did.
4 · Kill switch
One control pauses every agent, every recurring automation and every auto-post — and shows what is being held. Turning it back on never posts a backlog behind your back.
5 · Nightly verifier
Balances, clearing accounts, bank feeds and reconciliations are re-proven every night. A failure demotes the agents to draft-only until a human looks.
And the boring parts
Invoices, bills, estimates and contracts with retainage, bank statements and PDFs, recurring templates, payroll import, cash flow, 1099s, QuickBooks import and full export.
What agents can never do
- Delete a document, or edit one that has posted — at any amount.
- Reopen a closed period.
- Execute a refund, commit an import, or release retainage.
- Touch a recurring template or a payroll mapping — standing authorisations are human-only.
- Change your billing, your users or your policy.