For audit firms and in-house accounting teams. The agents load the whole general ledger, run the controls and the journal-entry tests on every line, reconcile it with the bank, match invoices, and assemble the workpapers. Every result can be run again and gives the same answer.
Four agents, coordinated by one that reports to the partner or the head of accounting.
Tracks every engagement and deadline, reviews them every morning and sends the weekly client status.
Loads the ledger and runs the controls, the journal-entry tests and the cut-off tests, then keeps them running on a schedule.
Prepares tax workpapers and the deadline calendar from cited official sources. It does not advise or elect.
Requests documents from clients or departments, chases what is missing and keeps the calendar.
Each test records its method, its result and who signed it off, so the workpaper can be produced again from the same data.
The controls, the anomaly detection and the reconciliations are rules and seeded models. The same ledger always flags the same lines, each with its source reference.
A test that ran once can run every day, week or month. A new finding raises an alert to the team, not a surprise at year end.
An accounting or tax rule is used only with the official text it comes from, and the date it was checked. A rule without a source is marked as unverified.
For a one-off audit, the Audit Workbench project takes the same tests from planning to the signed report.
FEC, SAF-T (XML), AICPA audit data standard, spreadsheets
NetSuite, SAP (read)
Qonto (read-only), or statements uploaded as files
Google Workspace, Microsoft 365, email
Each agent gets only the access its role needs, granted by you and listed in the access register. Another system can be connected through its API or an MCP server.
Each agent has its own SPIFFE identity and signing key. Every action records which agent acted, under which rule, and for whom.
Approvals where you want them. Autonomy widens only on measured results and steps back by itself when results slip.
Agents reach only what you grant. One switch stops an agent or the whole team, and the stop is on the record. 32 governance packs apply your rules, on your own model keys.
Which engagements or ledgers the team takes on, and your methods and rules, reviewed by you.
A charter per agent with its limits, and your tools connected with least access.
Real work, every action approved by a person, until the team earns autonomy.
At the autonomy it earned, measured against where you started.
Three to six weeks from scope to go-live. How a team launch works · Finance team · Other teams