An AI bookkeeper that works inside the accounting system.

Tally works from the firm and company context LedgerHQ already maintains. That can include bank-feed work, statement coverage, reconciliation readiness, requests, report availability, journal activity, and supported close checks.

The goal is not a mysterious score. A useful signal should lead back to a company, a record, a job, an approval, or another reviewable source inside LedgerHQ.

Tally can inspect context, research supported records, prepare coding and other bookkeeping actions, explain exceptions, and complete work when the firm's configured authority allows it. The exact boundary depends on the duty, the evidence, the user's permissions, and whether the action requires confirmation.

If Tally says it prepared something, the team should still be able to see what remains to approve. If it says the work is complete, it should be able to point to the resulting record or activity trail.

A generic assistant can describe bookkeeping. Tally is connected to the supported LedgerHQ accounting surface, so a request can be resolved against the actual company, chart, bank activity, reconciliation, and report context the firm has authorized.

That connection also creates constraints. Tally does not get a separate path around tenant scope, role checks, period locks, posting validation, or confirmation requirements.

Firms can choose the duties Tally is expected to help with and the authority available for those duties. Ambiguous, externally visible, destructive, sensitive, billing-related, or unusually broad work can stop for explicit intent or confirmation.

Activity and accounting records provide the evidence layer. The operating standard is simple: the model's prose is not proof that the books changed; the resulting LedgerHQ record is.