Structured context
Tally reads supported LedgerHQ accounting and workflow data through typed tools, not a hidden copy of the books.
Inside Tally
How the Tally AI bookkeeper works.
Tally turns a bookkeeping request or operating signal into a scoped, tool-based workflow: understand the firm or company, gather current evidence, act within authority, and leave a result people can inspect.
Configurable duties
The firm chooses work modes, duties, limits, and advanced permissions that shape what Tally should do.
Durable supervision
Jobs, automations, actions, approvals, activity, and accounting records make operational work inspectable.
01
1. Resolve the scope before the work
A firm-wide question can look across the companies the user and Tally are allowed to access. A company-specific task resolves to that company before protected accounting context is read or changed.
This prevents a conversational shortcut from becoming a tenant or company-boundary shortcut.
- Firm-wide conversation for broad supervision
- Company-scoped context for book changes
- Membership and permission checks
- No secret credentials in model context
02
2. Gather current bookkeeping evidence
Tally can use supported tools to inspect bank-feed queues, chart accounts, journal records, statements, reconciliations, reports, requests, jobs, and other approved surfaces.
The model is expected to use current records rather than claim that a task is ready based on a generic assumption. If the evidence is missing, that absence becomes part of the result.
- Typed tool inputs
- Organization-scoped results
- Accounting records instead of pasted summaries
- Missing evidence surfaced explicitly
04
4. Prove the outcome
A useful Tally answer distinguishes explanation, prepared work, and completed work. Prepared work should point to the approval or next action. Completed work should point to the resulting company, record, job, action, or activity.
This evidence boundary is especially important when a task spans several model turns or continues in a background worker.
- Durable conversation turns
- Jobs and automation runs
- Actions and approvals
- Accounting and audit records
How the workflow moves
A visible sequence, not a black box.
The exact action depends on permissions, company context, and the evidence available. The workflow stays inspectable from intake through review.
- 1The firm asks a question, gives an instruction, or runs an intended automation.
- 2Tally resolves scope and reads the supported LedgerHQ context needed for the task.
- 3It explains, prepares, or performs work within the firm's configured authority and hard safeguards.
- 4The response links the outcome to durable approval, activity, or accounting evidence.
Human control
The firm can review Tally settings, Activity, approvals, and resulting accounting records, and can pause Tally when behavior or data quality needs investigation.
Product boundary
Tally is not an unrestricted database agent and does not create support for product modules that LedgerHQ itself does not support.
Questions
What firms usually ask.
Does Tally have direct database access?
Tally is designed to work through supported LedgerHQ tools and services with scope and policy checks, not an unrestricted conversational back door.
What happens when Tally needs a person?
The work can pause for an approval, decision, missing document, confirmation, warning resolution, or clarification. Activity is the supervision view for that work.
Can a firm pause Tally?
Yes. Authorized firm users can pause new Tally work and use the problem-report path when behavior or data quality needs investigation.
How do I know a task completed?
Look for the resulting LedgerHQ record, job, action, approval state, or activity evidence. A chat statement by itself is not proof of a book change.
Keep exploring