Platform · platform & data

One search box that reaches everything you may see

Finding things is an underrated share of finance work — which invoice, which contract, which approval, which version of the schedule. In most systems that means knowing which module to look in first, which is a piece of institutional knowledge nobody should need.

Rows
DeliverySalesG&A
New York$412K$286K$104K
Austin$238K$141K$62K
Remote$176K$88K$39K
Every dimension is on the journal line, so any combination is a query rather than a rebuild.
⌘K from anywhereScoped to your permissionsNatural language when useful

What it does

Six things, specifically.

Everything, one box

Transactions, invoices, bills, journal entries, contracts, projects, vendors, customers, employees, documents, and audit events — without choosing a module first.

Inside documents

Full text within attached PDFs and images, so finding the contract that mentions a specific termination clause does not require having tagged it.

Scoped by construction

Results are computed within your permissions in the data layer rather than filtered afterwards, so an out-of-scope record does not appear as a hint.

Natural language where it helps

AWS invoices over $5,000 in the last six months. Unbilled projects for one client. A phrase where a filter chain would be slower to construct than to describe.

Jump to the trail

From any record straight to its audit history, including the agent action that produced it, without navigating there separately.

Any result becomes a view

Save it, schedule it, or export it. A search you run monthly should become a report rather than a habit.

Knowing where to look is not knowledge worth having

In most ERP systems, finding something requires knowing that vendor bills live here, that attachments are searched separately, that contracts are in a different module, and that the audit history is somewhere else again. That knowledge takes months to acquire and is worth nothing outside the specific product.

It is also the largest hidden cost of onboarding a new person into a finance team, and it is the reason so much institutional memory sits with whoever has been there longest.

Knowing which module to look in is a skill that takes months to learn and is worth nothing anywhere else. It should not be a skill.

Permissions applied at query time, not to results

This distinction matters more in search than anywhere else. A system that searches broadly and then filters what you may see leaks information through absence and through result counts — you learn that something exists even when you cannot open it.

Scope is applied when the query is composed, in the data layer, so an out-of-scope record is not merely hidden but genuinely not part of the result set. For customer portals and external auditors this is the whole requirement rather than a refinement.

Search that becomes a report

The useful pattern we see is people using search to discover the report they actually wanted. Someone searches for unbilled work on a client three months in a row, and on the third occasion it should become a saved view with a schedule.

Every result set can be pinned, scheduled, or exported for that reason. A recurring search is a report nobody has built yet.

Limits

Where this does not help.

Not a data warehouse

Search finds records. Large analytical queries across years of history belong in your warehouse, which we export into rather than competing with.

It cannot find what was never captured

Searching for a dimension that was never coded returns nothing, and it will say so rather than returning a partial set that looks complete.

Handwritten attachments are unreliable

Document text search depends on extraction. Printed and typed text is solid; a handwritten note photographed at an angle is not.

Questions

What people ask.

Does it search inside attachments?
Yes, full text within PDFs and images, which is how you find the contract mentioning a specific clause without having tagged it.
Can users see things they should not?
No. Scope is applied when the query is composed rather than to the results, so an out-of-scope record is not part of the result set at all — not hidden from it.
Is natural language required?
No. Structured filters work and are faster for many things. Natural language is there for the cases where describing what you want is quicker than building a filter chain.
Can external users search?
Portal users search within their own scope only — their invoices, their documents, their projects. Auditors search within the periods and entities they have been granted.
Does it replace reporting?
No, and the useful pattern is the opposite: people use search to discover the report they wanted, then pin it. A recurring search is a report nobody has built yet.

Stop learning where things live.

A read-only connection and about a week is enough to search your own records from one box.