AI CoreLive

Document Intelligence

Drop any document. We figure out what it is, what's in it, what it links to, and what records it creates.

Beyond receipts: timesheets, purchase orders, quotations, expense reports, delivery notes, credit notes. Gemini Vision extracts structured data (line items, hours, rates, PO numbers, totals). The document then auto-links to the matching transaction, invoice, bill, estimate, contact, or project. From there, one tool call turns a timesheet into an invoice, a PO into a bill, or a quotation into an estimate.

In short

AccountsOS Document Intelligence is the Phase 2 evolution of receipt scanning. It detects 6 document subtypes (timesheet, purchase order, quotation, expense report, delivery note, credit note) and extracts structured data β€” line items, hours, rates, PO numbers. Documents are auto-linked to the records they relate to via the document_links junction table. Three Finn tools create invoices, bills and estimates directly from documents.

6
Document subtypes detected
3
Create-from-doc tools
Auto
Linked on upload
Try Document Intelligence Free

Free during Early Access β€” no credit card required

Everything Document Intelligence can do

Subtype detection

We don't just read text β€” we know what kind of document it is.

  • Timesheet β€” employee/contractor hours by day, rate per hour, project allocation
  • Purchase order β€” supplier, line items, quantities, prices, PO number
  • Quotation β€” customer, line items, validity, terms
  • Expense report β€” claimant, line items, dates, totals, currency
  • Delivery note β€” items shipped, dates, references to a PO
  • Credit note β€” reason, original invoice reference, amount credited
  • Plus the original 'receipt' subtype for everything else

Structured data extraction

What Gemini Vision pulls out, ready to use elsewhere.

  • Line items array β€” description, quantity, rate, total
  • Hours / rates for timesheets
  • PO numbers, quotation references, invoice references for credit notes
  • Vendor, customer, date, currency, totals, VAT
  • Confidence score per field
  • Original image preserved alongside extracted data

Auto-linking on upload

Every document gets linked the moment it lands.

  • Match by reference number β†’ invoice / bill
  • Match by vendor + amount β†’ bill / estimate
  • Match by contact name β†’ contact / project
  • Stored in the document_links junction table β€” many-to-many, typed relationships
  • Re-runs on every ingest path: upload, email, WhatsApp, Slack, Telegram, Gmail

Create-from-document

Three Finn tools turn documents into records.

  • create_invoice_from_document β€” timesheet entries β†’ invoice line items, hours Γ— rate
  • create_bill_from_document β€” PO β†’ bill, line items preserved
  • create_estimate_from_document β€” quotation β†’ estimate, ready to send
  • All require Finn confirmation before they write
  • Source document remains linked to the created record for audit

Document detail & editing

Inspect everything Gemini found. Fix what's wrong inline.

  • Document detail page shows all extracted fields
  • Editable line items β€” correct hours, rates, totals per row
  • Linked entities sidebar (which invoice, which contact, which project)
  • Reprocess if extraction needs another pass

Unlinked transaction finder

See what still needs a document β€” and fix it from one place.

  • Lists every transaction without a matching document
  • Filter by amount range, vendor, date
  • Drag a document onto a row to link it
  • Surface as a Finn task at month-end

Capabilities at a glance

6 document subtypes detected (timesheet, PO, quotation, expense report, delivery note, credit note)

Structured extraction with line items, hours, rates, PO numbers

Auto-linking to transactions, invoices, bills, estimates, contacts, projects

Many-to-many junction table (document_links) with typed relationships

Three create-from-document Finn tools (invoice, bill, estimate)

Editable line items on document detail page

Payroll-from-timesheet flow

Unlinked transaction finder

GET /api/document-links endpoint for client-side queries

<LinkedDocuments> server component on every detail page

How It Works

1

Upload

Drop a document via any channel β€” web, email, WhatsApp, Slack, Telegram, Gmail.

2

Detect & extract

Gemini Vision identifies the subtype and pulls structured data β€” line items, hours, references, totals.

3

Auto-link

The document is matched to existing transactions / invoices / bills / contacts via reference number, vendor + amount, and contact name.

4

Create or attach

Ask Finn to turn the document into an invoice, bill, or estimate. Or leave it linked as supporting evidence. Either way, it's tracked.

By the numbers

Structured
Extraction
Line items, hours, rates, PO numbers, totals β€” not just text
Many-to-many
Linking
One document can link to many entities; one entity to many documents
Typed
Relationships
supporting, source, generated, related β€” every link knows its role
Editable
Line items
Fix what's wrong on the document detail page without re-uploading

Real-world Use Cases

Contractor timesheets β†’ invoices

Contractor sends a weekly timesheet PDF. Finn extracts hours per project, multiplies by the contractor's rate, and drafts an invoice ready to send. You confirm.

Supplier POs β†’ bills

Your supplier sends a PO confirmation. Drop it in; Finn creates the bill, line items intact, linked to the supplier contact.

Customer quotations β†’ estimates

You drafted a quote in Word. Drop the PDF in; Finn creates a proper estimate in AccountsOS, ready to be converted to an invoice when won.

Reconciliation evidence

Every transaction has its supporting document attached β€” receipt, contract, PO, delivery note β€” auto-linked. Your audit trail builds itself.

Plays nicely with

Gemini Vision

Live

Frontier OCR + structured extraction

Resend inbound

Live

Forward documents to documents@accounts-os.com

Gmail

Live

Auto-scan inbox for receipts and invoices

WhatsApp / Slack / Telegram

Live

Drop a document, get it linked

Claude Desktop (MCP)

Live

Query linked documents from your AI

AccountsOS vs OCR app + spreadsheet

CapabilityAccountsOSOCR app + spreadsheet
What it understandsReceipts, timesheets, POs, quotations, expense reports, delivery & credit notesReceipts only
Extracted fieldsStructured: line items, hours, rates, PO numbersVendor / total / date only
LinkingAuto-linked to transactions, invoices, bills, estimates, contacts, projectsManual
Create recordsInvoice / bill / estimate from one documentRe-type into accounting software
EditingEditable line items, no re-uploadEdit the export or start again
Audit trailSource document linked to every recordManual cross-reference

Why founders pick this over the spreadsheet

No more re-typing data from PDFs

Auto-linking eliminates the 'which transaction does this go with?' problem

Create-from-document turns paperwork into records, not work

Audit-ready: every record has its source document attached

Works across every ingest channel β€” upload, email, messaging

Editable extraction means you can fix without re-uploading

Frequently Asked Questions

How is this different from receipt scanning?

Receipt scanning gets vendor / date / total. Document Intelligence goes further β€” detecting the document subtype (timesheet, PO, quotation, expense report, delivery note, credit note), extracting structured data (line items, hours, rates), and auto-linking to existing records. Create-from-document tools then turn the document into an invoice, bill or estimate.

What's the document_links table?

A many-to-many junction table that links documents to transactions, invoices, bills, estimates, contacts, projects, expense claims. Each link is typed: supporting, source, generated, related.

Does this run on every upload?

Yes β€” auto-linking runs after every upload regardless of channel (web, email, WhatsApp, Slack, Telegram, Gmail). The same logic runs when an existing document is reprocessed.

Can I see the structured data?

Yes β€” every document has a detail page showing the extracted_data.structured_data payload. Line items are editable inline.

What if extraction is wrong?

Edit the line items inline on the document detail page. No re-upload needed. You can also reprocess to ask Gemini for another pass.

Can Finn turn a timesheet into payroll?

Yes β€” the payroll-from-timesheet tool maps timesheet entries to a payroll run. Hours Γ— rate, by employee, ready to submit.

Where do auto-linked documents show up?

On the linked entity's detail page β€” open an invoice, you see the source documents; open a contact, you see every document linked to them.

Can I query linked documents from outside?

Yes β€” GET /api/document-links?entity_type=invoice&entity_id=xxx returns the links. The same data is exposed through MCP for Claude Desktop.

Ready to try Document Intelligence?

Get started with AI-powered accounting for your UK limited company.

Included on every plan. Unlimited documents on Free Early Access.

Get Started Free

Free during Early Access β€” no credit card β€” cancel anytime