XRuns inside Excel

Fill your model. Prove every number.

modelflow reads your documents and proposes a value for every field in your Excel model. You check each one against its source and decide. No number is written without your approval, and even ten years later the journal lets you trace who wrote what, from which file.

  • Human in the loop
  • Your documents are never stored
  • EU servers, no AI provider
  • Right inside Excel

Why modelflow

Extraction is a commodity. Accountability isn’t.

Any AI can pull numbers out of a PDF. The hard part comes later, when someone asks where a number in your model came from and you have to answer.

01

Copy and paste leaves no trace

Hours of typing figures out of 300-page reports, and afterwards nothing shows which figure came from where.

02

Autopilot AI fills cells nobody checked

A tool that writes straight into your workbook turns one misread unit into a valuation that is off by a factor of a thousand.

03

The questions come years later

An auditor, a court, the other side’s advisor. When they ask, “the AI did it” is not an answer.

Autopilot tools
AI reads AI writes You hope
modelflow
AI proposes You verify You approve The journal remembers

How it works

From document to model in four steps. You decide at each one.

  1. 01

    Map your model once

    Open your workbook. modelflow suggests fields like revenue, EBITDA and total assets from your own labels, and you confirm each one. The one-time mapping takes minutes and applies to every extraction after that.

    AI-assisted mapping
  2. 02

    Load your documents

    Annual reports, scanned PDFs, Excel files, DATEV exports and XBRL filings, in any language. Your documents are not stored on our servers.

    PDF · XLSX · DATEV · XBRL
  3. 03

    Review field by field

    Each proposal shows the figure as printed and the value it becomes. Values that pass every check are accepted automatically. The rest wait for you. One click opens the source page with the number highlighted. Accept, correct or discard.

    Visual provenance
  4. 04

    Approve the write

    modelflow writes only what you approved, asks before it touches a formula and records every change in the journal. Changed your mind? One click undoes the whole write.

    One-click undo

Human in the loop

The AI proposes. A person decides.

One rule sits above everything else in modelflow: no value is written to a workbook without a person approving it. Everything else follows from that.

0

values written without a person approving them. That is not a setting, it is how the system is built.

Automatic only when every check passes

A value is only accepted automatically once modelflow can verify it in many different ways. If a check fails, it comes to you.

The figure as printed

You see “48,213” and “in EUR thousand” next to 48,213,000. A wrong unit or a wrong sign is obvious at a glance.

Source on click

One click opens the source page with the figure highlighted. You see the place itself, not just a page reference you have to look up.

Formulas are protected

A formula cell is never overwritten silently. modelflow warns you and asks first.

Model version checked

If a row moved since you mapped it, modelflow notices before writing and lets you pick the right cell.

One-click undo

Every write is one transaction. Undo restores exactly what was there before.

Audit trail

Who wrote what, when, from which file. For ten years.

Every approved write goes into a journal. It is kept for ten years and, if it comes to it, outlives the user account or the map it came from.

Proof without keeping the file
The journal stores a file’s fingerprint, never the document itself. Your archived copy still lets you trace the source.
Every change, in full
Old and new value, cell, map version, who approved it and when. Also whether the proposal was corrected or a formula overwritten.
Independent of file names
Rename it, save it under a new name, move it to OneDrive: the workbook keeps one continuous trail.
Export for your auditor
The journal as a CSV file, filtered if you like.
Change journal Nordwerk_DCF.xlsx · everyone
Export CSV
TimeApproved byFieldCellOld → newSource
14 Mar 2026, 10:42A. BeckerRevenueDCF!D4empty 48,213,000 Annual_Report_2025.pdf sha256 9f2c…e41a
14 Mar 2026, 10:42A. BeckerEBITDADCF!D5empty 7,904,000 Annual_Report_2025.pdf sha256 9f2c…e41a
14 Mar 2026, 10:42A. BeckerTotal assetsDCF!D6empty 61,337,000 correctedAnnual_Report_2025.pdf sha256 9f2c…e41a
14 Mar 2026, 10:42A. BeckerEmployeesDCF!D8empty 1,284 Annual_Report_2025.pdf sha256 9f2c…e41a
12 Mar 2025, 16:05M. JansenRevenueDCF!C4empty 44,870,000 Annual_Report_2024.pdf sha256 4b7e…07cd
12 Mar 2025, 16:05M. JansenEBITDADCF!C5empty 7,112,000 Annual_Report_2024.pdf sha256 4b7e…07cd
$ sha256sum Annual_Report_2025.pdf 9f2c81d0a4e7b95c3d1f0e8a7b6c5d4e3f2a1b0c9d8e7f6a5b4c3d2e3b61e41a matches the journal

Privacy by architecture

Your documents are read, not kept.

Privacy is not a promise on paper. It is built into the architecture.

Never stored

Documents stay in memory while you review them, then they are gone. Never written to disk, never kept in a database.

No AI provider in between

European open-weight models run on GPUs we operate ourselves. No model provider and no AI cloud sits in between.

Never trained on, never sold

No model is trained on your documents, and nothing you upload is shared, sold or reused.

Own servers

Dedicated servers in ISO 27001-certified data centres in the EU, run by a company based in Germany. A data processing agreement is available.

Who it is for

For anyone whose work lives in an Excel model.

Valuation

DCF and multiples models, filled with revenue, EBITDA and net debt from annual reports, across companies and years.

Accounting & tax

Financial statements, DATEV exports and XBRL filings, straight into the analysis and reporting workbooks you already use.

Anyone with a model

Analysts, investors, founders, private individuals. If your spreadsheet has fields and your numbers come from documents, modelflow fits.

For model providers

Sell an Excel model? Ship the map with it now.

Publish your maps to your clients. They drop in their documents and fill your model from day one, with no mapping or setup on their side.

  • Your clients use your maps without setting anything up
  • See which clients use which map, and how often
Talk to us about publishing maps

Free to start. Then one flat price per seat.

Map one model and fill it once for free, with no credit card and nothing to cancel. After that you pay one flat price per seat. No tokens, no credits, no per-page fees, and fair use applies. You never have to count pages or budget tokens.

Start now

FAQ

Questions and answers.

Does modelflow really run inside Excel?

Yes. modelflow is an add-in that opens as a side panel in Excel on Windows, Mac and the web. The workbook itself is never uploaded.

Which documents can it read?

PDFs, including scans, Excel files, DATEV exports and XBRL filings. They do not have to be in English or German.

Do you store my documents?

No. After processing, only the journal entry remains: file name and fingerprint, never the content.

Which AI do you use?

We run our own servers. We do not use OpenAI, Google, Anthropic or any other model provider, not even as a fallback.

What if the AI gets a number wrong?

You see it before it counts, not after. Every value comes with the figure as printed and the highlighted source. You can check every value. Correct a wrong value in place or discard it. The journal records the correction.

Can modelflow overwrite my formulas?

Not without asking first. Formula cells are detected before anything is written and need an explicit second confirmation.

How long can I prove what was written?

Every journal entry is kept for ten years. You can also export the journal as CSV at any time and keep it in your own archive.

Is there a data processing agreement?

Yes. We provide a data processing agreement under Art. 28 GDPR. Our servers run in ISO 27001-certified data centres in the EU.

What does it cost?

Starting is free: map one model and fill it once, no credit card needed. After that you pay one flat price per seat, with no tokens, credits or per-page fees. Fair use applies.

Get in touch

See modelflow on your own model.

Tell us what you work on and which documents you fill it from. We reply the same day.

Prefer email? info@jumoca.de

What is your focus?