Book a demo
Features

From a stack of documents to a reviewed data record – in six steps

Excalitract supports the entire process: you define what you need, upload documents, review the values against the original and hand reviewed data over to your systems.

Excalitract pulls data out of documentsA stack of documents on a stone base from which a glowing blade is drawn. Three marked source locations on the front document are linked to a result table: invoice amount, service period and contract number, each reviewed.ResultInvoice amount€1,284.50Service period1 Jan – 31 Dec 2025Contract numberV-2041-7Source: page 1Excalitract pulls data out of documentsA stack of documents on a stone base from which a glowing blade is drawn. Three marked source locations on the front document are linked to a result table: invoice amount, service period and contract number, each reviewed.ResultInvoice amount€1,284.50Service period1 Jan – 31 Dec 2025Contract numberV-2041-7Source: page 1
The Excalitract flow in six stations
1 / 6Intake

Documents arrive as a folder or ZIP – mixed or sorted, digitally created or scanned. Duplicates are detected.

  1. 1

    Create a project and document types

    A project brings together what belongs together: tenant, team and the document types you want to process – such as incoming invoices or supply contracts. Roles for reading, reviewing and managing are assigned per project.

    Project: Incoming invoicesInvoiceCredit noteRoles: Read · Review · Manage
  2. 2

    Define the schema – via AI draft or editor

    Describe the fields you need and provide an example. Excalitract drafts a schema from it and presents any ambiguities as notes for you to confirm one by one. Alternatively, you build fields, tables and layout in the visual editor – without any JSON.

    Example + descriptionSchemaAmount · NumberDate · DateLine items · ListNote: assumptionplease confirm
  3. 3

    Upload documents

    Drag an entire folder or a ZIP archive into the browser. The folder structure is preserved and duplicates are detected. Sample documents with explanations help the extraction with difficult cases.

    Your document chaosStructured resultsFolder or ZIPDocument → FieldsInvoiceAmount · DateContractTerm · PartiesReportFigures · Tables
  4. 4

    Start a run with a cost estimate

    Before every run you see a non-binding cost estimate. Scanned pages are first made machine-readable by state-of-the-art AI text recognition (OCR). During processing you follow progress live; processing times per document are measured and stay traceable.

    Cost estimate before startYour cost estimate is readyReview before startingWaiting for your confirmationNo documents processed yet.You decide when processing starts.
  5. 5

    Review against the original

    Every value is shown next to the original document. One click jumps to the source location. You confirm, correct or mark values as incorrect – every change is saved as a revision. Missing required values visibly stay open.

    Net amount€1,284.50Not reviewed yet
  6. 6

    Approve and export

    You export approved documents as Excel® with one sheet per list level, as JSON or via the REST API. For open fields, you start targeted follow-up runs – values that have already been accepted remain untouched.

    Excel®JSONREST API
Project: Incoming invoicesInvoiceCredit noteRoles: Read · Review · Manage
The core

A source location for every value

The model cites the passages it takes a value from, and Excalitract matches these citations against the recognized text. For digitally generated PDFs, a box highlights the exact location in the document. Where no reliable position is available, it honestly says “Source location unresolved” – instead of showing a guessed box.

Extracting a value with its source locationIn the original document, the location of a value is highlighted. A line connects the highlight to the extracted field “Consumption 2025: 48,210 kWh”, which carries the source location page 2, line 14.Page 2Field: Consumption 202548,210 kWhSource: p. 2, line 14From value to sourceHighlighted in the originalExtracting a value with its source locationIn the original document, the location of a value is highlighted. A line connects the highlight to the extracted field “Consumption 2025: 48,210 kWh”, which carries the source location page 2, line 14.Page 2Field: Consumption 202548,210 kWhSource: p. 2, line 14From value to sourceHighlighted in the original
In detail

What else Excalitract offers

Native MCP server

AI agents use Excalitract as a tool via the Model Context Protocol™: submit documents, start runs, fetch results with source locations and trigger reviews.

State-of-the-art text recognition

AI-based text recognition (OCR) makes scanned pages and photos machine-readable – including tables and multi-column layouts. Digital PDFs can use their text layer directly.

Recognition profiles

Choose between fast and thorough profiles for each run. Text recognition and the model chain are configured, not programmed.

Examples with explanations

Add sample documents with notes on difficult cases. Excalitract takes them into account during extraction.

Bulk approval with preview

Approve many documents at once. Open required fields, errors or values marked as incorrect block approval.

Tables and lists

The schema models line items, meters or periods as lists. The Excel® export gets one sheet per list level, linked via IDs.

REST API with OpenAPI

All features are available via a documented REST API – with an OpenAPI description and API keys per tenant.

Microsoft® sign-in and roles

Your team signs in with a Microsoft® account. Access is granted only with an explicitly assigned role: read, review or manage.

Revisions and audit log

Corrections and review decisions are stored as immutable revisions and can be traced in the audit log.

Costs under control

Before every run, Excalitract shows a non-binding cost estimate and afterwards the actual usage – per tenant, with a quota.

Duplicates and folder structure

During upload, Excalitract detects duplicate files and keeps the folder paths so that the results match your filing structure.

Targeted rework

Open fields? Only those are reprocessed.

Follow-up runs specifically target open or incorrect fields – optionally with a more thorough recognition profile. Text recognition is reused, values that have already been accepted remain untouched, and for every field you can see whether it came from the model, was carried over or was set by a human.

“Not found” never becomes zeroA field that is not in the document is not filled with 0. The zero is crossed out, the field visibly stays open and is marked “Required value missing”. The document goes to review, and approval is only possible once the value has been resolved.no entryField: Meter number0— not foundRequired value missing→ to a human for reviewApproval blocked until the value is resolved“Not found” never becomes zeroA field that is not in the document is not filled with 0. The zero is crossed out, the field visibly stays open and is marked “Required value missing”. The document goes to review, and approval is only possible once the value has been resolved.no entryField: Meter number0— not foundRequired value missing→ to a human for reviewApproval blocked until the value is resolved
Demo

See Excalitract with your documents

No standard show: we prepare the demo for your documents and fields. Afterwards, you will know how Excalitract handles your document collection – and how you can measure it on a test set of your documents.

Book a demoSee features