Ledger
From statement to spreadsheet,
without the re-keying
Ledger pairs a meticulous AI parser with a workflow built for review: upload in bulk, watch each file extract, preview the rows, and walk away with clean workbooks - through the browser or the batch API.
A pipeline tuned for financial documents
Read
Ledger pulls the text layer from each PDF. If a document turns out to be a scan, the pages are read as images with built-in OCR - automatically, with no separate step.
Extract
Claude parses the document as a "meticulous financial-document parser": every line item, values exactly as printed, nothing invented, account numbers masked to the last four digits.
Deliver
Structured rows become a two-sheet Excel workbook with a smart file name. Preview in the browser, download, batch-download, or email a secure link.
Each document type gets the columns it should have: bank and card statements come back as Date / Description / Amount / Balance; 401k and brokerage as Date / Fund / Units / Price / Value; check registers as Date / Check # / Payee / Amount / Balance; general ledgers as Date / Account / Reference / Debit / Credit / Balance.
The product tour
What you see between dropping a PDF and opening the workbook.
Upload in bulk, hint once
Drag a folder's worth of PDFs onto the dropzone - "multiple allowed" is the default, duplicates are skipped automatically, and a single optional Document type hint applies to the whole batch. Or leave it on auto-detect and let each file identify itself.
- Text-based PDFs up to 40 MB each; scans OCR'd up to ~16 MB
- Six document types plus auto-detect
- Live progress as the batch extracts: "Extracting 2/5..."
Review before you rely on it
Every file in the batch gets a live status, the detected document type, a row count, and an engine tag showing exactly how it was processed. Click any result to preview the account metadata and the first rows inline - then download the full workbook.
- Per-file status: queued, extracting, done, or a clear error
- Inline preview of the first 25 rows, full detail in the download
- Download files one at a time or all at once
Ninety days of history, always re-downloadable
Every extraction lands in your History: when it ran, the original file, the smart workbook name, type, row count, whether it came in through the web or the API, and when it was last downloaded. Re-download anything for 90 days - then it is automatically and permanently purged.
- Per-user history with one-click re-download
- Web and API extractions side by side
- Automatic 90-day retention sweep - storage and records both
Automate the volume work
For engagements measured in boxes of statements, the key-authenticated batch API takes up to 100 files per batch, processes them in the background, and holds the results until you fetch them. A ready-made PowerShell client handles upload, polling, and smart-named downloads.
- Asynchronous: upload returns immediately, results wait for you
- Optional email-on-complete with secure, expiring links
- Access restricted to allowlisted networks, key checked on every call
Built for documents you can't afford to leak
Ledger runs as a single-sign-on web app inside a private AWS environment, and treats your statements like the sensitive records they are.
Single sign-on
Corporate SSO in front of every page - no app-managed passwords - with access gated by user and admin groups in your identity provider.
In-account AI
Extraction runs on Claude via Amazon Bedrock within a private AWS account - your documents are processed inside that trust boundary, not shipped to a third-party SaaS.
Retention with an end date
Workbooks are kept 90 days, then swept - files and records both. Email links expire after 24 hours. Source PDFs are removed after processing.
Audit trail
Every extraction, download, email, and batch event is logged with who did it, which engine ran, and what came out.
Masked by default
Account numbers are masked to the last four digits in every workbook and preview - full numbers are never transcribed.
Locked-down API
The batch API is reachable only from allowlisted networks and validates a per-client key on every request - defense in depth, not either-or.