Local OCR cache (optional)
For scanned PDFs, ExtractMint may OCR pages. You can persist an OCR cache locally so repeat exports are faster.
Core conversion runs in your browser. The optional assistant and site analytics use separate, limited data flows.
The selected statement and extracted rows are processed in the extension and held temporarily in memory. They are not transmitted to ExtractMint or another service, and the extension does not save them. The extension does not read bank pages or monitor downloads.
When you submit a question with a converted document active, the browser sends the question and a compact context to ExtractMint’s /api/agent/ endpoint. The context includes the file name, document type, currency, statement period, row count, overall confidence and validation details, plus up to five extracted rows with dates, descriptions, amounts, deposits, withdrawals, balances, and row confidence. The original PDF is not included.
If an AI provider is configured for the assistant, the question and context are forwarded to that provider. Otherwise, the endpoint uses built-in guidance. Avoid asking about a document if you do not want these selected details sent to the endpoint and, when configured, the AI provider. Review the provider’s privacy terms for its handling and retention.
Financial records and imported backups stay in this browser. For exchange rates, the tracker sends the selected base-currency code to Frankfurter; it does not send records or backup files. ExtractMint measures page views without receiving tracker records or backups.
Receipt photos and PDFs are read in your browser. ExtractMint stores imported receipt copies and expense or income records in a separate IndexedDB database in this browser profile so you can review them later. This tool does not send receipt files, extracted fields, categories, notes, or amounts to ExtractMint servers or an AI provider, and it does not sync records to another device.
You can delete a record with its saved receipt copy or clear all records from the tool. Clearing this browser's site data also removes these local records. Page-view analytics do not receive receipt or ledger details.
ExtractMint uses Vercel Web Analytics for page views and three fixed product events: upload_started, conversion_completed, and export_started. The synthetic example has four separate demo-only events: sample_conversion_started, sample_conversion_completed, sample_conversion_failed, and sample_conversion_cancelled when a real upload takes precedence. These demo-only events carry no properties. All seven events include no custom properties, so no filenames or statement contents are sent with them. Vercel describes Web Analytics as cookie-free and based on anonymized data.
ExtractMint also uses Minilytics, hosted at data.iangoh.com, to measure visits and session engagement. It sets first-party visitor and session cookies (ml_vid for up to one year and ml_sid for up to 30 minutes) and records page paths, referring domains, campaign labels, browser and device details, approximate location, and time spent on pages. These analytics do not receive uploaded files, filenames, statement contents, receipt images, or financial records. You can opt out in this browser by visiting a page with ?ml_ignore=true; visit with ?ml_ignore=false to opt back in.
For scanned PDFs, ExtractMint may OCR pages. You can persist an OCR cache locally so repeat exports are faster.
For review-mode matching, ExtractMint can cache per-page text indices locally in your browser.
Use the in-app Trust & privacy panel and click “Clear all local caches” for accountant workflows.