Extraction is the easy part. Shipping it means idempotent requests, signed webhooks, broad coverage, and the security and deployment controls a real team depends on.
PDFs, scans, images, and Office files in; fields, tables, and relations out, mapped to a schema you define.
We don't just flatten text. Tables are detected, reconstructed, and emitted with their structure intact: headers, merged cells, and row/column spans are preserved rather than flattened into a wall of text, even nested tables spanning multiple pages. Export to JSON, CSV, or Excel.
Native and scanned PDFs processed with state-of-the-art OCR. Complex layouts, multiple columns, and rotated or low-quality scans are handled for you.
Define the fields you want extracted with a template, or let the engine identify the data relevant to your use case. Typed output mapped to your schema.
Recognizes the financial semantics of a document (distinguishing line items from totals, taxes, and currency across invoices and receipts), so reconciliation gets typed numbers instead of strings.
Native support for 20+ languages out of the box, with no per-language configuration. Automatic format detection spans native and scanned PDFs, images, and Office files.
Every value comes back with a per-field confidence score and pixel-level traceability: the coordinates and page it was read from, so you can validate automatically or show a user exactly where a number came from.
A documented REST API, webhooks, and async batch processing, so the engine fits the systems you already run.
A fully documented REST API with intuitive endpoints and semantic versioning so upgrades never surprise you. Predictable typed JSON responses you can validate against your own schema.
Real-time notifications when documents finish processing. Register multiple endpoints and filter by event type; payloads are signed so you can verify origin, and delivery is retried until acknowledged.
Submit large documents and receive results via callback, with no blocking of your application. Bulk jobs run asynchronously and report back through webhooks, so a thousand-page batch never ties up a request thread.
A predictable request budget per plan (from 10 req/min on Starter to unlimited on Enterprise), so throughput scales with your tier instead of throttling unexpectedly.
The operational guarantees a team needs to ship and keep shipping.
An idempotency key per request means a retried call never double-charges or double-processes, so a flaky network never turns into duplicate work.
Run in our infrastructure, or deploy on-prem for sensitive workloads. The same engine, wherever your compliance posture requires it to live.
99.95% uptime observed across all tiers over the last 12 months; a contractual SLA is set per plan (99% to 99.99%). The numbers are measured, not promised.
We keep uploaded files and extracted data only as long as needed to deliver the service. Retention terms are defined per plan; Enterprise retention is set in your agreement.
The essentials at a glance. The full posture, subprocessors, and DPA live on the compliance page.
All traffic is protected with TLS; uploaded documents and extracted data are encrypted at rest in our storage and database layers.
Access to production data is restricted, authenticated, and logged, following least-privilege principles.
Extraction runs on our own engine. You control whether your documents ever reach external model providers; nothing is exposed to third parties without your choice.
Waitlist and contact forms use Cloudflare Turnstile, a privacy-friendly anti-bot challenge that does not track users across sites (Cloudflare is a disclosed subprocessor).
Join the waitlist and we will send your invite when self-serve opens. 500 pages to test the API, no credit card to start.