Optional add-on module
OCR software that turns scanned paper into searchable truth.
The enterprise OCR software extracts text from scanned plans and documents across alphabets, produces searchable PDFs, and feeds full-text search. It runs on your hardware or ours, and never touches the protected originals.
Board_Minutes_2026_Scan.pdf
412 pages · searchable PDF + .txt
Technical_Report_ES.pdf
page 268 of 391 · engine: es
Plant_Overview.pptx
queued · engine: auto-detect
## Page 268 of 391, confidence 0.97
…the design pressure of the vessel shall not exceed 24 bar in accordance with…
invisible text layer injected · original scan untouched
OCR capabilities
Extraction without exposure
Asynchronous extraction
Documents are uploaded by streaming, queued, and processed in the background with page-by-page progress. You can close the browser — the job continues on the node.
Bilingual EN/ES
The node ships in English and Spanish, and its recognition catalog covers the Latin family — Spanish, English, Portuguese, French, German, Italian — plus Cyrillic, Chinese, Japanese, Korean, Arabic, and Devanagari. Spanish output gets a generic correction pass for accents, acronyms, and table artifacts.
Full-text indexing
Extracted text feeds the search index directly, enabling contextual discovery down to the exact page across the entire repository.
Queued, throttled, observable
Workers adjust automatically to CPU and RAM every 30 seconds, and resource guards reject or shrink jobs when the node is saturated instead of degrading it. Every task reports status, progress, and errors.
Machine-to-machine Scan API
A server-to-server API lets your platform submit scans, track progress, and download the searchable PDF under its original filename. Jobs are idempotent per document reference, and completion can arrive by webhook or polling.
Native in the viewer
Users request OCR from the file browser — one document or a whole folder — and admins can queue every scanned PDF that has no text. Progress is visible per page, and the searchable PDF lands back in place, gated by existing permissions.
What it does
From dead pixels to living text
Scanned documents are images — invisible to search, impossible to audit. Document OCR turns them back into text your platform can actually use.
Searchable PDF generation
Scanned documents come back as searchable PDFs with an invisible OCR text layer under the original image — same look, every word selectable and indexable. The .txt and the PDF come from the same recognition pass, so they always agree, and the original filename is preserved.
Hybrid per-page extraction
Pages are evaluated individually. Native text is kept when it is clean; fragmented or image-only pages are OCR'd. Native PDFs can also pass through a layout-aware extractor (OpenDataLoader, with an optional docling-fast hybrid mode) that preserves reading order for complex tables and technical drawings.
Multi-script UTF-8 recognition
Latin (Spanish, English, Portuguese, French, German, Italian), Cyrillic, Chinese, Japanese, Korean, Arabic, and Devanagari — the engine picks the alphabet that reads each page best, and models (~10 MB each) download on demand. The invisible text layer uses fonts with real glyph coverage, including logical ordering for right-to-left scripts.
Adaptive capture quality
Short documents — maps, blueprints, forms — render at 400 DPI for maximum glyph fidelity; long documents run at 200 DPI. Low-confidence pages are automatically retried at high DPI before the text layer is written.
Enterprise operation
Built for production loads, not demos
The OCR node manages its own resources, protects itself from overload, and leaves no residue behind.
Bulk queue processing
An asynchronous queue processes documents in bulk with live page-level progress, persistent history, cancellation, and a one-hour per-task timeout. Files stream in up to 3 GB, bounded dynamically by the node's RAM and disk.
Self-tuning workers
The worker pool re-evaluates CPU and RAM every 30 seconds. Under overload the node rejects or shrinks jobs — stricter limits kick in below 512 MB of RAM or 10 GB of disk — instead of degrading.
Authenticated & audited
HMAC-signed sessions, bcrypt-hashed credentials, per-user task and file isolation, and an audit trail of every upload, download, and deletion with IP and user-agent. API keys for integrations are stored hashed.
24-hour automatic retention
Processed tasks and files are purged automatically after 24 hours by default, and any task can be deleted on demand. The OCR node never becomes an uncontrolled copy of your document repository.
Integration
Two ways in: your systems or your viewer
Aegis OCR is designed to be called, not babysat — by platform services over a documented API, or by users directly from the viewer.
Server-to-server Scan API
Native in the viewer
Deployment options
Run it in our cloud — or on your hardware
Aegis OCR is offered fully managed in our cloud, with zero infrastructure on your side. If your policy requires on-premise processing, these are the minimum requirements for a dedicated node.
Aegis Cloud Recommended
On-premise node
Minimum requirements when contracting local processing.
No GPU is required — inference runs entirely on CPU. The node can share the same server as your Aegis Secure View deployment, and the optional layout-aware companion runs alongside it.
Bring your hardest documents
Send us a sample of your scanned plans and we will show extraction quality on your own files.