PDF Intelligence API

Recognition
is not
comprehension.

The only REST API that reads engineering drawings — not just sees them.

Vision models recognize patterns. The PDF Intelligence API extracts typed vector geometry — every line, arc, polygon, and text label with real-world coordinates, layer names, and color attributes. Structured JSON. One API call.

PDF INTELLIGENCE API
# Turkish Airlines cabin layout — page 1 curl -X POST https://api.visual-integrity.com/v1/extract \ -H "X-API-Key: $VI_KEY" \ -F "[email protected]" → 200 OK { "total": 8723, "polylines": 3821, "polygons": 3591, "text": 1059 ... 8,722 more objects }

30 yrs
PDF R&D
Production-proven since 1995

18 formats
Vector, image and text
DXF, DWG, SVG, WMF, JPG, TIFF+

5 types
Object Geometry
Line, polyline, polygon, image, text

100+
API Parameters
Precision control over every output

One click on pink.
Every Metro stop — highlighted.

We ran the Obama inauguration route map — a public PDF from the National Archives — through the PDF Intelligence API. Here is what came back versus what a vision model sees.

The document’s own structure becomes the index.

When you click the pink swatch in the color panel — every Metro stop illuminates simultaneously. Not because we told the system what Metro stops look like. Because the cartographer used pink consistently, and the API surfaces color as a queryable semantic signal.

Each stop is a compound object: the M symbol, the text label, and the individual station entrances — nested together, extracted as a group. This is not recognition. This is comprehension.

Vision model output
Pattern recognition "This appears to be a detailed street map of Washington D.C., centered on the National Mall area. The map shows major landmarks including the White House, the Capitol, and various monuments..."
✕  Cannot identify object types
✕  Cannot extract coordinates
✕  Cannot query by color or layer
✕  Cannot identify compound objects
PDF Intelligence API — inauguration map
PATHS
street geometries, boundaries, route
61,513
POLYGONS
parks, plazas, federal buildings
1,314
TEXT LABELS
location names with coordinates
1,234
POLYLINES
inauguration route
40
✓  All objects typed, layered, coordinates-exact
✓  Queryable by color, type, or layer
✓  Compound objects identified as groups

Three endpoints. Complete intelligence.

The API reads native PDF operators using the same geometry the original CAD system wrote. No rasterization. No OCR. The document’s actual structure, returned as JSON.

01 — EXTRACT

Vector object extraction

POST any PDF. Receive every vector object as structured JSON.

POST /v1/extract
-F “[email protected]
02 — CONVERT

Format conversion

Convert to DXF, DWG, SVG, PNG and 14 more formats.

POST /v1/convert
-F “profile=cad-layered”
03 — SUMMARIZE

AI drawing intelligence

Geometry-grounded AI summaries. Every claim traceable.

POST /v1/summarize
-F “[email protected]

Built on 30 years of PDF R&D.
Read the technical brief.

Architecture, extraction methodology, and benchmark comparisons — for engineering and technical due diligence teams.

Your Name(Required)

 

No sales calls. We’ll send the PDF and follow up only if you ask.