Search engines, language models and AI agents only understand what has been written down. FullText reads everything else — scans, handwriting, faxes, photos, video, sound and 3D space — and turns it into clean, verified, searchable text.
Pixels are for eyes. Text is for machines. Every search box, every model, every agent reasons in words. Whatever hasn't been turned into text — the scanned, the handwritten, the photographed, the spoken — simply doesn't exist to them.
If a record can't be read, it can't be indexed, searched, or cited.
AI is only as trustworthy as the text underneath it. Every word needs a confidence score and a source.
Before analysis, before answers, before agents — something has to read the page.
Anything a person can see or hear goes in on the left. On the right comes text a machine can trust — structured, searchable, and traceable back to the exact pixel it came from.
Each stage is useful on its own. Together they close the gap between what an organization has and what its AI can actually use.
Finds the structure before the words: tables, stamps, signatures, seals, margins, photos and the regions that matter.
Printed type, cursive, typewriter carbons, faded microfilm and fax noise — read with a confidence score on every character.
Turns pages into records: parties, dates, amounts, parcels, clauses and codes, mapped to the schema you already use.
Exact phrase when you need precision, meaning when you don't know the words. Every hit opens on the highlighted source.
Ask questions across millions of pages and get answers that cite the book, page and line they came from.
Every word keeps its coordinates, page and confidence. Low-confidence text routes to human review instead of into your model.
The proof point for "everything becomes text" is the record that was never meant for a computer: the handwritten deed book at the county courthouse.
Every parcel has a paper trail — deeds, mortgages, liens, easements, plats, surveys and notary acknowledgments. Much of it lives in handwritten ledgers, typewriter carbons, microfilm reels and image-only scans that no search box can see.
So title work is still done by eye: an examiner paging book by book to prove a chain of ownership. FullText reads all of it — every hand, every era, every format — and turns a county's history into one searchable, citable record.
grantor: Calloway
easement: 15 ft Elender: First Fed.
ref: DB 312/44releases: OR 88/210
status: satisfiedgrantee: utility co.
width: 10 ftsetback_E: 1.2 ft
flag: encroachmentA pixel is a square of light. A voxel is a cube of space. FullVoxel takes LiDAR, drone photogrammetry, CT and building scans and writes them down as text an AI can search: what's there, how big it is, and what it's too close to.
Send a file, a folder or a stream. Get back text, fields, coordinates and confidence — ready for your search index, your database or your model's context window.
# Read a 1923 quitclaim deed and pull the fields that matter curl https://api.fulltext.ai/v1/read \ -F file=@deed_book_312_p44.tif \ -F schema=land_record # → { "text": "THIS QUITCLAIM DEED, executed this 14th day of March…", "fields": { "type": "quitclaim_deed", "date": "1923-03-14", "grantor": "Josiah P. Calloway", "parcel": "NE1/4 SW1/4 S14 T28S R24E", "easements": [{ "width_ft": 15, "side": "east" }] }, "provenance": { "page": 44, "line": 7, "bbox": [212,88,604,112], "confidence": 0.97 } }
The faded, the handwritten, the scanned-sideways. Tell us what you need read and we'll show you what's inside it.