Any into a queryable sheet.
Guides for verifiable document OCR
Practical, code-checked walkthroughs — receipts and invoices, line items, CSV export, and the audit trail behind every extracted value.
Scans & photos, turned into data
Photos and scans land as rows in a sheet, every cell traces back to where it came from, and one click drops a CSV — all of it, in one place.
Any layout into the sheet you want
Receipts, faxes, handwritten notes — set the columns once and they land in a familiar sheet, just like Excel.
Click a cell, see the area on the photo
Wonder where the AI got that value from? Click the cell and it lights up on the photo. If something's off, fix it right there — the change history saves itself.
Specify only the fields you want extracted
Purchase orders, invoices, bankbooks — each job needs different fields. Set the columns yourself, or let AI suggest them from a sample photo. Then just upload — the AI finds what you asked for.
Drop your photos right onto the sheet
Fifty purchase orders, a whole stack of receipts — drop them straight onto the sheet. Each photo becomes one row, and the results stack up automatically.
Pull scattered docs into one search
Drop everything into Spaces, attach memos, and search across all of it at once. "Where did I put that?" — gone.
From document to data you can verify
All real product, no mockups: drop a document and rows fill in, hover any value to see its exact place on the page, then search every scan to land on the verified cell.
One API key, your whole space
One API key, no SDK. Define fields, get structured JSON — every value with verified coordinates. Unconfirmed cells are flagged for review, not silently wrong.
One call, structured JSON
curl -X POST https://api.space-ocr.com/ocr/fields \
-H "Authorization: Bearer YOUR_API_KEY" \
-d '{ "image": "receipt.jpg", "imageType": "url",
"fields": [{ "name": "total" }, { "name": "date" }] }'
# → "total": "12,000", "date": "2025-04-10"
# field_bboxes.total: { text_verified: true, needs_review: false }Why build on it
Structured JSON, one call
Define fields once;
POST /ocr/fieldsreturns schema-mapped JSON, nested arrays included.Verifiable values
Every field carries
bbox+vertices, cross-checked by two engines (text_verified).Flags what it can't trust
Unverified cells return
needs_reviewinstead of a silent wrong value.Consistent output
Sheet text accuracy ~0.98 run-to-run, up from a jittery ~0.73.
Async + signed webhooks
Batch, poll
/jobs, HMAC-signed webhook. REST · JSON, OpenAPI 3.1.