Your documents never train a model, ours or a provider's

Extract document data. Check it against the source.

Extract structured data from invoices, receipts, purchase orders, delivery notes and e-invoices. Check a value against the place it was read from, correct exceptions, then export it or send it on by API and webhook.

View API

No card required · 100 free pages

sample-invoice.pdf Illustrative sample

Document previewPage 1

Every outlined value was located on the page. The stronger outlines are the ones Source check can select.

Extraction complete

Source checkSelected fields. Choose one to see its source.

Result

17 fields including a line-item table of 2 rows

Output · Excerpt 
{  "invoice_number": "INV-2024-0892",  "issue_date": "2024-12-01",  "seller_name": "TechSupply GmbH",  "buyer_name": "Horizon Logistics sp. z o.o.",  "total_amount": "3456.30",  "due_date": "2024-12-31",  "line_items": [
    { "description": "Barcode Scanner Model XS-400", "quantity": "5" }
  ]}

An excerpt of the structured result. The complete one carries every field of this sample and every line-item row. Open this sample in the app

  • 0 Documents used for AI training Ours or a provider's. Content sent to a provider is used only to produce your result.
  • 100 Free pages, no card A one-time allowance on the free plan, with every feature included.
  • $29/ month 1,000 pages per month The Pro plan. Predictable page-based pricing; the same features on every plan.
  • XML E-invoices read from their XML Factur-X · ZUGFeRD · KSeF FA(3) · UBL 2.1

Clean results move on. Uncertain ones get your attention.

Every result is validated. A document that passes is completed and delivered. A failed check or a low-confidence value marks the document as needing review and raises document.review_required in addition to document.completed. A reviewer opens the source, corrects the value and verifies it. Correcting a field does not verify it; a person does.

invoice-2026-0851.pdf Document Reviewed

4 of 4 required fields verified

  • Invoice number INV-2026-0851 Verified
  • Seller TechSupply GmbH Verified
  • Seller VAT ID DE123456789 Verified
  • Total gross 3 456.80 EUR Verified

Total gross · the exception

Source
3 456.80
Initially extracted
3 458.80 58%
Outcome
Corrected, then verified by the reviewer
Invoice number, seller and seller VAT ID were verified. The total was extracted with a confidence of 58%, so the document was routed to review. The reviewer inspected the source, corrected 3 458.80 to 3 456.80 and verified the field. The correction alone did not verify it. With the last required field verified, the review completed.

Correct it once. See what DocSolved remembers.

Correct a field on a supplier's digital PDF in the review workspace and DocSolved saves a supplier note. The next digital PDF from that supplier you analyse there is extracted with the note applied, and the result says so. Notes are visible in the workspace and you can remove any of them.

March · digital PDF

IBAN DE89 3704 0041… DE89 3704 0044…

Corrected · supplier note saved

April · digital PDF

IBAN DE89 3704 0044…

Supplier notes: 1 applied for TechSupply GmbH

Two digital PDFs from the same supplier: the IBAN corrected on the first, the note applied on the second.

When the data is already structured, DocSolved reads it.

Factur-X, ZUGFeRD, KSeF FA(3) and UBL 2.1 invoices carry a structured XML layer. DocSolved takes their values from that XML: no OCR and no model call for those values, and no page box, because the value comes from an element rather than a page. If the XML is embedded in a PDF, the total and the payment account are also compared with the printed page, and a mismatch is flagged. AI extraction runs only for documents without such a layer.

XML

<ram:GrandTotalAmount>3456.80</ram:GrandTotalAmount>

JSON

"total_amount": "3456.80"
Read from the XML element, not from a page.

See a KSeF FA(3) e-invoice in the live demo KSeF in Poland: how DocSolved reads FA(3)

DocSolved handles the document. Your stack handles the business.

Send documents in by upload, email or the REST API. DocSolved extracts, validates, routes uncertain results to a person, keeps the corrections, and ends every job with an event. Posting, paying, filing and notifying stay in the tools you already run.

A job's status is completed, review_required or failed. The webhook events are not one-per-job: every successful job emits document.completed, and document.review_required is sent in addition when the result needs a person.

DocSolved document lifecycle

  1. receive
  2. extract
  3. validate
  4. route
  5. review when needed
  6. correct
  7. optional approval
  8. deliver

Webhook events

  • document.completedevery successful job; the result is ready
  • document.review_requiredalso sent, with document.completed, when validation or routing marks that result for a person
  • document.approvedlater, only when an approver approves the document in a workspace with approval enabled
  • document.failedprocessing failed; no result, nothing charged

REST · async jobs · HMAC-signed webhooks · JSON · CSV · XLSX · UBL 2.1 XML · MCP

Your business workflow

  • ERP
  • CRM
  • payment
  • Slack
  • database
  • internal service
  • automation tool
  • human attention
The document lifecycle ends at the event. Everything after it is yours.

n8n: a workflow template and HTTP guidance. Make and Zapier: their webhook and HTTP modules, no native app. MCP: coding agents can discover the integration, inspect its capabilities and wire extraction into your project. Tested with Codex, Claude Code and Gemini CLI.

npx -y @synairodocai/docai-mcp@next install

API documentation Webhook events and signing Explore MCP integration

Check the product, not the claim.

DocSolved is a small vendor, so nothing on this page asks to be taken on faith. Everything it says can be inspected.

  • Sample documentsthe review workspace on real files, no account needed
  • API referenceevery endpoint, the compatibility policy, webhook signing
  • Benchmarkswhat is measured, how, and what is not measured yet
  • Securitypractices, subprocessors, what leaves your account
  • Changelogwhat changed, and when

Predictable, page-based pricing

Priced by pages. Every feature on every plan. A failed analysis is not charged. Your remaining balance is preserved. Each billing cycle tops it back up to your plan allowance; it is not added on top.

Free

$0 100 free pages to start

  • 100 pages, one-time, yours until used
  • 100 MB storage · 50 documents

Recommended

Pro

$29 / month

  • 1,000 pages per month
  • 5 GB storage · 2,000 documents

Business

$99 / month

  • 4,000 pages per month
  • 20 GB storage · 10,000 documents

See all plans

Frequently asked questions

What happens when extraction is uncertain?

Every result is validated. A failed check or a low-confidence value marks the document as needing review, and a job then raises document.review_required next to document.completed. In the review workspace you open the source, correct the value and verify it. Correcting a field does not verify it by itself.

Do I need to review every document?

No. A document that passes validation is completed and delivered; only documents that fail a check or carry a low-confidence value are marked as needing review. Fields are never verified automatically unless an administrator turns on confidence-based auto-confirmation, which is off by default.

Does DocSolved integrate with n8n, Make or Zapier?

Through HTTP and signed webhooks. For n8n there is a ready-made workflow template and a step-by-step guide; Make and Zapier connect through their webhook and HTTP modules. DocSolved does not ship its own n8n node, Make app or Zapier app.

How is DocSolved different from a raw OCR or document AI API?

A raw API returns values and leaves the rest to you. DocSolved also validates each result, routes uncertain ones to a person, keeps supplier notes from your corrections, reads structured e-invoices from their XML instead of a page, and ends every job with a signed event, so your workflow tool does not have to rebuild the review step.

Where are my documents processed?

AI analysis sends up to the first three pages of a document to the AI provider configured for the deployment: OpenAI (United States) or OVH AI Endpoints (European Union). The privacy policy lists every subprocessor and what each one receives.

Do my documents train AI models?

No. Your documents and extracted data are never used to train AI models, ours or a provider's. Content sent to a provider is used only to produce your result.

How much of a long document does the AI read?

DocSolved renders the first 5 pages and sends the first 3 to the model. Long documents are read in part, not in full. The same limits are stated on the pricing page.

What happens when I upload an e-invoice?

If the file carries a structured XML layer (Factur-X, ZUGFeRD, KSeF FA(3), UBL 2.1), DocSolved reads the values from the XML: no OCR and no model call for those values. If the XML is embedded in a PDF, the total and payment account are compared with the printed page and a mismatch is flagged. AI extraction runs only for documents without that layer.

Is there an API?

Yes. A REST API with asynchronous jobs, email ingestion, and signed webhooks, on every plan. The API reference is public.

Is the free plan really free?

Yes. Create an account and the first 100 pages are free, one-time, no card required. On a paid plan your remaining balance is preserved, and each billing cycle tops it back up to your plan allowance; it is not added on top.

Which languages and countries does DocSolved work in?

The interface is available in English, German, French, Spanish, Polish and Italian, and documents in these languages are extracted with their own field labels understood. Regional variants add country-specific fields automatically: EU VAT details, US sales tax, India GST. You can also pick the region yourself before analysing.

See it on your own documents.

Upload a document to extract, review, and organize its data. Start free, no card required.

See it on a sample
AI extraction