Schema-constrained AI parsing API

AI that turns any document into structured data with evidence.

Q: What does Fieldry do?

Fieldry turns any PDF/JPG/PNG into structured JSON/XML matching your schema. Every value includes confidence and evidence (page + source text + bounding box).

Q: How is this different from OCR?

OCR returns raw text. Fieldry returns your schema: grammar-constrained generation plus format validators, with null-honesty — missing fields return null instead of hallucinations.

Q: How does pricing work?

$1 = 100 credits. 4 credits per page, +2 for tables. Free 200 credits/month. Usage is transparent in the X-Credits-Used header. Top-ups from $20 with up to 25% bonus, batch saves 50%.

Q: How do agents use Fieldry?

Via REST API, Python/JS SDKs, n8n node, or the MCP server — Claude, ChatGPT and Cursor call it as a tool and get cited values back.

Q: What about PII and compliance?

EU hosting option, DPA from day one, and a self-host path for sensitive workloads. Evidence and audit logs ship by default.

Q: Which documents work best?

Invoices, receipts, IDs, bank statements, proof-of-address, BOL/CMR and insurance claims — across RO/TR/DE/BG and Cyrillic scripts.

POST a PDF/JPG/PNG + your schema — Fieldry's AI returns cited JSON/XML back, value, confidence and bbox on every field. Missing stays null, never invented.

Start free · 200 credits/mo See the API

POST /extract?wait — invoice.pdf + schema

{
  "invoice_no": {
    "value": "INV-2026-041",
    "confidence": 0.99,
    "evidence": [{
      "page": 1,
      "text": "INV-2026-041",
      "bbox": [0.21, 0.08, 0.38, 0.11]
    }]
  },
  "total": {
    "value": 1240.50,
    "confidence": 0.97,
    "status": "ok"
  },
  "vat": { "value": null,
    "status": "missing" }
}

X-Credits-Used: 40How evidence works →

≥99%
schema-valid output
<10s
p95 API latency
~$0.0003
infra cost / doc
4
credits / page (+2 table)

Platform

One API for the whole document lifecycle

Four hubs, one credit meter. Start with extraction, add trust and automation when you need them.

ExtractFile + schema in, validated JSON out.ExtractParseTablesExplore Pre-processSplit, classify and batch before extraction.SplitClassifyBatchExplore Trust LayerEvidence, validation and forensics.EvidenceValidationForensicsExplore DevelopersREST, MCP, SDKs, n8n and webhooks.REST APIMCP ServerSDKs & WebhooksExplore

Evidence on every field, schema guaranteed

Pick a capability to see what it returns and how it looks in Fieldry.

invoice_no0.99

total · 1.240,50 EUR0.97

iban · RO49…review

missing_vat → nullnull

Schema-constrained extract

POST a PDF/JPG/PNG with a JSON/XML schema. Grammar-constrained generation plus validators guarantee ≥99% schema-valid output — no repair loops.

Custom fields + tables as rows
Null-honesty: missing = null, never invented
JSON and XML output

Feature 1 of 5

How it works

Live in an afternoon, not a quarter

01
Upload
POST a PDF/JPG/PNG + your JSON/XML schema. Sync with ?wait or async 202 with webhook callback.
02
Parse & interpret
RapidOCR extracts boxes + text, the interpreter maps them to your schema — multilingual, no dictionaries.
03
Validate with evidence
Format validators run, every value ships with confidence, page and bbox. Missing stays null.

Pricing

$1 = 100 credits. No contracts.

4 credits per page, +2 for tables. Free 200/month. Every response carries X-Credits-Used.

See $20 / $100 / $500 tiers Batch saves 50%

Use cases

Documents teams parse every day

Each pattern has a dedicated page with schema tips, evidence handling and setup.

KYC Onboarding
IDs and proof-of-address to verified JSON in one call.
Solved with:Extract ·Classify ·Evidence ·Validation
Proof of Address
Utility bills and bank statements mapped to address schema.
Solved with:Extract ·Validation ·Evidence
Invoice AP
Invoices to ERP-ready line items with table extraction.
Solved with:Extract ·Tables ·Batch
Bank Statements
Lending decisions on cited transactions, not OCR dumps.
Solved with:Extract ·Tables ·Validation
Logistics Docs
BOL, CMR and customs forms across RO/TR/DE/BG languages.
Solved with:Split ·Classify ·Extract
Insurance Claims
Claim packs with forensics signals before payout.
Solved with:Extract ·Forensics ·Evidence
Agent & RAG
Give Claude/Cursor grounded doc tools via MCP.
Solved with:Parse ·MCP Server ·Evidence
Freight Invoice Audit
Carrier invoices checked against rate con, BOL and POD — duplicates held.
Solved with:Split ·Tables ·Validation
Financial Spreading
Tax returns, pay stubs and statements to spread-ready rows + income check.
Solved with:Extract ·Tables ·Evidence

Industries

Built for regulated volume

Plugs into your stack

Start with the API, call it from agents. Same evidence-rich output everywhere.

Have more questions?

The questions parsing teams ask most before they sign up.

Talk to us

Fieldry turns any PDF/JPG/PNG into structured JSON/XML matching your schema. Every value includes confidence and evidence (page + source text + bounding box).

OCR returns raw text. Fieldry returns your schema: grammar-constrained generation plus format validators, with null-honesty — missing fields return null instead of hallucinations.

$1 = 100 credits. 4 credits per page, +2 for tables. Free 200 credits/month. Usage is transparent in the X-Credits-Used header. Top-ups from $20 with up to 25% bonus, batch saves 50%.

Via REST API, Python/JS SDKs, n8n node, or the MCP server — Claude, ChatGPT and Cursor call it as a tool and get cited values back.

EU hosting option, DPA from day one, and a self-host path for sensitive workloads. Evidence and audit logs ship by default.

Invoices, receipts, IDs, bank statements, proof-of-address, BOL/CMR and insurance claims — across RO/TR/DE/BG and Cyrillic scripts.

Parse your first document today

Start free with 200 credits/month.

An ops team drowns in invoice PDFs on Friday. With Fieldry, each one returns cited JSON the same afternoon — no template setup, no repair loops.

Request a Demo See pricing