productivity

Nanonets Review: Schema-Driven Document Extraction (2026)

Name: Nanonets
Availability: InStock
Author: AI Demos

Schema-first PDF extraction that produces usable exports, but dense table rows still need review.

Visit Nanonets

Custom JSON schemasBank statementsInvoice line itemsCSV/JSON exports

TL;DR — our verdictUpdated July 2026 · 6 test artifacts

Strong schema extraction, weaker row fidelity

Where it wins

You need direct PDF uploads with custom JSON schema extraction for bank statements or invoices.
You need downloadable CSV/JSON exports and a review UI for inspecting extracted rows.
You can tolerate some row-level cleanup in exchange for strong document-level structure.

Main limitation

You need bank-statement transaction rows to be lossless on the first pass.

Pricing (verified plans)

Free $200 in usage credits includedGrowth CustomEnterprise Custom

Strongest test artifacts

Schema-Driven Document Extraction →Row-Level Table Extraction →Multi-Format Export →

Our take

Nanonets handled direct PDF uploads, custom schemas, and downloadable exports well, and the earlier page positioned it for plain-English questions over extracted data. In this report, the weak spot is record-level fidelity: bank-statement transactions were often merged or incomplete, and one invoice line item came through partially, so it works best with human review before downstream use.

Bank statement extraction demo recording.

In-Depth Review

Our detailed analysis of Nanonets — features, performance, and real-world testing.

AI Demos Team

Expert Reviewer

Verified Review

Feature-by-Feature Breakdown

Schema-Driven Document Extraction

Strong for known schemas and top-level fields.

▾

Test Summary

Feature tested: Schema-Driven Document Extraction

Result: Partial — Strong for known schemas and top-level fields.

Nanonets accepts direct PDF uploads plus a custom JSON schema and populates nested document objects rather than flattening everything into generic OCR text. In the tested bank-statement and invoice runs, it reconstructed metadata, account details, balances, summary fields, and invoice metadata.

file

Bank Statement PDF.pdf

↓→

file

docstrange-nanonets-bank-statement-output.json

Loading file...

The bank-statement output follows the requested schema closely, with top-level metadata, account details, statement period, balances, and summary objects populated in structured JSON rather than returned as flat OCR text.

file

Invoice PDF.pdf

↓→

file

docstrange_nanonets-invoice-output.json

Loading file...

The invoice output also follows the supplied schema, filling invoice metadata, advertiser, station, account, billing, remit, line_items, and summary sections as structured JSON.

Bottom Line

Best when the schema is known ahead of time; document-level structure came through reliably.

Row-Level Table Extraction

Mixed: invoice rows were mostly clean, but bank-statement transactions were noisy.

▾

Test Summary

Feature tested: Row-Level Table Extraction

Result: Partial — Mixed: invoice rows were mostly clean, but bank-statement transactions were noisy.

The tool extracts tabular data as records from documents such as the bank statement and invoice. In the tested runs, transaction rows and invoice line items were produced, though some dense rows needed cleanup.

file

Input artifact for "Row-Level Table Extraction" test: INPUT, bank-statement-18-jun-transactions.png

↓→

file

docstrange-nanonets-bank-statement-output.json

Loading file...

The bank-statement extraction preserved row structure but not row fidelity: some transaction descriptions merged neighboring content, several rows were missing dates, and transaction_type remained null across the dataset.

file

Input artifact for "Row-Level Table Extraction" test: INPUT, invoice-line-item-8.png

↓→

file

Output artifact for "Row-Level Table Extraction" test: Line item 8 was extracted as a separate record, but it came through partially: the description merged adjacent text, and flight-period / frequency fields were missing even though similar records were populated., docstrange-nanonets-line-item-8.png

Line item 8 was extracted as a separate record, but it came through partially: the description merged adjacent text, and flight-period / frequency fields were missing even though similar records were populated.

Bottom Line

Good enough for cleaner invoice tables, but bank-statement transactions needed the most cleanup.

Multi-Format Export

Strong export support across common handoff formats.

▾

Test Summary

Feature tested: Multi-Format Export

Result: Passed — Strong export support across common handoff formats.

The tool can download or copy extracted data in multiple downstream formats, including JSON, CSV, HTML, and Markdown. The tested report confirms handoff files in CSV and JSON for extracted bank and invoice data.

file

Bank Statement PDF.pdf

↓→

file

docstrange-nanonets-bank statement-output.csv

Loading file...

The bank-statement extraction was exported as CSV, and the report says the same run also supported JSON, HTML, Markdown, and copy-to-clipboard output with consistent data across formats.

file

Invoice PDF.pdf

↓→

file

docstrange-nanonets-invoice-output.csv

Loading file...

The invoice extraction was exported as CSV, matching the report's claim that JSON, CSV, HTML, and Markdown exports were available and stayed consistent with the structured extraction.

Bottom Line

A practical downstream handoff layer, assuming the upstream extraction quality is acceptable.

Free

$200 in usage credits included

No platform fees or commitments. Includes Data Extraction AI, API access, email integration, cloud storage connectors, community support, and supports up to 3 users. Pay only for usage beyond the included credits.

Growth

Custom (Volume-based pricing)

Quote-based plan for scaling teams. Adds Classification AI, barcode & signature detection, Generative AI blocks, custom Python blocks, ERP/database integrations, AI reporting & analytics, team-wide credit sharing, custom integrations, and up to 40% volume discounts.

Enterprise

Custom (Tailored to processing volume)

Includes everything in Growth plus enterprise security and deployment features such as SAML SSO, SCIM, RBAC, HIPAA & SOC 2 compliance, private cloud/on-prem deployment, regional data residency, Salesforce/SAP/Oracle connectors, audit logs, SIEM integration, dedicated support & SLAs, and white-label UI.

✓ Use This If

●You need direct PDF uploads with custom JSON schema extraction for bank statements or invoices.

●You need downloadable CSV/JSON exports and a review UI for inspecting extracted rows.

●You can tolerate some row-level cleanup in exchange for strong document-level structure.

✕ Skip This If

●You need bank-statement transaction rows to be lossless on the first pass.

●You need every invoice line item to be fully populated without post-check review.

●You need a documented correction or audit workflow from the tested interface.

productivityother-productivitytext

Yes. The report says both the bank statement and invoice were uploaded directly through the UI with no preprocessing needed.

Yes. Both test runs used custom JSON schemas, and schema validation passed on the first try.

The report explicitly says JSON, CSV, HTML, and Markdown exports were available, and it also mentions download and copy-to-clipboard support.

High-level metadata came through well, but transaction rows were noisy. The report says descriptions were often garbled or concatenated, transaction_type was null, and dates were missing on 15+ transactions.

Mostly good at the metadata and line-item level, but not perfect. The report says all eight line items were extracted, while program descriptions were concatenated and line 8 was missing several fields such as flight period and frequency.

The bank statement took 1 minute 9 seconds, and the invoice took 32.7 seconds.