Mindee review, pricing and verdict

Developer API that turns invoices, receipts, IDs and other documents into structured JSON, with prebuilt models, custom schemas and confidence scores.

Best for
Backend and platform engineers building document ingestion pipelines
Not for
Non-technical operations teams without engineering support to integrate an API

By , Founder and editor

Should you use Mindee?

Pick Mindee when developers need to turn invoices, receipts and IDs into JSON through an API without training a model, and want confidence scores to route doubtful fields to a reviewer. Skip it if you want a finished accounts-payable app, free production use, or output built mainly for LLM pipelines.

Prebuilt models for common business documents plus an editable schema, which Mindee says remove the need to train a model.

Prebuilt models and schema editing shorten setup, though the build-or-buy answer in Mindee's pricing FAQ puts highly proprietary documents and strict sovereignty rules on the build side.

Also for

  • Product managers at fintech and insurtech companies automating accounts-payable workflows
  • Engineering teams building KYC or customer-onboarding flows that need ID and passport verification
  • Accounting and expense-management software vendors embedding receipt and invoice capture

Also not for

  • Teams needing a finished, no-code AP automation product with a built-in UI

Main trade-off

No free plan and per-page credits that expire each period make large or irregular volumes costly to budget.

Mindee pricing

PlanMonthly priceWhat it includes
Starter$53/mo1 full seat, Unlimited models, Confidence scores and polygons, Utilities and chaining, RAG unlimited, Community support
Pro$134/moEverything in Starter, Live chat support, Unlimited seats, Data processing localization (Europe or US), Annual billing: $107 platform fee plus $27 for the smallest credit bundle
EnterpriseCustomEverything in Pro, Custom pricing and custom SLAs, Dedicated account manager and premium technical support, 30 minutes of free setup and 2 hours of solution engineering, Annual contract, 500,000+ credits
How the pricing works, in full

Prices are in USD, and every plan is a platform fee plus a credit bundle. At the smallest bundle, 500 credits a month (6,000 a year on annual billing), Starter totals $53 a month billed annually ($636 a year: a $26 platform fee plus $27 for credits, about $0.054 a credit) and Pro totals $134 a month billed annually ($1,608 a year, with a $107 platform fee). Monthly billing costs $72 a year more on Starter and $180 more on Pro, per the savings the page shows. Larger bundles run up to 300,000 credits a year, and Enterprise is custom-priced with an annual minimum of 500,000 credits. There is no free plan; the pricing page offers a 14-day free trial with no credit card.

What each plan includes

FeatureStarterProEnterprise
Seats1UnlimitedUnlimited
SupportCommunityLive chatPremium, account manager
EU or US processing choice—✓✓
Custom SLAs——✓
CommitmentMonthly or annualMonthly or annualAnnual, 500,000+ credits

Before you commit

  • Credits are billed per page, and a page returned with confidence scores costs 1.5 credits instead of 1.
  • Overage is billed at the plan rate plus a surcharge, and unused credits lapse each period.
  • EU or US processing choice and live chat are gated to Pro.

What Mindee does

Mindee is a Paris-based document-processing API, founded in 2018 and part of Y Combinator's Winter 2021 batch, that returns structured JSON from invoices, receipts, ID cards and other files. It offers prebuilt models, custom schemas and classify, split and crop utilities, with SOC 2 Type II certification and a choice of EU or US processing on Pro.

  • Data extractor with custom schemas

    Start from a prebuilt model or define your own fields; each value comes back typed, normalised, located on the page and scored.

  • Confidence scores and polygons

    Every value can carry a score and bounding coordinates so doubtful fields go to a reviewer, on every plan including Starter.

  • Classifier, splitter and edge detector

    Label each upload, split a long scan into separate documents of up to 200 pages, and crop several documents photographed on one page.

  • Intent prediction

    Uploading one sample file builds a schema for the document's purpose, such as any proof of address rather than one electricity bill; Mindee reports that over 80% of users start from a sample.

Show all 7 features

  • SDKs and no-code connectors

    Official client libraries across six languages, with Zapier, Make and n8n connectors for no-code flows.

  • Retention controls

    Extracted results are kept 12 hours by default, 1 to 24 configurable, with a delete-when-fetched option; the original file is never written to disk.

  • Any alphabet and common formats

    Accepts PDF, JPEG, PNG, TIFF, HEIC and WEBP files, including handwritten forms, in any alphabet.

Works with

  • Python
  • Node.js
  • Java
  • .NET
  • PHP
  • Ruby
  • Zapier
  • Make
  • n8n

Capabilities

  • Document AI
  • OCR
  • Data extraction
  • Invoice processing
  • Kyc
  • Computer vision
  • API first
  • Custom model training

Strengths and limits

Strengths

  • Prebuilt invoice, receipt, ID and passport models plus an editable schema let most teams skip training, according to Mindee.
  • Retention is documented in concrete terms: a short default window, a delete-on-fetch option and a statement that customer documents are not used for training.
  • Six official SDKs and three no-code connectors cover most stacks without raw REST work.
  • Confidence scores come with every plan, so review routing needs no upgrade.

Limits

  • No free plan: a time-limited trial is the only free path, and even Starter carries a paid minimum before any overage.
  • Credits are per page and expire at the end of each period, with a higher rate for pages returned with scores.
  • Starter is one seat with community support; live chat, unlimited seats and the EU or US processing choice need Pro.
  • It is an extraction API, so review screens, approvals and accounting sync have to be built around it.

How Hokai rates it

  • Ease of use8.0 / 10
  • Features7.5 / 10
  • Value for money6.5 / 10
  • Support6.5 / 10

Mindee alternatives

ToolPrice fromFree planRuns onRating
Mindee This page$53/moTrialWeb3.6
ReductoPick Reducto if your pipeline feeds RAG or AI agents; pick Mindee if you need pretrained invoice, receipt and ID extraction out of the box.--YesWeb3.9
LlamaIndexPick LlamaIndex if you're building a broader agentic document-and-retrieval workflow; pick Mindee for a dedicated, pretrained document-extraction API.FreeYesWeb--
VeryfiPick Veryfi for receipts, invoices and expenses with a free plan to start; pick Mindee for custom schemas and classify, split and crop utilities.$500/moYesWeb · iOS · Android--

Mindee facts

Deployment
Cloud
Compliance
SOC 2 Type II · GDPR · CCPA
Works offline
No
Minimum skill level
Intermediate
Price band
Mid-range
Maker founded
2018
Training-data policy
Customer documents are not used to train Mindee models (docs.mindee.com data processing policies); results are held for a short, configurable window.
Training
Documentation · Quickstart · Tutorials · Community feedback board
Total funding
$14M Series A (October 2021, led by GGV Capital; Alven, Serena Capital and Bpifrance Digital Venture took part); total raised not confirmed

Data and privacy

Proprietary

Questions about Mindee

How much does Mindee cost in 2026?

Starter costs $53 a month on annual billing at the smallest credit bundle, and Pro costs $134 a month on the same terms; both include a platform fee and a credit bundle, and larger bundles are available. Starter has one seat and community support, while Pro adds unlimited seats, live chat and a choice of Europe or US processing. Enterprise is quoted individually and requires an annual contract.

Is Mindee free to use?

No. Mindee offers a free trial, but its live pricing page lists no free plan, and every paid plan starts at the Starter tier. Mindee's llms.txt file still mentions a free tier of 500 pages a month, which the pricing page no longer shows, so read the pricing page rather than summaries of it.

How does Mindee count credits?

One processed page is one credit, and a page returned with confidence scores is 1.5 credits. Utilities such as classify, split and crop cost 0.25 credits a page, waived for pages that go on to a linked extraction model. Credits expire at the end of each billing period, and usage beyond the bundle is charged at the plan's rate plus a 20% surcharge so the API never stops.

Does Mindee store my documents or train on them?

Mindee's data processing policy says the original file is never stored to disk and extracted results are kept for 12 hours by default, adjustable from 1 to 24. An option deletes results as soon as they are fetched or delivered by webhook, which suits zero-retention workflows. The same documentation states that customer documents are not used for training Mindee models.

Which document types and file formats does Mindee handle?

Prebuilt models cover invoices, receipts, ID cards, passports, bank statements, resumes and bills of lading, and a schema editor handles other documents. Mindee reads PDFs and image files such as JPEG, PNG, TIFF, HEIC and WEBP in any alphabet, including handwritten forms in its homepage demo, and the splitter is built for long multi-document scans.

About Mindee

Mindee is a Paris document-processing company, founded in 2018, whose API turns invoices, receipts, IDs and other files into structured JSON. It came out of stealth in October 2021 with a $14 million Series A led by GGV Capital, joined by Alven, Serena Capital and Bpifrance's Digital Venture fund; Y Combinator lists it in its Winter 2021 batch with a team of 60, and TechCrunch named Qonto, Payfit, Spendesk and Lucca as early customers.

Read the full background

The platform has four building blocks. A data extractor reads fields into a schema you define or start from a prebuilt model such as invoice, receipt, ID card or passport, a classifier labels each upload, a splitter separates a multi-document PDF (up to 200 pages and 100 MB per file), and an edge detector crops several documents photographed on one page. Mindee says no model training is needed, and a recent product update, intent prediction, builds a broader schema from one sample file instead of one tuned to that exact document.

Each extracted value can return a confidence score and a bounding polygon, so low-scoring fields go to a human reviewer. Developers get official SDKs for Python, Node.js, Java, .NET, PHP and Ruby, and no-code connectors for Zapier, Make and n8n; accepted files include PDF, JPEG, PNG, TIFF, HEIC and WEBP.

On data handling, Mindee's documentation says customer documents are not used to train its models, the original upload is never written to disk, and extracted results are kept only briefly by default, with the window configurable. Pro plans can pin processing to Europe or the United States. Mindee lists SOC 2 Type II and GDPR on its site, and CCPA on its homepage.

Where it loses: teams that feed documents into an LLM or retrieval pipeline may prefer Reducto or LlamaIndex, and expense and receipt capture buyers can compare Veryfi. Mindee's own January 2026 roundup of OCR APIs also names Nanonets, ABBYY, Google Document AI, Amazon Textract, Rossum and Doxis, and its build-or-buy page says proprietary documents or strict data-sovereignty rules favour building. It is an API for engineers, not a finished accounts-payable product, and it has no free plan. The AI SDK directory lists related developer tools.

Sources and guides

Official sources

Product facts on this page (pricing, plans, integrations, compliance) are read from Mindee's own pages and documentation. The verdict is Hokai's reading of them, not the vendor's claim.

Hokai guides that mention Mindee

Lists that include Mindee

More AI Tools on Hokai

  • BAMLProgramming language for agents that generates typed SDKs for your own language and records every call; free, open source, with hosted tracing in early access.
  • InsForgeOpen-source AI agent backend with 12,800+ GitHub stars: $25/mo Pro, free tier. 2.4x fewer tokens than Supabase MCP on MCPMark v2. YC Spring 2026, 6-person team.
  • TiptapOpen source rich text editor framework backed by a 6,000-member Discord community; Tiptap Cloud adds AI-agent editing and paid collaboration plans.
  • DailyDaily provides WebRTC SDKs and the open-source Pipecat framework for building realtime voice and video AI agents, with 75+ global points of presence.
  • AgentPhoneAgentPhone gives any AI agent its own US or Canadian phone number via 1 API call, with unified voice, SMS, real-time transcription, and built-in MCP support.
  • AstrocadeFree phone-first game maker and feed: describe a game, AI builds a playable mini game, and it joins more than 1,000 new games published each day.