Mindee

Mindee is a Paris-founded document AI company offering a REST-based intelligent document processing (IDP) platform that extracts structured data from invoices, receipts, passports, IDs, payslips, bank checks, resumes, and custom document types. Its positioning sits between the hyperscaler OCR services (AWS Textract, Google Document AI, Azure Document Intelligence) and newer LLM-based extraction: it ships pre-trained, off-the-shelf models that work without a data-science team, plus custom model building and a 'Continuous Learning' (RAG) capability that lets models improve from corrected examples without full retraining. The current API generation (api-v2.mindee.net) is organized into five model families - Extraction, Classification, Crop, OCR, and Split - with a unified asynchronous Jobs/polling endpoint and webhook support, signaling a maturing, batch-oriented architecture aimed at production document pipelines.

Commercially, Mindee runs a transparent page-credit pricing model: one credit per physical page regardless of document type, format, or extraction complexity, which makes cost forecasting unusually predictable compared to per-field or LLM-token billing. The published tiers (Starter EUR44, Pro EUR179, Business EUR584, Enterprise custom) bundle progressively richer features - team seats, RAG document counts, confidence scores, polygon/localization data, and faster support. There is no permanent free plan, only a 14-day trial, which is a friction point for hobbyists relative to Textract or open-source options. Notably, Mindee also stewards docTR, a well-regarded MIT-licensed open-source OCR library (~6k+ GitHub stars) built on PyTorch, which strengthens its developer credibility and gives self-hosting teams a path that competitors lacking an open-source arm cannot match.

Sentiment is strongly positive among the developer and ops teams who adopt it. On G2 and Capterra it carries a 4.8/5 rating, with reviewers consistently praising the API-first design, clean documentation, fast integration, and responsive support; Capterra cites multi-year retention and price-performance wins over Microsoft Document Intelligence. The main criticisms are fluctuating accuracy on certain languages and layouts (e.g. typos on Polish-language invoices), occasional connection/latency issues, and slower support response on lower-tier plans. Third-party and competitor benchmarks place its field-level accuracy around 96% with mid-range latency, making it a credible specialist choice - though buyers should weigh the lack of a free tier and the comparatively smaller review base against the entrenched cloud incumbents.