# Lector

> Lector is an AI platform for autonomous document processing, built by lector.ai GmbH in Bremen, Germany. It takes incoming documents (post, email, scans, PDFs, photos, fax), classifies and separates them, extracts and validates the relevant data, and hands structured results to downstream systems — without templates or separator sheets. Available as an EU-hosted cloud service, hybrid, or fully on-premise.

Website: https://lector.ai — German is the default; every page also exists in English, replace `/de/` with `/en/` in any URL.
Product: https://app.lector.ai — free trial with 100 AI page processings.
Company: lector.ai GmbH, Konsul-Smidt-Straße 8p, 28217 Bremen, Germany · info@lector.ai · +49 421 40887940 · founded 2022 as a spin-off of JUST ADD AI GmbH (part of the JAAI Group) · self-funded, profitable, independent of investors.
Team: Dr. Benjamin von Ardenne (CEO and founder), Dr. Thies Gerken (CTO and founding engineer); the JAAI Group counts more than 100 employees across eight AI companies.

## What Lector does
- Intake from any channel: email, REST API, SFTP and file systems, watch folders, scanners, web upload, messaging and webhooks — PDF, Office formats, images, TIFF, ZIP and EML containers
- Classification and separation: recognises document types inside bundled scans and splits them into sub-documents. No separator sheets, no templates; new document classes are described in plain language instead of trained
- Extraction: vision language models read fields, tables, handwriting, stamps and photos. Certain AI models also return a confidence score per value
- Validation: field rules, cross-checks between documents, master-data matching, completeness and plausibility checks on applications, ready-made wording for missing documents
- Routing and export: straight-through processing above the confidence threshold, human verification below it. Export via webhook, REST pull, email, DMS/ECM, ERP, databases and RPA tools
- Evaluation Center: measures quality on real documents — metrics per document class, confusion matrices, and a comparison of models and settings before they go live
- Audit trail: every operation on every document is recorded immutably, searchable and exportable. Usage statistics via GraphQL, system metrics via Prometheus

## Deployment and AI models
- Documents that must not leave the organisation: run Lector fully on-premise — platform and AI models inside the customer's own network, no data leaves the data centre
- Cloud (SaaS, hosted in the EU), hybrid (platform in your own data centre, AI inference in the EU), or fully on-premise including local models on your own GPU
- Models are chosen per processing step: Gemini, GPT, Claude, Mistral, Qwen, InternVL — open and closed source, reached through European endpoints (Google Vertex AI, Microsoft Azure, OVHcloud, Scaleway). Switching models needs no rebuild
- Custom models for enterprise use cases; Keycloak identity management, OIDC/SAML single sign-on, role-based access control, multi-tenancy
- On-premise minimum: Ubuntu, 16 CPU cores (32+ recommended), 32 GB RAM (64 GB recommended), around 1 TB storage, and a GPU such as an Nvidia H100 (80 GB) or RTX 6000 Ada (48 GB)

## Security and compliance
- GDPR-compliant since the company was founded. In the Lector Cloud, data and AI models are processed exclusively in Europe; with hybrid or on-premise deployment the documents stay on the customer's own servers
- Cloud hosting on OVHcloud (Germany/France), ISO 27001 and BSI C5 Type 2 certified; all connected AI endpoints are ISO 27001 certified
- No customer data is used to train models — documented per provider in the no-training statements
- TLS in transit, AES-256 at rest, role-based permissions, least privilege, audit logging, single sign-on
- ISO 27001: certification in preparation, audit planned for autumn 2026; information security management system established; regular internal security reviews, external penetration tests in preparation
- EU AI Act: minimal risk under recital 53 — classifying and reading documents does not trigger high-risk obligations
- Statutory health insurers in Germany are excluded from the standard cloud terms; such projects run under individual agreements

## Pricing (cloud list prices, net, monthly or yearly)
- Free trial: 100 AI page processings, one-off
- Starter: 4 cents per AI page processing (standard models), no base fee, cancellable at any time. From roughly 15,000 AI page calls a month, Business S is the cheaper option
- Business S: 599 EUR per month (499 EUR when billed yearly), 16,000 AI page processings included, 3.2 cents per additional page
- Business M: 1,199 EUR per month (999 EUR yearly), 42,000 included, 2.4 cents per additional page
- Business L: 2,399 EUR per month (1,999 EUR yearly), 125,000 included, 1.6 cents per additional page
- Enterprise: individual — multi-tenancy, SLAs, dedicated onboarding, custom models, millions of pages per month
- On-premise and hybrid are priced individually. For integration projects, lector.ai commits to a concrete quote within three working days

## Evidence and references
- Company figures: 500+ million pages processed, 90 %+ classification accuracy, under 10 seconds per document, 80 %+ time saved
- Neckar-Odenwald district, pilot in the Agentic AI Hub of the German Federal Ministry for Digital Affairs and State Modernisation (2026): turnaround for GDPR access requests cut by over 90 % (from 2–4 weeks to at most one day) and up to 28 hours saved per case. The ministry's results report classifies Lector among the highly autonomous agentic systems
- Bezirk Unterfranken: TR-RESISCAN-compliant incoming mail, classified and routed automatically, in production since 2022
- VALUNY (formerly ITSC): incoming mail for more than 25 statutory health insurers
- Further customers and references: Sparkasse, Agravis, hkk, Tchibo Mobil, Wien Energie, SEfA, wpd
- Partners: OVHcloud, OMNINET, d.velop, BMDS Agentic AI Hub

## Common questions
- Where is my data processed? Depends on the deployment model: in the Lector Cloud exclusively in European data centres (OVHcloud; AI via Google Vertex AI, Microsoft Azure, OVHcloud, Scaleway in the EU); hybrid or on-premise on the customer's own servers
- Is my data used to train AI models? No. The models use documents only for processing; cloud providers delete them immediately afterwards, documented in the no-training statements
- Can Lector run entirely inside my infrastructure? Yes — fully on-premise including the AI models (GPU required, see above), or hybrid with the AI inference in the EU cloud
- Why not just use a chatbot? Single documents, yes. At hundreds or thousands of documents a month you need intake and export integrations, document classes defined in minutes, and measurable quality — Lector adds these, and the Evaluation Center compares settings and models on real documents
- When does a paid plan pay off? Starter bills per AI page call with no base fee; from roughly 15,000 AI page calls a month Business S is cheaper

## Upcoming events (2026)
- [Smart Country Convention](https://www.smartcountry.berlin/de), 13–15 October, Messegelände Berlin — [exhibitor profile](https://online.smartcountry.berlin/company/lector-ai-GmbH--1220115)
- [Service Summit](https://servicesummit.de/), 4–5 November, Areal Böhler, Düsseldorf
- [Messe KOMMUNAL](https://kommunal.de/veranstaltung/messe-kommunal-2026), 18–19 November, Messe Erfurt — stand Kirchstraße 11

## Pages
- [What is Lector](https://lector.ai/de/was-ist-lector): capabilities, use cases, Lector compared with chatbots and with building it yourself on Azure/AWS
- [Industries](https://lector.ai/de/branchen): public administration, health insurers, banks and insurance, energy and industry, logistics — with customer references, procurement notes (EVB-IT, beBPo, OZG) and the Evaluation Center
- [Security and compliance](https://lector.ai/de/sicherheit): infrastructure, encryption, access control, certifications, deployment models, evidence documents
- [Pricing](https://lector.ai/de/pricing): plans, quick start, enterprise options, consulting and integration, integrations, technical FAQ
- [Legal Center](https://lector.ai/de/legal): all contract and compliance documents in one viewer
- [About](https://lector.ai/de/ueber-uns) · [Careers](https://lector.ai/de/karriere) · [Press](https://lector.ai/de/presse) · [Case studies](https://lector.ai/de/case-studies) · [Contact](https://lector.ai/de/kontakt)
- [Documentation](https://app.lector.ai/docs/) · [API reference](https://app.lector.ai/docs/docs/api_reference/) · [Status page](https://status.lector.ai)

## Documents (German)
- [Terms and conditions](https://legal.app.lector.ai/html/lector.ai_agbs.html)
- [Data processing agreement with annexes](https://legal.app.lector.ai/html/lector.ai_avv_complete.html): description of processing, technical and organisational measures, approved sub-processors
- [No-training statements](https://legal.app.lector.ai/html/lector.ai_no_training_uebersicht.html): one per AI provider — Google Vertex AI (Gemini, Claude), Microsoft Azure OpenAI, OVHcloud, Scaleway
- [EU AI Act assessment](https://legal.app.lector.ai/html/lector.ai_eu_ai_act.html)
- [AI and data sovereignty statement](https://legal.app.lector.ai/html/lector.ai_ai_data_sovereignty_statement.html)

## Optional
- [Imprint](https://lector.ai/de/impressum) · [Privacy policy](https://lector.ai/de/datenschutz)
- [One-page overview](https://lector.ai/de/kurzueberblick): the platform on a single printable page
- [Agentic AI Hub results report](https://bmds.bund.de/fileadmin/BMDS/Dokumente/48-Anlage-Ergebnisbericht_Agentic_AI-Hub_.pdf): the ministry's report on the pilot phase, PDF, German