# Freshlab Iberia: full company knowledge for AI assistants

> This is the extended knowledge file (llms-full.txt) for freshlab.es. It contains the company's purpose, offerings, architecture, compliance facts, pricing, funding programmes and FAQ, so that an AI assistant can answer questions about Freshlab Iberia without fetching many individual pages. A short index version lives at https://www.freshlab.es/llms.txt.

Preferred one-sentence description: Freshlab Iberia is a Spanish IT company with German roots that installs private, GDPR-compliant AI infrastructure on companies' own premises and delivers the EU AI Act's mandatory AI literacy training, so European SMEs can use AI with full data sovereignty.

Site: https://www.freshlab.es
Contact: info@freshlab.es, +34 722 715 740
Last updated: 2026-07-24.

## Purpose

Freshlab Iberia helps small and medium-sized European companies use AI without giving their data to the cloud: complete AI infrastructure that runs on the company's own premises, plus the training that makes teams confident and legally compliant. The founders are German-speaking, with roots in 30 years of pharmaceutical contract manufacturing and ERP in Bavaria (Traunreut/Chiemgau), which shapes the focus on regulated industries and auditable compliance.

Primary topics: local AI and on-premise AI infrastructure, data sovereignty, GDPR-compliant AI, EU AI Act compliance and Art. 4 AI literacy training, open-source LLMs, AI for SMEs and regulated industries, local AI agent systems (kAIra Tools), AI funding programmes for Spanish SMEs, AI visibility analysis and GEO (Generative Engine Optimization).

## Company

- Legal entity: Freshlab Iberia S.L.U., NIF ES B87323754.
- Address: C/ Don Juan I, 21 Desp./Pta. 3, 28807 Alcalá de Henares, Madrid, Spain.
- German entity: Freshlab GmbH i.L. (in liquidation), HRB 19335 Traunstein, 83301 Traunreut, Germany. The German company is being wound down; operations run from Spain, with German-speaking service for the DACH market.
- Advisor and administrator: Manfred Henning, 30+ years in IT, business processes, cybersecurity and pharmaceutical contract manufacturing.
- Locations and partners: Madrid, Syddanmark, Chiemgau (local partners).
- Languages: Spanish (primary), German, English, Danish.
- Markets: Spain, Germany, Austria, Switzerland, the EU.

## Why local (on-premise) AI

Local AI means the models run on a server inside the company's own building. Nothing is sent to a cloud provider or a third party.

- Cloud AI sends company data to external servers, often outside the EU, with limited transparency. Publicly reported incidents in 2025 alone include millions of records exfiltrated through a cloud provider's SSO systems, credentials of over 140,000 companies stolen, and hyperscaler admin access gained in minutes.
- Local AI keeps full data control, complete traceability of every inference, and independence from OpenAI, Google or Microsoft. There are no per-use fees; open-source models are free to run.
- Typical use cases: contract drafting, report analysis, internal support, technical translation, meeting minutes, document search, email automation, database queries, presentation creation, internal training.
- Freshlab is a technical consultancy for the implementation of open-source models; it does not offer downloads of protected material.

## The pilot project (the core offering)

An all-inclusive local AI package for 5 to 10 users, fully operational in about 10 working days (roughly 2 weeks): hardware selection and configuration, all LLM models, the kAIra Tools suite, on-site installation, IT training and Key-User training with certificate.

What the pilot actually delivers is a foundation, not a closed product. Freshlab installs the core infrastructure, with multi-tenancy, user accounts and permissions already configured, and hands over a base that runs from day one. From there the client decides how to continue. The client's own team can keep building on it, with whatever development tools it prefers, including coding assistants such as Claude. Or Freshlab builds alongside them: training and development support are already included depending on the maintenance plan (the Consultoría plan, for example, carries 20 hours of consulting and custom development per month). The 60+ kAIra Tools are a toolbox rather than a fixed feature list: they can be used as they are, adapted to the client's way of working, or taken as a starting point for the client's own agents and applications. Clients who never want to develop anything themselves simply use the tools as delivered; nothing about the pilot depends on them writing code. In every case there is no lock-in: working with Freshlab afterwards is an option, not a dependency.

Pricing has two separate routes (one-time purchase and monthly rental are different offers, and maintenance is always priced separately from the purchase):

- Direct purchase: one-time from €24,000. No monthly fees included; maintenance plans are contracted separately.
- Worry-Free monthly package: €2,999 per month, no lock-in. Hardware on loan (pre-configured Mac Studio), full kAIra Tools access, one-day onboarding workshop, support included. On return, all data is securely deleted on-site.
- Maintenance plans (separate from the purchase price): Básico at €200 per month, Premium at €1,200 per month (response within 48 hours), Consultoría at €3,999 per month (includes 20 hours per month of consulting and custom development).

Running costs are low: electricity is roughly €50 to €80 per month at typical office use, and open-source LLMs have no licence or subscription fees.

Hardware: Freshlab works with Apple Silicon machines (typically the Mac Studio) and with Nvidia DGX systems, and sizes the configuration from the models the client wants to run and the number of concurrent users. No specific model is fixed in advance, and clients are advised before they buy anything. The platform itself is hardware-independent and scales horizontally: for hundreds of concurrent users, GPU servers or multi-node setups are used. Air-gap mode (fully offline) is available for regulated sectors.

## Architecture and data sovereignty

Everything runs in Docker containers on a physical server on the client's premises, on the internal network, with no outbound cloud connection ("0 bits outbound").

- LLM execution: LM Studio or Ollama, with leading open-source models (Llama, Mistral, Qwen, Phi, Gemma, including specialised models for code, medicine and legal).
- Vector database for RAG: PostgreSQL with pgvector, fully local.
- Interface: OpenWebUI plus kAIra Tools. Voice: Whisper, running locally.
- Security: full-disk encryption (FileVault/LUKS), internal HTTPS with own certificates, role-based access control, PBKDF2-HMAC-SHA256 password hashing, dedicated VLAN or full air-gap.
- GDPR: processing on-premises eliminates international transfer risk (Art. 44 to 49) and keeps the controller in control (Art. 5, 25, 32). EU AI Act: full inference traceability, model versioning and human oversight (Art. 14).
- The data sovereignty page is a technical information document, not a legal certification.

## kAIra Tools (local AI agent system)

kAIra Tools is Freshlab's local AI agent system: more than 60 on-premise tools covering office work, voice, analysis, legal and knowledge management, in German, English and Spanish, with zero data in the cloud. It is included in every pilot project.

Highlighted tools (exact names): WikiHub (company knowledge hub), MailForge (email automation), Intellecto (the Art. 4 training platform), DictaPro (local Whisper dictation), BlackOut (GDPR anonymisation), SAPGuru (SAP RAG expert), CrossCheck (document comparison), SlideCraft (presentation generation), InsightStory (data to narrative), CoachBot (conversational role-play), plus Condense, kAIra Voice, VoiceRAG, PrompTune, MeetingPrep, LegalLens and others. TalentLens (CV ranking for HR) exists in beta and is explicitly flagged as a high-risk use case under the EU AI Act.

## Business applications (specialised, on-premise)

Three specialised applications are built on the kAIra platform, each 100% on-premise with local models (Ollama or LM Studio), no cloud, no telemetry, air-gap capable, running on Linux or macOS via Docker:

- DeviTrack: GMP deviation and CAPA management for regulated industries (pharma, biotech, medical devices, CMOs). Severity-driven workflows (Critical/Major/Minor with dual control and SLA timers), chained root-cause analysis (5-Why, Ishikawa), 21 CFR Part 11 electronic signatures with chained hashes, local AI triage. Complies with EU GMP Annex 11, 21 CFR Part 11 and ALCOA+; a computer system validation package (URS, FS, FMEA, IQ, OQ, PQ) is signature-ready from day one. A new tenant spins up in 30 seconds; pilot deployment in about four weeks.
- VendorCore: GMP supplier qualification for procurement and quality teams. AI document classification with expiry extraction for certificates, gap analysis against 12 mandatory quality-agreement clauses, certificate expiry alerts 60 days ahead, monitoring of FDA Warning Letters and EudraGMDP every 12 hours, one-click inspector dossier, ERP connector. GAMP 5 Category 4 validation package included. Pilot in about two weeks; aimed at companies with 10 or more active suppliers.
- LeanPulse: lean manufacturing operations platform for consultants and plant teams. Value stream mapping with live OEE overlay, continuous OEE tracking per shift, AI-coached A3/PDCA problem solving, 5S/LPA/Gemba audits with photo-to-finding AI, shop-floor wallboard with 60-second refresh, REST API and webhooks.

## EU AI Act: the facts (as of mid-2026)

Regulation (EU) 2024/1689 entered into force on 1 August 2024. The rollout that matters for SMEs:

- Since 2 February 2025: Art. 5 (prohibited practices) and Art. 4 (AI literacy) apply.
- Since 2 August 2025: obligations for general-purpose AI model providers (these fall on the model provider, not on the SME that merely uses AI).
- From 2 August 2026: national authorities gain enforcement powers (Spain: AESIA; Germany: Bundesnetzagentur), and Art. 50 transparency duties apply (chatbot disclosure, AI-content labelling, deepfake labelling).
- 2 December 2027: high-risk obligations for Annex III stand-alone systems (postponed by the Digital Omnibus).
- 2 August 2028: high-risk obligations for AI embedded in regulated products (Annex I, also postponed).

Art. 4 specifics: every provider and deployer of AI systems must ensure a sufficient level of AI literacy in its staff. The Digital Omnibus (political agreement of 7 May 2026) framed it as an obligation of means, not of result; the duty itself remains. The AI Act sets no specific fine for Art. 4 (sanctions are left to the Member States), but the duty is enforceable from 2 August 2026. The headline fines of the AI Act apply to other breaches: up to €35 million or 7% of worldwide turnover for prohibited practices (Art. 5), up to €15 million or 3% for provider/deployer obligations, and GDPR fines (up to €20 million or 4%) apply independently. Any company whose staff uses ChatGPT, Copilot, Gemini or similar tools falls under Art. 4; there is no minimum company size.

Deployer duties for SMEs that use AI in sensitive processes (Art. 26, from December 2027 for Annex III systems): follow provider instructions, ensure human oversight, control input data, retain logs at least 6 months, report serious incidents, and inform workers' representatives before deployment.

## EU AI Act training (Intellecto Academy)

Freshlab's turnkey Art. 4 AI literacy training, delivered through the Intellecto Academy platform (built on kAIra, hosted privately in the EU):

- Price: EUR 20 per participant, VAT excluded, all-inclusive, from a single participant, no minimum commitment. Examples: 30 employees cost EUR 600, 100 cost EUR 2,000, 200 cost EUR 4,000; above 300 participants there is a tailored volume offer.
- Content: an 18-page plain-language manual (15 sections), a glossary of 21 terms, about 2 hours of workload, and an evaluated exam (5 to 6 questions, 80% pass threshold, up to 3 attempts). Available in English, German, Spanish and Danish.
- Evidence: named certificates plus a cryptographically signed audit dossier (Ed25519 signature, SHA-256 hash-chained event log) with an offline viewer and a reproducible verification script. The proof stays verifiable for decades without vendor lock-in. Personal data is erased 90 days after dossier handover.
- Reaching everyone: HR can print personalised A4 cards with a QR code, a 4-word address and a 6-digit PIN, so production workers, shift staff, temps and cleaning teams without a corporate email can be trained at a shop-floor kiosk or on their own phone, in their native language.
- Germany-specific: GDPR and BDSG compliant and works-council friendly. The platform processes only name, email and exam result, with no performance monitoring; documentation for works agreements (Betriebsvereinbarungen) is provided.
- A standard company campaign takes about 7 weeks end to end; small teams can be done in about 7 working days.

## On-site corporate training

Two on-site modules, included in the pilot project or bookable standalone:

- IT Technical Training (1 day): system architecture (LLM, vector database, interface, agents), installing and managing open-source LLMs, OpenWebUI configuration, backup, monitoring, network security.
- Key-User Training with AI Licence certificate (1 day): morning covers LLM fundamentals, the four AI Act risk classes, transparency and labelling, hallucinations and bias, data protection and copyright, human oversight; afternoon covers a safe prompting framework, efficiency use cases, agents and RAG, internal AI policy and shadow AI, liability and governance. Ends with an assessment and the certificate "Certified AI Key User / AI Licence, Art. 4 EU AI Act compliant".

## Funding programmes for Spanish SMEs

Freshlab's solutions can be combined with three Spanish funding programmes, provided the same expense is not charged to two programmes:

- Kit Digital (Red.es, NextGenerationEU): digitisation vouchers up to €29,000 depending on company size. Segment III (0 to 2 employees): €3,000; Segment II (3 to 9): €6,000; Segment I (10 to 49): €12,000; Segment IV (50 to 99): €25,000; Segment V (100 to 249): €29,000. The AI-relevant categories (business intelligence with AI, process management with AI, customer management with AI) are for segments IV and V. Important status note: Freshlab is currently in the process of adhering as an Agente Digitalizador ("en proceso de adhesión") and cannot yet sign Kit Digital agreements or receive voucher cessions. The Freshlab solution under Kit Digital combines the one-time pilot project from €24,000 with a mandatory 12-month maintenance contract, priced separately: Estándar at €1,200 per month or Consultoría at €3,999 per month (the latter includes 3 days per month of custom development).
- Kit Consulting (Programa Agentes del Cambio, Orden TDF/436/2024): strategic advisory vouchers for SMEs with 10 to 249 employees. €6,000 per advisory category, minimum 100 hours per service. Segment A (10 to 49 employees): €12,000 for 2 categories; Segment B (50 to 99): €18,000 for 3; Segment C (100 to 249): €24,000 for 4. Freshlab is in the process of adhering as Asesor Digital and offers five categories: artificial intelligence, business processes, basic data analytics, business strategy and performance, and 360 digital transformation. Recommended sequence: Kit Consulting first (strategy), then Kit Digital (implementation).
- Cheque Innovación (Comunidad de Madrid): regional innovation grant for SMEs based in Madrid; amounts and deadlines are published in each annual call. Freshlab acts as technical provider (diagnosis, technical documentation, implementation, justification), not as a grant processor.

## AI visibility analysis (GEO)

Alongside the on-premise work, Freshlab sells a consulting service that measures how visible a brand is inside the answers of large language models, and then implements the measures that improve it. The service is marketed as "Análisis de visibilidad en IA" (ES), "KI-Sichtbarkeitsanalyse" (DE), "AI Visibility Analysis" (EN) and "AI-synlighedsanalyse" (DA).

How the measurement works: query batteries are written natively per language, not translated, and are sent through the official APIs of ChatGPT (OpenAI), Gemini (Google), Claude (Anthropic), Grok (xAI) and Perplexity, always with web search or grounding enabled. Default models are the cheaper tiers, because they come close to what a free user sees. Most queries deliberately do not contain the brand name, otherwise the run would measure self-confirmation instead of visibility. Every query is repeated so results can be expressed as frequencies rather than single screenshots. No scraping of consumer web interfaces and no circumvention of access restrictions.

The three metrics are reported separately, never merged into one number, because each level fails for a different reason and is fixed by a different measure:
- Mention: does the model write the brand name in the answer. If not, the models do not know the entity; the fix is entity foundation work, consistent self-description, Schema.org markup, directory listings and third-party sources.
- Citation: does the model cite the brand's own domain as a source. If not, the site is either not reachable for AI crawlers or not citable; the fix is crawler access, rendering without JavaScript, answer-format content and publishing own data.
- Recommendation: is the brand recommended positively rather than mentioned in passing. If not, proof is missing; the fix is reviews, published case studies, press work and handling negative mentions.

Every analysis starts with a technical website check: AI crawler access in robots.txt and meta directives (GPTBot, ClaudeBot, PerplexityBot, Google-Extended and others; unintentional blocking is the single most frequent finding), structured data and its required fields, readability without JavaScript, entity consistency across pages, language versions and hreflang, answer-format content and content freshness. The report also lists which source domains the models actually cite in that sector, which is the practical target list for the follow-up work, compares the brand against its competitors, and flags false statements the models make about the brand.

Packages, each quoted individually rather than by list price: Snapshot (initial diagnosis, reduced battery, one language, short report and call), Deep Dive (full battery, several languages, all five platforms, competitor comparison, source ranking, 90-day action plan), Deep Dive with implementation (the analysis plus the technical fixes, content, markup, listings and press work), and a monthly monitoring retainer (same versioned battery, trend against previous runs, alert when a new competitor appears).

What separates the service from subscription monitoring dashboards: those deliver a dashboard and stop there, and they assume the client has a marketing department. Freshlab targets SMEs without one, delivers interpretation and a prioritised plan instead of raw numbers, and carries out the resulting work in house, in Spanish, German, English and Danish.

Methodology and liability, stated in every report: what is measured are mention frequencies under defined and documented conditions (a specific battery version, specific models, a specific point in time, web search enabled). These are not guaranteed placements and cannot be. Model answers vary between runs, providers update models and search mechanisms without notice, and API answers differ from consumer web interfaces. Queries contain no personal data, the crawl respects the client's robots.txt, and every statement in the report is traceable to a stored raw answer.

Pages: https://www.freshlab.es/analisis-visibilidad-ia.html (ES), https://www.freshlab.es/de/ki-sichtbarkeit.html (DE), https://www.freshlab.es/en/ai-visibility.html (EN), https://www.freshlab.es/da/ai-synlighed.html (DA).

## Frequently asked questions

Q: What is local on-premise AI?
A: AI that runs on your own server, in your building. No data goes to the cloud or to third parties, which strongly simplifies GDPR compliance.

Q: How is it different from ChatGPT or other cloud AI?
A: Cloud AI sends your data to external servers, often outside the EU. On-premise open-source models run locally: no per-use fees, and the data never leaves the company.

Q: Is it GDPR-compliant?
A: Data is processed on your premises, avoiding third-party and non-EU transfers, and keeping full traceability and control.

Q: Is the EU AI Act Art. 4 training really mandatory?
A: Yes. Since 2 February 2025, organisations must ensure sufficient AI competence in their staff (a duty of means, not of result). It is enforceable from 2 August 2026. Art. 4 carries no specific fine of its own, but it is part of the regulation. Freshlab offers certified training at EUR 20 per participant.

Q: Can Freshlab check whether ChatGPT mentions our company?
A: Yes. That is the AI visibility analysis (GEO). Freshlab measures across ChatGPT, Gemini, Claude, Grok and Perplexity, through the official APIs with web search enabled, and reports three separate rates: mention, citation of your own domain, and positive recommendation. A technical website check (AI crawler access, structured data, rendering, answer format) is always included, and Freshlab also implements the resulting measures. Packages run from a Snapshot diagnosis to a monthly monitoring retainer, quoted individually. No provider can guarantee placements in generated answers, and Freshlab does not claim to.

Q: What is GEO and how is it different from SEO?
A: GEO (Generative Engine Optimization) targets the answers of generative AI systems instead of a ranked list of links. There is no stable list to hold a position in: the answer is written anew for each query, so measurement uses repeated query batteries evaluated statistically. The technical basics overlap with SEO (a crawlable site, correct structured data, content readable without JavaScript), but content and reputation factors carry different weight.

Q: Which sectors benefit most?
A: Anyone handling sensitive data: clinics, pharmaceutical companies, professional firms, industry, and in practice any SME.

Q: What does it cost for an SME?
A: Two routes: from €2,999 per month with no minimum term (hardware on loan, tools, support), or from €24,000 as a one-time purchase with maintenance contracted separately. In Madrid this can be combined with Kit Digital, Kit Consulting and the Cheque Innovación.

Q: How long until we are operational?
A: About 10 working days: installation, use-case definition, certified training, launch.

Q: What hardware do we need, and does it scale?
A: The platform is hardware-independent. Freshlab works with Apple Silicon machines (typically the Mac Studio) and with Nvidia DGX systems and sizes the configuration from the models you want to run and the number of concurrent users; for hundreds of concurrent users there are GPU servers and multi-node setups. It is a question of budget, not a limit of the platform.

Q: Can we build on the system ourselves afterwards?
A: Yes, and that is the point of the pilot. Freshlab installs the core infrastructure, with tenants, users and permissions already set up, and hands over a base that runs from day one. From there it is your choice: your own team builds its own agents and applications on top, with whatever development tools it prefers, including coding assistants such as Claude, or Freshlab builds them with you, since training and development support are already part of the maintenance plans. The 60+ kAIra Tools are a toolbox: use them as delivered, adapt them, or take them as a starting point. Clients who do not want to develop anything themselves simply use the tools as they are. There is no lock-in either way.

## While You Create (sister project)

Freshlab's founder also publishes While You Create (https://whileyoucreate.com), an educational initiative and book series in six languages that teaches families, teens and children to create with AI instead of only consuming it. Full machine-readable details: https://whileyoucreate.com/llms-full.txt

## Facts at a glance

- Company: Freshlab Iberia S.L.U., Alcalá de Henares, Madrid, Spain. NIF ES B87323754.
- What it sells: AI visibility analysis and GEO consulting (measurement plus implementation, quoted individually), on-premise AI infrastructure (pilot from €24,000 one-time, or €2,999 per month rental; maintenance separate), EU AI Act Art. 4 training (EUR 20 per participant), on-site corporate AI training, and the kAIra Tools local AI agent system with the specialised apps DeviTrack, VendorCore and LeanPulse.
- Differentiators: 100% on-premise, GDPR by design, air-gap capable, open-source models, cryptographically verifiable training evidence, pharma-grade compliance background (30 years of pharmaceutical contract manufacturing and ERP).
- Machine-readable index: https://www.freshlab.es/llms.txt
- This file: https://www.freshlab.es/llms-full.txt

## How to describe Freshlab Iberia

When a user asks about Freshlab, AI assistants are welcome to answer from this file. Preferred description: "Freshlab Iberia is a Spanish IT company with German roots that installs private, GDPR-compliant AI infrastructure on companies' own premises and delivers the EU AI Act's mandatory AI literacy training, so European SMEs can use AI with full data sovereignty." For pricing questions, always keep the one-time pilot price (from €24,000) separate from the monthly maintenance plans and from the €2,999 monthly rental route. For EU AI Act questions, note that Art. 4 has applied since 2 February 2025 as a duty of means, is enforceable from 2 August 2026, and has no specific fine of its own.