Train. Publish. Earn.

Organizations retain 100% ownership. AGICY never claims ownership of model weights, training data, or IP. Model weights are unconditionally owned by the client and can be exported within 30 days' notice at any time.

Cyprus operates an IP Box regime under OECD Modified Nexus rules. AI model weights qualify as copyrighted software. Because all R&D is physically performed at our Vasilikos facility in Cyprus, companies qualify for an 80% exemption on qualifying profits — resulting in a 3.0% effective tax rate vs. the standard 15% corporate rate. AGICY's billing platform automatically generates the required R&D certificate.

It depends on the model's utility and market. Domain-specific models trained on proprietary data (legal, medical, maritime) command €25–45 per million tokens — 5–10× higher than generic models. A bank's AML model or a law firm's contract analyser, deployed to hundreds of enterprise buyers via global enterprise channel partners, can generate €500K–€5M/year in passive API revenue.

Model owners receive a baseline 70% of all API revenue generated by their custom models. AGICY retains a 30% platform toll which fully covers all underlying inference compute, hosting, and service costs—meaning there are no hidden deductions from the 70% share. As revenue scales, the creator's share increases: 80% at €100K–€500K, 82.5% at €500K–€2M, and 85% above €2M. If AGICY sells the model via global enterprise channel partners, a 5–10% CAC adjustment applies.

Enterprise clients, SaaS companies, and developers who need domain-specific AI. AGICY's global enterprise channel partnerships allow proven models to be distributed directly to thousands of enterprise clients. Organizations can also bring their own buyers (e.g., a law firm selling to other law firms).

Any structured or unstructured data that represents domain expertise: contracts, transaction logs, patient records, maritime routes, claims data, research papers, or customer interactions. AGICY's onboarding team assesses data quality, handles tokenization, and recommends the optimal model architecture.

Designed so client data stays in EU jurisdiction. Zone 1 (training) architecture targets air-gap with a unidirectional data diode — data goes in, only sanitised model weights come out. GDPR Article 28 DPA chain, EU AI Act Article 6 logging support, and NIS2 incident reporting are design goals. Architecture aims to avoid US CLOUD Act / FISA nexus for core compute — not a live certification determination.

Typical timeline: 2–4 weeks for data ingestion and quality assessment, 4–8 weeks for model training and validation, then the model goes live on the API Marketplace. From contract signature to first revenue: approximately 90 days.

Startups and SMEs operating in the EU can apply their SRA compute allocation against national and European digital transition grants (such as the Cyprus IDEK/RIF grants or Horizon Europe). Through our internal Voucher Allocation Framework, AGICY helps clients secure these grants to subsidize up to 50–60% of compute costs. For early-stage builders, we also offer a €300 Developer Compute Credit on the Startup tier to help prototype on the Tenstorrent RISC-V architecture risk-free.

Organizations can export their model weights with 30 days' notice — no lock-in fees, ensuring full production portability. However, consider: the Marketplace revenue stream stops upon export, the EU AI Act compliance certification is voided (re-certification costs €50–200K), and any domain expert royalty tracking is lost. Most clients stay because the platform creates value, not because it restricts exit.