AI Data & Model Handling

Your infrastructure, your rules

All models are deployed on your hardware, inside your network perimeter. Air-gapped, network-isolated, or your cloud environment — the choice is yours. No data ever touches a third-party API. No cloud egress fees. No platform tax.

No training on your data

• Training data used to fine-tune a model for your engagement belongs exclusively to you. It is processed in isolated compute environments with no cross-engagement data leakage.
• EYwALINK does not use client data to train, benchmark, or improve its own models or tools.

Inference data is not persisted

• Data submitted to AI models (prompts, queries, context) is processed in real-time. For self-hosted models on your infrastructure, no inference data is transmitted to EYwALINK.
• Data is not persisted unless your Statement of Work explicitly requires it.

Vector stores stay with you

• Embeddings and vector indices created from your data are stored on your infrastructure under your security controls. No cross-tenant vector stores. No shared embeddings.

Model weights are yours

• Fine-tuned model weights produced during an engagement belong to you. EYwALINK does not retain model weights after delivery.

Open-source models only

• EYwALINK uses open-source AI models exclusively — no OpenAI, no Anthropic, no Google models for your data.
• Inference runtimes: vLLM, Ollama. Frameworks: LangGraph. All under open-source licences.
• A Software Bill of Materials (SBOM) listing all model sources and licences is provided with every engagement deliverable, so you can audit every component before it enters your infrastructure.

Synthetic data

Synthetic data generated during an engagement is treated as personal data if it can be linked back to an identifiable individual, and handled in accordance with the applicable data-protection obligations.

What this means in practice

Contact

Questions about how your data and models are handled? Contact info@eywalink.org.