ZILO Migrate

Security and deployment

A migration model that lives in your data centre

Migration data is the most sensitive copy of a firm's book that will ever exist. So ZILO™ Migrate is never SaaS: the model and the platform run on your infrastructure, and nothing leaves.

The model

Frontier capability, without the frontier model

Hosted frontier models cannot be used on a regulated book. So we built a model that does not need to be hosted: reinforcement-learned and fine-tuned on migration work, small enough to run on-prem, and benchmarked to perform on par with the frontier for this task.

  • Fine-tuned for one job

    A small model, fine-tuned and reinforcement-learned on data migration work: mapping, transformation, trial runs and reconciliation. It does one thing, and it does it to frontier standard.

  • On par with frontier models

    Benchmarked against the leading hosted frontier models on complex migration scenarios drawn from real books, and matching them, at a fraction of the compute and the cost.

  • Small enough to run inside

    Because it is small, it runs on a GPU host or appliance in your data centre, inside the air gap. No hosted inference, no outbound calls, no third party between the agent and your data.

Deployment

On-prem, air-gapped, yours

The harness comes to the data, not the other way round. One deployment model: air-gapped, with the model running on-prem beside your data.

Public internet
Sealed · air-gappedOn-prem appliance
  • Air-gapped, always

    Platform and model install inside your estate as a single isolated tenant with no outbound connection, aligned to ISO 27001 controls and inheriting your own. There is no hosted option.

  • Appliance or your own hardware

    Delivered as a hardware appliance, or as platform and model weights installed on GPU hosts you already run. Either way, your network boundary is the only one.

  • The model comes with it

    Inference runs on your GPU host or appliance. No call to a hosted model is ever made, so there is no third party between the agent and your data.

  • The whole loop inside

    Because the model is inside, the entire loop runs there: ingestion, filtering, transformation, reconciliation and load, with the evidence pack alongside.

Assurance

Controls your risk function will recognise

Designed with the due-diligence process in mind, so the answers to the questionnaire are properties of the product rather than promises in a contract.

  • Never SaaS

    The platform and the model are installed in your environment as a single, isolated tenant. There is no hosted or multi-tenant deployment of ZILO™ Migrate.

  • No egress

    The deployment is air-gapped: no client records leave your estate and no external service is reachable, let alone called, during a migration.

  • Access control to the agent

    Role-based access applies to people and agents alike. An agent holds a role and a set of permissions the same way a person does, so you decide what it can read, change and run.

  • Every action on the record

    Full audit trails and version history across mappings, runs, controls and approvals. A written reason for every agent decision, retained with the evidence.

  • Model risk, answered

    Human-in-the-loop gates, confidence and variance reporting, and a model that never leaves your environment address the questions in SR 11-7, PRA SS1/23, DORA and EU AI Act due diligence.

  • Aligned to ISO 27001 and SOC 2

    Controls are aligned to ISO 27001 and SOC 2 Type II, and an on-prem deployment inherits your own. Formal attestation is on our audit roadmap; ask us for the current status and the due-diligence pack.

Request the due-diligence pack

Talk to the team.