Hybrid AI uses different models for different tasks. In Arvocatus, anything involving client or privileged material stays on private, open-source models in your cloud, with small models answering first and larger models when needed. Tasks with no client material, such as tracking new legislation or summarising published guidance, can use managed models such as Claude, GPT or Grok.

Private-only means lawyers miss out on the most capable models for public research. Managed-only sends client material to a vendor and bills every page by the token. Hybrid gives each task the right model: confidentiality where it matters, frontier capability where it helps, and the lower cost for each workload.

No. Every request passes a policy check before any model sees it. Anything involving client or privileged material, or where the check is unsure, stays private. You decide which task types may use managed AI at all, and you can switch it off entirely.

Running legal AI is far more than provisioning a server with a GPU. It needs infrastructure provisioning as code, model serving, retrieval over your own documents, staging environments, release engineering with evaluation gates and rollback, observability, autoscaling, backup and security policy, all kept running every day. Arvocatus delivers that complete stack, built on XePlatform, inside your own cloud account and operated for you, so a supervised pilot can start within weeks without new hires.

XePlatform, the engine under Arvocatus, runs the full stack in your account: managed Kubernetes and node pools, infrastructure provisioning, model serving and scaling, the policy check that routes each request, staging environments, the release pipeline that tests every model against your benchmark, observability, backup and recovery, and security policy enforcement. Your team keeps ownership; we keep it running.

Arvocatus is designed to support GDPR and the EU AI Act. Client data is processed only inside your own account, your organisation remains the data controller, and we provide documentation to support your impact assessments and security reviews. Your organisation defines the intended use of each workflow, and compliance depends on how you use the platform.

It is designed to. Client material never leaves your account, so there is no vendor copy to disclose or explain. We supply documentation you can use with your regulator, insurer and clients’ security questionnaires.

No. Arvocatus supports legal work with human oversight: a qualified lawyer reviews every output. Every answer cites its source and states its confidence, so reviewers can check each point quickly.

Open-source general and legal-tuned models run inside your account for all client work, adopted only after they pass evaluation on your own matters. Managed models such as Claude, GPT or Grok are used only for tasks with no client material.

Yes. Arvocatus runs on the major public clouds in the region you choose, in Europe or North America, or in a private cloud, built on open standards so you can move without rebuilding.

A fixed platform subscription plus your own cloud costs, with no per-token fees for client work. Managed AI is billed per token only for the tasks you route to it, under a monthly cap. Our break-even calculator shows when owning capacity becomes cheaper than paying per use.

The environment is set up in hours and your knowledge loaded within days. A supervised pilot with your lawyers typically runs about four weeks, while your risk and data protection teams complete their review in parallel.