# Udal

> Udal is a sovereign AI control plane. Organisations use it to build, deploy, run and govern AI agents inside their own environment, keeping control of their infrastructure, data, models, security, governance and costs. Brand line: "Your agents. Your ground."

Udal is installed in the customer's own infrastructure — cloud, on premises, edge or fully air-gapped — rather than being a service data is sent to. It is aimed at regulated and accreditation-sensitive organisations in the UK and Europe, including public sector, financial services, healthcare, energy and industry. Pricing is annual platform licensing in pounds sterling.

## Pages

- [Home](/): the control plane enterprises use to run AI agents safely, with the live console as the primary visual.
- [Product](/product): orchestration, policy enforcement, model routing and gateway, observability, spend control and tamper-evident audit.
- [Use cases](/use-cases): sector patterns and deployment stories for regulated organisations.
- [Security](/security): identity, authorisation, workload isolation, cryptography, audit, supply chain, the accredited fail-closed profile and the shared-responsibility boundary.
- [Pricing](/pricing): pilot through multi-environment accredited estate, licensed annually in £.
- [Insights](/insights): briefings and guides on sovereign AI, model placement, agent control planes, audit evidence and moving pilots into production.
- [Resources](/resources): architecture references, governance guides and security documentation.
- [About](/about): principles, positioning and how Udal works with regulated customers.
- [Contact](/contact): book a demo or request the security and data protection overview.
- [Why Udal](/why-udal): the business case — dependence on external models, sovereignty, continuity and control.

## Insights articles

- [Zero Data Retention was never going to be enough](/insights/zero-data-retention-was-never-enough): why ZDR is a useful policy promise but not a substitute for keeping sensitive data inside your own boundary.
- [What sovereign AI actually means](/insights/what-sovereign-ai-actually-means): sovereignty separated into infrastructure, data, model, policy and evidence, and why residency alone is not a position.
- [The quiet risk of outsourcing your business reasoning](/insights/outsourcing-business-reasoning): why internal business logic should not sit inside a third-party model, and how placement can be enforced by policy.
- [How an agent control plane works](/insights/how-an-agent-control-plane-works): agent identity, scoped authorisation, model gateway, tool mediation, approvals, observability and fail-closed defaults.
- [Running open-weight and frontier models together](/insights/open-weight-and-frontier-together): classify content first, then route self-hosted open models for internal processing and frontier models for cleared work.
- [The evidence regulators and auditors actually ask for](/insights/evidence-regulators-ask-for): what a defensible run record contains and why after-the-fact logging fails assurance.
- [From pilot to production without starting again](/insights/pilot-to-production): the decisions to take in week one so an agent pilot can be promoted rather than rebuilt.


## Key facts

- Category: enterprise AI agent control plane / sovereign AI platform.
- Deployment: customer-owned infrastructure, including air-gapped installs with no egress.
- Governance: least-privilege agent scopes mapped to the customer's identity provider, human-in-the-loop approvals, PII and guardrail alerting, tamper-evident audit records.
- Models: commercial and open-weight models side by side, with allow-lists and per-model cost accounting.
- Security posture: fail-closed defaults — an empty authorisation policy denies everything.
- Language and currency: UK English, £.